en
×

分享给微信好友或者朋友圈

使用微信“扫一扫”功能。
参考文献 1
American Psychiatric Association. Dsm-5 Task Force. Diagnostic and statistical manual of mental disorders : DSM-5. 5th ed. Washington, D.C. 2013: xliv, 947 p
参考文献 2
Gottesman, Ii, GouldT D. The endophenotype concept in psychiatry: etymology and strategic intentions. Am J Psychiatry, 2003, 160(4): 636-645
参考文献 3
LiL J. Functional magnetic resonance imaging diagnostic marker for depression and its clinical application prospects. Chin J Psychiatry, 2015, 48(2) : 65-67
参考文献 4
DilsizianS E, SiegelE L. Artificial intelligence in medicine and cardiac imaging: harnessing big data and advanced computing to provide personalized medical diagnosis and treatment. Curr Cardiol Rep, 2014, 16(1): 441
参考文献 5
JentschM C, Van BuelE M, BoskerF J, et al. Biomarker approaches in major depressive disorder evaluated in the context of current hypotheses. Biomark Med, 2015, 9(3): 277-297
参考文献 6
LuiS, ZhouX J, SweeneyJ A, et al. Psychoradiology: the frontier of neuroimaging in psychiatry. Radiology, 2016, 281(2): 357-372
参考文献 7
LinW. Principles of magnetic resonance imaging: a signal processing perspective [Book Review]. IEEE Signal Process Mag, 2000, 19(5): 129-130
参考文献 8
Le-BihanD, ManginJ C, ClarkC, et al. Diffusion tensor imaging: concepts and applications. J Magn Reson Imaging, 2010, 13(4): 534-546
参考文献 9
DeyoeE A, BandettiniP, NeitzJ, et al. Functional magnetic resonance imaging (FMRI) of the human brain. J Neurosci Methods, 1994, 54(2): 171-187
参考文献 10
RiveM M, VanR G, VeltmanD J, et al. Neural correlates of dysfunctional emotion regulation in major depressive disorder. a systematic review of neuroimaging studies. Neurosci Biobehav Rev, 2013, 37(10): 2529-2553
参考文献 11
MarchettiI, KosterE H, Sonuga-BarkeE J, et al. The default mode network and recurrent depression: a neurobiological model of cognitive risk factors. Neuropsychol Rev, 2012, 22(3): 229-251
参考文献 12
ChenM C, BurleyH W, GotlibI H. Reduced sleep quality in healthy girls at risk for depression. J Sleep Res, 2012, 21(1): 68-72
参考文献 13
JoormannJ, CooneyR E, HenryM L, et al. Neural correlates of automatic mood regulation in girls at high risk for depression. J Abnorm Psychol, 2012, 121(1): 61-72
参考文献 14
DrevetsW C. Orbitofrontal cortex function and structure in depression. Ann N Y Acad Sci, 2007, 1121(1): 499-527
参考文献 15
DrevetsW C, PriceJ L, FureyM L. Brain structural and functional abnormalities in mood disorders: implications for neurocircuitry models of depression. Brain Struct Funct, 2008, 213(1-2): 93-118
参考文献 16
FonsekaT M, MacqueenG M, KennedyS H. Neuroimaging biomarkers as predictors of treatment outcome in Major Depressive Disorder. J Affect Disord, 2017, 233(6): 2049
参考文献 17
FuC H, CostafredaS G. Neuroimaging-based biomarkers in psychiatry: clinical opportunities of a paradigm shift. Canadian Journal of Psychiatry Revue Canadienne De Psychiatrie, 2013, 58(9): 499
参考文献 18
KozaJ R, BennettF H, AndreD, et al. Automated design of both the topology and sizing of analog electrical circuits using genetic programming. Dordrecht: Springer Netherlands, 1996: 151-170
参考文献 19
WilsonS M, OgarJ M, LaluzV, et al. Automated MRI-based classification of primary progressive aphasia variants. Neuroimage, 2009, 47(4): 1558-1567
参考文献 20
FungG, DengY, ZhaoQ, et al. Distinguishing bipolar and major depressive disorders by brain structural morphometry: a pilot study. BMC Psychiatry, 2015, 15(1): 298-298
参考文献 21
ZhangJ, ZongC. Deep neural networks in machine translation: an overview. IEEE Intelligent Systems, 2015, 30(5): 16-25
参考文献 22
HintonG E. Deep belief networks. Scholarpedia, 2009, 4(6): 5947
参考文献 23
LecunY, BengioY, HintonG. Deep learning. Nature, 2015, 521(7553): 436-444
参考文献 24
ShenD, WuG, SukH I. Deep learning in medical image analysis. Annu Rev Biomed Eng, 2017, 19(1): 221-248
参考文献 25
GeorgeviciA I, TerblancheM. Neural networks and deep learning: a brief introduction. Intensive Care Med, 2019, 45(5): 712-714
参考文献 26
SymmsM, JagerH R, SchmiererK, et al. A review of structural magnetic resonance neuroimaging. J Neurol Neurosurg Psychiatry, 2004, 75(9): 1235-1244
参考文献 27
BoraE, FornitoA, PantelisC, et al. Gray matter abnormalities in Major Depressive Disorder: a meta-analysis of voxel based morphometry studies. J Affect Disord, 2012, 138(1-2): 9-18
参考文献 28
ChenT, KendrickK M, WangJ, et al. Anomalous single-subject based morphological cortical networks in drug-naive, first-episode major depressive disorder. Hum Brain Mapp, 2017, 38(5): 2482-2494
参考文献 29
MwangiB, EbmeierK P, MatthewsK, et al. Multi-centre diagnostic classification of individual structural neuroimaging scans from patients with major depressive disorder. Brain, 2012, 135(5): 1508-152
参考文献 30
SacchetM D, LivermoreE E, IglesiasJ E, et al. Subcortical volumes differentiate major depressive disorder, bipolar disorder, and remitted major depressive disorder. J Psychiatr Res, 2015, 68: 91-98
参考文献 31
DosenbachN U F, SchlaggarB L. Prediction of individual brain maturity using fMRI. Science, 2010, 329(5997): 1358-1361
参考文献 32
GuoH, ChengC, CaoX, et al. Resting-state functional connectivity abnormalities in first-onset unmedicated depression. Neural Regen Res, 2014, 9(2): 153-163
参考文献 33
ZengL L, ShenH, LiuL, et al. Identifying major depression using whole-brain functional connectivity: a multivariate pattern analysis. Brain A Journal of Neurology, 2012, 135(Pt 5): 1498
参考文献 34
MaximV, SendurL, FadiliJ, et al. Fractional Gaussian noise, functional MRI and Alzheimer's disease. Neuroimage, 2005, 25(1): 141-158
参考文献 35
WeiM, QinJ, YanR, et al. Identifying major depressive disorder using Hurst exponent of resting-state brain networks. Psychiatry Res, 2013, 214(3): 306-312
参考文献 36
VersaceA, AlmeidaJ R, HasselS, et al. Elevated left and reduced right orbitomedial prefrontal fractional anisotropy in adults with bipolar disorder revealed by tract-based spatial statistics. Arch Gen Psychiatry, 2008, 65(9): 1041
参考文献 37
QinJ, WeiM, LiuH, et al. Abnormal hubs of white matter networks in the frontal-parieto circuit contribute to depression discrimination via pattern classification. Magn Reson Imaging, 2014, 32(10): 1314-1320
参考文献 38
FangP, ZengL L, ShenH, et al. Increased cortical-limbic anatomical network connectivity in major depression revealed by diffusion tensor imaging. Plos One, 2012, 7(9): e45972
参考文献 39
MclaughlinN C, PaulR H, GrieveS M, et al. Diffusion tensor imaging of the corpus callosum: a cross-sectional study across the lifespan. International Journal of Developmental Neuroscience the Official Journal of the International Society for Developmental Neuroscience, 2007, 25(4): 215-221
参考文献 40
VersaceA, AlmeidaJ R, QuevedoK, et al. Right orbitofrontal corticolimbic and left corticocortical white matter connectivity differentiate bipolar and unipolar depression. Biol Psychiatry, 2010, 68(6): 560-567
参考文献 41
JacobsR H, OrrJ L, GowinsJ R, et al. Biomarkers of intergenerational risk for depression: a review of mechanisms in longitudinal high-risk (LHR) studies. J Affect Disord, 2015, 175: 494-506
参考文献 42
FolandrossL C, SacchetM D, PrasadG, et al. Cortical thickness predicts the first onset of major depression in adolescence. Int J Dev Neurosci, 2015, 46: 125-131
参考文献 43
BeldenA C, BarchD M, OakbergT J, et al. Anterior insula volume and guilt: neurobehavioral markers of recurrence after early childhood major depressive disorder. Jama Psychiatry, 2014, 72(1): 40
参考文献 44
KoolschijnP C, Van HarenN E, Lensvelt-MuldersG J, et al. Brain volume abnormalities in major depressive disorder: a meta-analysis of magnetic resonance imaging studies. Neuroimage, 2009, 47(11): S152-S152
参考文献 45
HajekT, CookeC, KopecekM, et al. Using structural MRI to identify individuals at genetic risk for bipolar disorders: a 2-cohort, machine learning study. Journal of Psychiatry & Neuroscience Jpn, 2015, 40(2): 140142
参考文献 46
SchnackH G, NieuwenhuisM, HarenN E M V, et al. Can structural MRI aid in clinical classification? A machine learning study in two independent samples of patients with schizophrenia, bipolar disorder and healthy subjects. Neuroimage, 2014, 84(1): 299-306
参考文献 47
RocharegoV, JogiaJ, MarquandA F, et al. Examination of the predictive value of structural magnetic resonance scans in bipolar disorder: a pattern classification approach. Psychol Med, 2014, 44(3): 519-532
参考文献 48
CuijpersP, BeekmanA, SmitF, et al. Predicting the onset of major depressive disorder and dysthymia in older adults with subthreshold depression: a community based study. Int J Geriatr Psychiatr, 2010, 21(9): 811-818
参考文献 49
KimB, KimM-K, YooE, et al. Comparison of panic disorder with and without comorbid major depression by using brain structural magnetic resonance imaging. Prog Neuropsychopharmacol Biol Psychiatry, 2013, 43: 188-196
参考文献 50
MaZ, LiR, YuJ, et al. Alterations in regional homogeneity of spontaneous brain activity in late-life subthreshold depression. Plos One, 2013, 8(1): e53148
参考文献 51
WuT, LongX, ZangY, et al. Regional homogeneity changes in patients with Parkinson's disease. Hum Brain Mapp, 2010, 30(5): 1502-1510
参考文献 52
PaakkiaJ J, RahkobJ, LongcX, et al. Alterations in regional homogeneity of resting-state brain activity in autism spectrum disorders. Brain Res, 2010, 1321(12): 169-179
参考文献 53
EmmrichA, BeesdobaumK, GlosterA T, et al. Depression does not affect the treatment outcome of CBT for panic and agoraphobia: results from a multicenter randomized trial. Psychother Psychosom, 2012, 81(3): 161-172
参考文献 54
LuekenU, StraubeB, YangY, et al. Separating depressive comorbidity from panic disorder: A combined functional magnetic resonance imaging and machine learning approach. J Affect Disord, 2015, 184: 182-192
参考文献 55
JiaZ, HuangX, WuQ, et al. High-field magnetic resonance imaging of suicidality in patients with major depressive disorder. Am J Psychiatry, 2010, 167(11): 1381-1390
参考文献 56
MacqueenG M. Magnetic resonance imaging and prediction of outcome in patients with major depressive disorder. Journal of Psychiatry & Neuroscience Jpn, 2009, 34(5): 343
参考文献 57
CostafredaS G, ChuC, AshburnerJ, et al. Prognostic and diagnostic potential of the structural neuroanatomy of depression. Plos One, 2009, 4(7): e6353
参考文献 58
NouretdinovI, CostafredaS G, GammermanA, et al. Machine learning classification with confidence: application of transductive conformal predictors to MRI-based diagnostic and prognostic markers in depression. NeuroImage, 2011, 56(2): 809-813
参考文献 59
DavatzikosC, ShenD, GurR C, et al. Whole-brain morphometric study of schizophrenia revealing a spatially complex set of focal abnormalities. Arch Gen Psychiatry, 2005, 62(11): 1218
参考文献 60
GammermanA, NouretdinovI, BurfordB, et al. Clinical mass spectrometry proteomic diagnosis by conformal predictors. Stat Appl Genet Mol Biol, 2008, 7(2): Article13
参考文献 61
GongQ, WuQ, ScarpazzaC, et al. Prognostic prediction of therapeutic response in depression using high-field MR imaging. Neuroimage, 2011, 55(4): 1497-1503
参考文献 62
FuC H, WilliamsS C, CleareA J, et al. Neural responses to sad faci al expressions in major depression following cognitive behavioral therapy. Biol Psychiatry, 2008, 64(6): 505-512
参考文献 63
MarquandA F, Mourao-MirandaJ, BrammerM J, et al. Neuroanatomy of verbal working memory as a diagnostic biomarker for depression. Neuroreport, 2008, 19(15): 1507-1511
参考文献 64
ScholkopfB, PlattJ C, Shawe-TaylorJ, et al. Estimating the support of a high-dimensional distribution. Neural computation, 2001, 13(7): 1443-1471
参考文献 65
HoyK E, ThomsonR H, CherkM H, et al. Effect of magnetic seizure therapy on regional brain glucose metabolism in major depression. Psychol Med, 2013, 211(2): 169-175
参考文献 66
ShimizuY, YoshimotoJ, TokiS, et al. Toward probabilistic diagnosis and understanding of depression based on functional MRI data analysis with logistic group LASSO. Plos One, 2015, 10(5): e0123524
参考文献 67
RosaM J, PortugalL, HahnT, et al. Sparse network-based models for patient classification using fMRI. Neuroimage, 2015, 105: 493-506
参考文献 68
PatelM J, AndreescuC, PriceJ C, et al. Machine learning approaches for integrating clinical and imaging features in late‐life depression classification and response prediction. Int J Geriatr Psychiatry, 2015, 30(10): 1056
参考文献 69
YuanG X, ChangK W, HsiehC J, et al. A comparison of optimization methods and software for large-scale L1-regularized linear classification. J Mach Learn Res, 2010, 11(2): 3183-3234
参考文献 70
PinayaW H, GadelhaA, DoyleO M, et al. Using deep belief network modelling to characterize differences in brain morphometry in schizophrenia. Sci Rep, 2016, 6(1): 38897
参考文献 71
UlloaA, PlisS M, ErhardtE B, et al. Synthetic structural magnetic resonance image generator improves deep learning prediction of schizophrenia. Proceedings of the International Workshop on Machine Learning for Signal Processing. 2015: 1-6
参考文献 72
HazlettH C, GuH, MunsellB C, et al. Early brain development in infants at high risk for autism spectrum disorder. Nature, 2017, 542(7641): 348
参考文献 73
HeinsfeldA S, FrancoA R, CraddockR C, et al. Identification of autism spectrum disorder using deep learning and the ABIDE dataset. Neuroimage Clinical, 2018, 17(C): 16-23
参考文献 74
SukH I, WeeC Y, LeeS W, et al. State-space model with deep learning for functional dynamics estimation in resting-state fMRI. Neuroimage, 2016, 129: 292-307
参考文献 75
IslamJ, ZhangY. A novel deep learning based multi-class classification method for alzheimer’s disease detection using brain MRI data. Cham. Springer International Publishing, 2017: 213-222
参考文献 76
SukH I, LeeS W, ShenD. Deep sparse multi-task learning for feature selection in Alzheimer's disease diagnosis. Brain Struct Funct, 2016, 221(5): 2569-2587
参考文献 77
HanX, ZhongY, HeL, et al. The unsupervised hierarchical convolutional sparse auto-encoder for neuroimaging data classification. cham. Springer International Publishing, 2015: 156-166
参考文献 78
ChoiH, JinK H. Predicting cognitive decline with deep learning of brain metabolism and amyloid imaging. Behav Brain Res, 2018, 344: 103-109
参考文献 79
DeshpandeG, WangP, RangaprakashD, et al. Fully connected cascade artificial neural network architecture for attention deficit hyperactivity disorder classification from functional magnetic resonance imaging data. IEEE Transactions on Cybernetics, 2015, 45(12): 2668-2679
参考文献 80
MunsellB C, C-YWee, KellerS S, et al. Evaluation of machine learning algorithms for treatment outcome prediction in patients with epilepsy based on structural connectome data. Neuroimage, 2015, 118: 219-230
参考文献 81
DuY, FuZ, CalhounV D. Classification and prediction of brain disorders using functional connectivity: promising but challenging. Front Neurosci, 2018, 12: 525
参考文献 82
DeshpandeG, LiZ, SanthanamP, et al. Recursive cluster elimination based support vector machine for disease state prediction using resting state functional and effective brain connectivity. PLoS One, 2010, 5(12): e14277
参考文献 83
KimJ, CalhounV D, ShimE, et al. Deep neural network with weight sparsity control and pre-training extracts hierarchical features and enhances classification performance: Evidence from whole-brain resting-state functional connectivity patterns of schizophrenia. Neuroimage, 2015, 124(Pt A): 127-146
参考文献 84
BermanM G, MisicB, BuschkuehlM, et al. Does resting-state connectivity reflect depressive rumination? A tale of two analyses. Neuroimage, 2014, 103: 267-279
参考文献 85
ChoiH, JinK H. Predicting cognitive decline with deep learning of brain metabolism and amyloid imaging. Behav Brain Res, 2018, 344: 103-109
参考文献 86
WestmanE, MuehlboeckJ S, SimmonsA. Combining MRI and CSF measures for classification of Alzheimer's disease and prediction of mild cognitive impairment conversion. Alzheimers Dement, 2011, 7(4): e16-e16
参考文献 87
KriegeskorteN. Deep Neural Networks: a new framework for modeling biological vision and brain information processing. Annu Rev Vis Sci, 2015, 1(1): 417
参考文献 88
AcharyaU R, OhS L, HagiwaraY, et al. Automated EEG-based screening of depression using deep convolutional neural network. Comput Methods Programs Biomed, 2018, 161: S0169260718301494
参考文献 89
LinE, KuoP H, LiuY L, et al. A deep learning approach for predicting antidepressant response in major depression using clinical and genetic biomarkers. Front Psychiatry, 2018, 9: 290
参考文献 90
JanA, MengH, GausY F A, et al. Artificial intelligent system for automatic depression level analysis through visual and vocal expressions. IEEE Transactions on Cognitive & Developmental Systems, 2017, PP(99: 1-1
参考文献 91
OrabiA H, BuddhithaP, OrabiM H, et al. Deep learning for depression detection of twitter users. Proceedings of the The Workshop on Computational Linguistics & Clinical Psychology: From Keyboard To Clinic. New Orleans: Association for Computational Linguistics, 2018: 88-97
参考文献 92
KangY, XiaoJ, YeY, et al. Deep transformation learning for depression diagnosis from facial images. Chinese Conference on Biometric Recognition ,2017, 13-22
参考文献 93
GeraciJ, WilanskyP, LucaV D, et al. Applying deep neural networks to unstructured text notes in electronic medical records for phenotyping youth depression. Evid Based Ment Health, 2017, 20(3): 83-87
参考文献 94
TranT, KavuluruR. Predicting mental conditions based on “History of present illness” in psychiatric notes with deep neural networks. J Biomed Inf, 2017, 75: S138-S148
参考文献 95
DwyerD B, FalkaiP, KoutsoulerisN. Machine learning approaches for clinical psychology and psychiatry. Annu Rev Clin Psychol, 2018, 14(1): 91-118
参考文献 96
MuellerS G, WeinerM W, ThalL J, et al. Ways toward an early diagnosis in alzheimer's disease: the alzheimer's disease neuroimaging initiative (ADNI). Alzheimers Dement, 2005, 1(1): 55-66
参考文献 97
PlisS M, HjelmD R, SalakhutdinovR, et al. Deep learning for neuroimaging: a validation study. Front Neurosci, 2014, 8(8): 229
参考文献 98
WintermarkM, ColenR, WhitlowC T, et al. The vast potential and bright future of neuroimaging. Brit J of adiol, 2018, 91(1087): 20170505
参考文献 99
GhafoorianM, KarssemeijerN, HeskesT, et al. Location sensitive deep convolutional neural networks for segmentation of white matter hyperintensities. Sci Rep, 2017, 7(1): 5110
参考文献 100
DolzJ, DesrosiersC, BenA I. 3D fully convolutional networks for subcortical segmentation in MRI: A large-scale study. Neuroimage, 2017, 170: 456-470
参考文献 101
HerrmannL, Le MasurierM, EbmeierK. White matter hyperintensities in late life depression: a systematic review. J Neurol Neurosurg Psychiatry, 2008, 79(6): 619-624
参考文献 102
Van UdenI W, TuladharA M, De LaatK F, et al. White matter integrity and depressive symptoms in cerebral small vessel disease: The RUN DMC Study. Am J Geriatr Psychiatry, 2015, 23(5): 525-535
参考文献 103
PapakostasG I, SheltonR C, KinrysG, et al. Assessment of a multi-assay, serum-based biological diagnostic test for major depressive disorder: a pilot and replication study. Mol Psychiatry, 2013, 18(3): 332-339
参考文献 104
JentschM C, BuelE M, Van, BoskerF J, et al. Biomarker approaches in major depressive disorder evaluated in the context of current hypotheses. Biomark Med, 2015, 9(3): 277-297
参考文献 105
ZhangW, LiR, DengH, et al. Deep convolutional neural networks for multi-modality isointense infant brain image segmentation. Neuroimage, 2015, 108: 214-224
参考文献 106
WuG, KimM, WangQ, et al. Scalable high-performance image registration framework by unsupervised deep feature representations learning. IEEE Trans Biomed Eng, 2016, 63(7): 1505-1516
参考文献 107
SukH I, LeeS W, ShenD. Hierarchical feature representation and multimodal fusion with deep learning for AD/MCI diagnosis. Neuroimage, 2014, 101: 569-582
参考文献 108
ShinH, RobertsK, LuL, et al. Learning to read chest X-rays: recurrent neural cascade model for automated image annotation. proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 2497-2506
参考文献 109
GreenspanH, Van GinnekenB, SummersR M. Guest editorial deep learning in medical imaging: overview and future promise of an exciting new technique. IEEE Trans Med Imaging, 2016, 35(5): 1153-1159
参考文献 110
KrogelM A, SchefferT. Multi-relational learning, text mining, and semi-supervised learning for functional genomics. Mach Learn, 2004, 57(1-2): 61-81
参考文献 111
VieiraS, PinayaW H L, MechelliA. Using deep learning to investigate the neuroimaging correlates of psychiatric and neurological disorders: methods and applications. Neurosci Biobehav Rev, 2017, 74(Pt A): 58-75
参考文献 112
PoldrackR A, BarchD M, MitchellJ P, et al. Toward open sharing of task-based fMRI data: the OpenfMRI project. Front Neuroinform, 2013, 7: 12
参考文献 113
Hosseini-AslE, GhazalM, MahmoudA, et al. Alzheimer's disease diagnostics by a 3D deeply supervised adaptable convolutional network. Front Biosci, 2016, 23: 584-596
参考文献 114
PanS J, YangQ. A survey on transfer learning. IEEE Trans Knowl Data Eng, 2010, 22(10): 1345-1359
参考文献 115
TajbakhshN, ShinJ Y, GuruduS R, et al. Convolutional neural networks for medical image analysis: fine tuning or full training? IEEE Trans Med Imaging, 2017, 35(5): 1299-1312
参考文献 116
GaoM, BagciU, LuL, et al. Holistic classification of CT attenuation patterns for interstitial lung diseases via deep convolutional neural networks. Comput Methods Biomech Biomed Eng Imaging Vis, 2018, 6(1): 1-6
参考文献 117
XuY, JiaZ, AiY, et al. Deep convolutional activation features for large scale Brain Tumor histopathology image classification and segmentation. proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, Brisbane. IEEE, 2015: 947-951
参考文献 118
LinderH P, JohnsonS D. The netflix recommender system: algorithms, business value, and innovation. Acm Transactions on Management Information Systems, 2015, 6(4): 1-19
参考文献 119
JohanssonR, SjobergE, SjogrenM, et al. Tailored vs. standardized internet-based cognitive behavior therapy for depression and comorbid symptoms: a randomized controlled tial. Plos One, 2012, 7(5): e36905
目录 contents

    摘要

    抑郁症是当今社会上造成首要危害且病因和病理机制最为复杂的精神疾病之一,寻找抑郁症的客观生物学标志物一直是精神医学研究和临床实践的重点和难点,而结合人工智能技术的磁共振影像(magnetic resonance imaging,MRI)技术被认为是目前抑郁症等精神疾病中最有可能率先取得突破进展的客观生物学标志物. 然而,当前基于精神影像学的潜在抑郁症客观生物学标志物还未得到一致结论. 本文从精神影像学和以机器学习(machine learning,ML)与深度学习(deep learning, DL)等为代表的人工智能技术相结合的角度,首次从疾病诊断、预防和治疗等三大临床实践环节对抑郁症辅助诊疗的相关研究进行归纳分析,我们发现:a. 具有诊断价值的脑区主要集中在楔前叶、扣带回、顶下缘角回、脑岛、丘脑以及海马等;b. 具有预防价值的脑区主要集中在楔前叶、中央后回、背外侧前额叶、眶额叶、颞中回等;c. 具有预测治疗反应价值的脑区主要集中在楔前叶、扣带回、顶下缘角回、额中回、枕中回、枕下回、舌回等. 未来的研究可以通过多中心协作和数据变换提高样本量,同时将多元化的非影像学数据应用于数据挖掘,这将有利于提高人工智能模型的辅助分类能力,为探寻抑郁症的精神影像学客观生物学标志物及其临床应用提供科学证据和参考依据.

    Abstract

    Depression is one of the most complex psychiatric diseases that cause the most serious harm in today's society. Searching for objective biomarkers of depression has always been the focus and difficulty of psychiatric research and clinical practice. Numerous studies have shown that magnetic resonance imaging (MRI) combined with artificial intelligence technology might be currently the most likely biologic marker to find breakthroughly in mental illness such as depression. However, the current potential objective biomarkers of depression based on psychiatric imaging have not been consistently concluded. From the perspective of combining psychoradiology with artificial intelligence technology represented by machine learning (ML) and deep learning (DL), this paper summarizes and analyses the related studies on depression from three components of the clinical practice including disease diagnosis, prevention and treatment for the first time. We found that a.the brain areas with diagnostic value are mainly concentrated in: precuneus, cingulate gyrus, inferior parietal lobule, insula , thalamus and hippocampus; b.the brain regions with preventive value are mainly concentrated in: precuneus, central posterior gyrus, dorsolateral prefrontal cortex, orbitofrontal cortex, middle temporal gyri; c.brain regions with predictive therapeutic response are mainly concentrated in: precuneus, cingulate gyrus, inferior parietal lobule, middle frontal gyrus, middle occipital gyrus, lingual gyrus. Future research can be improved by enlarging the sample size through multi-center collaboration and data transformation, and at the same time non-imaging data can be applied to data mining, which will help to improve the classification accuracy of artificial intelligence models, and provide scientific evidence and reference for the studies on exploring psychoradiological objective biomarkers for depression and its clinical application.

    抑郁症是导致非致命健康损失最主要的原因之一,其主要症状包括自我评价低、情绪低落、快感缺失,同时伴有认知、睡眠及饮食障[1]. 目前,抑郁症的防治工作依然十分艰巨,主要在于其发病机制不清、缺少客观的临床诊断标准. 抑郁症的临床症状具有高度异质性,但现今临床医师主要依据量表进行主观判断来诊断抑郁症,使得误诊或漏诊的现象频发. 为解决这一问题,基于大数据下的人工智能通过对已有的数据和关系进行重新组织,不断优化自身计算性能,实现了高精度的数据提取与准确分类,极大程度弥补了临床医师主观判断的不足.

    人工智能辅助识别抑郁症客观生物学标志物有赖于内表型的桥梁作用. 内表型作为基因型与表现型之间的中间标志,既避免了原发基因机制的高度争议以及患者临床症状的异质性,又能在基因与临床表现的关联分析中表现出较高的灵敏性. 以磁共振影像数据为代表的精神疾病内表型研究不仅能提供脑组织的结构信息,还可以刻画反映病变组织功能随时间变化的动态信[2]. 同时,其高分辨、非侵入性、时空连续性等优点也进一步确定了影像诊断技术在神经精神疾病中重要的临床价值. 因此,抑郁症内表型中最具有临床价值的便是精神影像学标志[3]. 精神影像智能分析一般采用深度学习和机器学习等人工智能技术,识别分析精神疾病患者的磁共振影像,并结合其他病历记录与精神疾病评估量表辅助做出诊断. 这种人工智能技术不仅能辅助医生作出较为准确的临床判断,而且大大降低了医生的工作负荷,使他们能去处理更棘手的病例,同时也可以帮助医疗机构提供更加个性化、整合化的医疗服[4].

    机器学习技术辅助识别精神影像学图片,使得脑部异常的定量测量成为可能,初步的研究已经充分验证了这类方法的有效性. 与此同时,另一种新兴人工智能算法——深度学习,凭借其数据驱动自动学习特征和所建立的深度模型优势受到了图形识别领域的青睐,在阿尔茨海默病、精神分裂症、多动症等神经精神疾病智能诊断中取得了初步研究进展,为抑郁症精神影像的深度学习研究提供了方法和途径. 鉴此,本文主要评价近年来机器学习及深度学习在抑郁症及其他神经精神疾病领域的研究,分析比较相关研究的一致性与差异性,总结分析目前人工智能算法研究局限,并对未来进一步研究提出合理建议.

  • 1 基于精神影像学的客观生物学标志物

    随着磁共振影像技术的飞速发展,特别是结构磁共振成像(structure MRI,sMRI)、功能磁共振成像(function MRI,fMRI)、扩散张量成像技术(diffusion tensor imaging,DTI)等先进成像技术的发展,为基础脑科学和应用脑科学研究领域的蓬勃发展提供了重要的技术支[5],也因而促进了精神影像领域的发[6]. 磁共振影像能够提供脑结构、功能和代谢等的区域化空间信息,以及详细的功能和结构连接图. 例如,sMRI能够显示高空间分辨率(1mm³或更少)的大脑结构,是研究大脑不同结构特征的理想影像模态,常用于检测脑区异常、病变和损[7]. DTI是唯一的活体非侵入地研究白质微结构的技术,通过对水分子扩散的建模,使不同脑区之间的解剖连接可视[8]. fMRI通过检测血氧变化监测颅内活动,以此来研究大脑的功能区域和网络以及它们之间的时间关[9]. 功能磁共振信号采集方式分为静息态扫描和任务态扫描. 任务态功能磁共振成像(task fMRI,tfMRI)是一种刺激诱发脑活动的扫描范式,作为传统方法被广泛应用. 但是近年来,静息态功能磁共振成像(rest fMRI,rfMRI)凭借分析方法多、不受任务设置影响的优点,正成为研究抑郁症的更有利武器. 在理论上,这些方法都可以作为抑郁症内表型被机器学习技术识别,从而能够定量测量脑部的结构和功能异常.

    通过以脑MRI为代表的影像学方法检测得到的生物学标志物可分为易感生物学标志物、疾病诊断生物学标志物和治疗生物学标志物,分别实现了抑郁症高危人群筛查、疾病诊断以及疗效预测. 目前的研究主要针对于疾病诊断标志物,涉及的脑区主要包括由杏仁核、海马、前额叶、纹状体腹内侧、苍白球和丘脑核团所形成边缘系统-皮层-纹状体-苍白球-丘脑环路(limbic-cortical-striato-pallidal-thalamic circuit,LCSPT环路)、皮层-边缘系统,以及默认网[10]. 这些区域出现的结构和功能改变与抑郁状态有密切关系,很有可能会成为高信度的疾病诊断标[11]. 易感生物学标志物也有了一些初步发现. 部分脑区异常可在抑郁发病之前就被发现,主要包括左侧海马和右侧海马旁回、右侧脑岛、杏仁核、腹外侧前额叶、背侧前扣带[12,13]. 这些区域往往是涉及抑郁症状发病机制的关键脑区,其体积减少和活动异常与更高的抑郁症发病风险有[14],这些研究成果为选取高预测性能的影像生物学标志物提供思路. 通过治疗前的影像特征预测疗效成为了近年来的研究热点,在接受治疗的患者当中,30%患者即使接受标准化的抗抑郁症治疗,仍然无法得到缓[15],具有预测能力的生物学标志物亟待开发. 与治疗效果较为相关的脑区为额叶边缘区域,尤其是前额叶、前扣带回、海马、杏仁核与脑岛,这可能与抗抑郁治疗能够逆转这些区域的病理过程有[16]. 更进一步的研究是在患者接受治疗之前检测这些脑区的预测能力,从而筛选出特异的治疗反应指标为个体患者制定治疗策略. 这些结果表明MRI影像能够直接反应抑郁症不同阶段的病理生理特征,而且可能成为抑郁症早期筛查、客观诊断和疗效评估的生物学标志物.

  • 2 基于人工智能的精神影像学研究新方法

    近年来,随着人工智能技术的发展,使得我们可以从群组水平研究生物学标志物转变为用脑影像在个体水平进行结果预[17]. 机器学习和深度学习可以客观准确地对海量多维的精神影像数据进行建模,从而量化了神经精神疾病引起的脑解剖与功能异常的程[8],有助于精神疾病诊断和预后生物学标志物的开发,从而有效辅助疾病的临床判断.

  • 2.1 机器学习

    机器学习是计算机科学领域中人工智能的一个主要部分,它使用统计技术给计算机“学习”的能力. 在没有明确编程的情况下,使用数据逐步提高特定任务的性[18]. 机器学习的核心是模式识别方法,主要流程包括特征提取和选择、分类模型构建、泛化能力测试等.

    目前较为常用的分类诊断模型有KNN分类诊断模型、Logistic回归分类诊断模型、支持向量机(support vector machine,SVM)分类诊断模型、高斯过程分类模型(Gaussian process classifiers,GPC)等. 其中,SVM目前是神经精神疾病研究中使用最广泛的模式分类模型. SVM的理论基础是结构风险最小化原则,它通过构造最大边缘超平面来搭建学习器,既保证分类器的精度又保证了较高的泛化能力. GPC是近年来机器学习领域十分关注的一类有监督的学习算法研究,具有计算精度较高、计算复杂度较低和稳定性较好等优[18]. 另外,交叉验证(cross validation,CV)是一种评估模型性能的重要方法,主要有K折交叉验证、留一法验证(leave-one-out cross-validation,LOOCV)、Holdout验证等,交叉验证的主要目的是减低模型潜在的过拟合风险,从而提高其在未知数据上的泛化能[19].

    机器学习对磁共振成像样本进行分类识别的整体过程如下:a. 磁共振成像首先帮助我们完成脑区域及网络结构、功能与连接的构建,然后进行相关模块指标与抑郁症组间差异分析,并选择出有显著差异的模块指标作为分类特征,将其按照贡献度进行排序,以方便进行接下来的分类模型构建. b. 利用降维技术对向分类器输入的数据进行特征选择与特征提取,从而剔除数据中的噪声并提高机器学习的性能. c.根据样本数量选择合适的交叉验证方法用于分类器泛化能力测试,从而避免学习过度的情况出[19].

    机器学习能够为找到精神疾病的神经脑区靶点提供有力的技术支持,可以归因于以下因素:a. 机器学习对多模态大样本影像数据的处理能力增强,解决了常规数据分析方法面对庞大数据无能为力的问题;b. 现代图形处理技术具有强大的并行处理能力;c. 优化技术的发展保障了学习结果的精确程度. 目前,机器学习加权因子等分析已经成为抑郁症特异生物特征多模态客观标志物分析的主要技术手[20].

  • 2.2 深度学习

    深度学习是机器学习系列方法中的子类别,它利用特定的网络结构以及训练方法对输入的图像学习有意义的表征,从而用于后续的图像分类. 深度学习可以是监督、半监督或无监[21]. 其主要流程(特征提取、选择、分类等)可以通过构建深度学习框架来实现,当前常用的深度学习模型包括:深度神经网络(deep neural network,DNN)、深度置信网络(deep belief network,DBN)[22]、递归神经网络(recurrent neural networks,RNNs)、深度卷积神经网络(convolutional neural networks,CNNs)[23]等.

    几个主要特征使深度学习对于机器学习领域的研究者们非常有吸引力. 首先,传统的机器学习方法需要医学专家对待处理图像特征进行人工干预设计,容易受工作者主观性影响并且医学专业化程度造成了机器学习专家介入困[24]. 相比之下,深度学习只需要一组经过简单预处理的数据就能能够进行数据驱动的自动特征学习,避免了以往相关特征筛选的主观[25]. 深度学习另一大特征就是模型的深度. 通过应用非线性层次结构,深度学习能够对非常复杂的数据模型进行建模,这是以往浅层模型无法实现的. 最后,上文已提及到的传统核心步骤(特征提取、选择、分类)可以在同一个最优化的深层结构中实现. 因此,在基于脑影像的疾病分类方面,深度学习模型能够在一定程度上提高计算机辅助诊断系统的准确率、灵敏度与特异度.

  • 3 基于机器学习的抑郁症精神影像学客观标志物的研究进展

    不同研究发现的抑郁症特征脑区大相径庭,其能否作为疾病的特征生物学标志物还需要进一步研究证实. 为此,笔者收集了近年来基于机器学习的抑郁症影像学研究,重点总结并列举其潜在的生物学标志物、机器学习方法、识别准确率、精神影像技术等,并按照预防生物学标志物、诊断生物学标志物及治疗生物学标志物的分类标准将文章分门别类(表1,2,3),展现研究成果及应用价值.

    表 1 基于精神影像和机器学习的抑郁症诊断生物学标志物研究汇总

    Table 1 Studies of diagnostic biomarkers for depression based on psychoradiology and machine learning

    参考文献疾病种类样本容量模式分类器交叉验证法特征脑区准确率/%
    Roberts et al., 2017

    BD

    HC=80,BD=49,BD高危组 = 71

    rsfMRI

    SVM

    不详

    腹外侧前额叶、内侧前额叶、额下回、脑岛、颞上回、豆状壳核

    64.30

    Sundermann et al., 2017

    MDD

    HC=180,MDD=180

    rsfMRI

    SVM

    CV

    不详

    47.50~53.60
    Hilbert et al., 2017GAD,MDDHC=24,GAD=19,MDD=14

    sMRI

    SVMs

    LOOCV

    额上回、额中回、额下回、前扣带回、尾状核、杏仁核

    68.05

    Ramasubbu et al., 2016

    MDD

    HC=19,MDD(轻微)=12, MDD(严重)=18,MDD(非常严重)=15

    rsfMRI,fMRI(情绪识别任务),sMRI

    SVM

    5-fold CV

    额极、额上回、额中回、额下回、眶额叶、内侧额叶、中央前回、前扣带回、脑岛

    52.00~66.00

    Rive et al., 2016

    MDD,BD

    BD=36,MDD=45

    sMRI, rsfMRI(DMN, SN, FPNs)

    GPC

    LOOCV

    内侧眶额叶、额中回、额前回、颞极、顶下缘角回、楔前叶、膝下前扣带回、海马旁回

    69.10

    Frangou et al., 2016BD,MDD家族史者HC=30,BD=30,MDD家族史者=30fMRI(工作记忆任务)

    GPC

    “leave 2 out”CV

    额上回、额中回、颞叶

    73.10

    Rosa et al., 2015

    MDD

    HC=19,MDD=19

    fMRI (性别区分与情感任务)

    SVM

    LOOCV

    额上回、颞下回、颞上回、顶上回、脑岛、前扣带回、壳核、丘脑、胼胝体78.90~85.00
    Sato et al., 2015

    MDD

    HC=21,MDD=25

    fMRI(社会观念任务)

    LDA

    LOOCV

    海马、丘脑和前脑岛

    78.10

    Shimizu et al., 2015

    MDD

    HC=31,MDD=31

    fMRI(言语流畅性任务)

    SVM

    10-fold CV

    舌回、小脑、楔前叶

    90.00~95.00

    Sacchet et al., 2015

    MDD,BD

    HC=61,BD=40,MDD=57, MDD(缓解)=35

    sMRI

    SVM

    不详

    杏仁核、伏隔核、尾状核、苍白球、壳核、丘脑、海马

    59.50~62.70

    Fung et al., 2015MDD,BDHC=29,MDD=19,BD=16

    sMRI

    SVM

    10-fold CV

    颞上沟、颞中回、楔前叶、顶上回、顶下回

    74.30

    Patel et al., 2015

    LLD

    老年HC=35,LLD=33sMRI,rsfMRI与DTI

    SVM

    LOOCV

    扣带回、内侧前额叶、前脑岛、楔前叶和顶下缘角回

    87.27

    Johnston et al., 2015

    TRD

    HC=21,TRD=20

    sMRI

    SVM

    LOOCV

    尾状核、脑岛、松果体缰和脑室周围灰质

    85.00

    Mwangi et al., 2015

    BD

    HC=16,BD=16

    DTI

    SVM

    LOOCV

    颞上沟、额极、海马旁回、颞横回、颞极、前扣带回

    78.12

    Wu et al., 2015

    UD

    HC=26,UD=25

    sMRI

    SVM

    LOOCV

    丘脑和颞极

    78.40

    Koutsouleris et al., 2015

    MDD,

    SC

    SC=158,

    MDD=104

    sMRI

    SVM

    CV

    额下回、脑岛、颞上回、颞极、小脑、脑室周围区域、顶叶、辅助运动区域

    76.00

    Cao et al., 2014

    MDD

    HC=37,

    MDD=39

    rsfMRI

    SVM

    LOOCV

    眶部额下回、缘上回、顶下缘角回、后扣带回、颞中回、颞下回

    76.60

    Guo et al., 2014

    首发UD

    HC=27,首发UD=36

    rsfMRI

    ANN

    CV

    海马、扣带回、豆状核和丘脑、顶下缘角回

    90.50

    Serpa et al., 2014

    MDD,BD

    HC-1组=33,HC-2组=38, MDD=19,BD=23

    sMRI

    SVM

    LOOCV

    不详

    54.6~66.1

    MacMaster et al., 2014MDD,BDHC=22,MDD=32,BD=14

    sMRI

    SVM

    CV

    不详

    81.00

    Chen et al., 2014

    BD

    HC=14,BD=9

    sMRI

    SVM

    LOOCV

    边缘叶、额叶、顶叶

    57.00

    Qin et al., 2014

    MDD

    HC=30,MDD=29

    sMRI,DTI

    SVM,GPC

    LOOCV

    背外侧额上回、颞下回、颞中回

    83.05

    Qiu et al., 2014

    MDD

    HC=32,MDD=32

    sMRI

    SVM

    LOOCV

    额中回尾部、前扣带回、颞极、颞中回、中央前回和枕外侧回

    78.00

    Dominik et al., 2014MDD,BDHC=22,MDD=22,BD=22fMRI(情绪识别任务)

    GPC

    LOOCV

    杏仁核的浅表核、杏仁核基底外侧

    79.60

    Rondina et al., 2014

    MDD

    HC=30,MDD=30

    fMRI(情绪识别任务)

    RFE-SVM

    CV

    额中回、额下回、颞中回小脑、基底神经节、前扣带回、眶额叶、基底神经节和梭状回

    67.00

    Wei et al., 2013

    MDD

    HC = 20,MDD=20

    rsfMRI

    SVM

    LOOCV

    腹内侧前额叶、额顶叶、内侧额上回、额中回、角回、扣带回、楔前叶、小脑

    90.00

    Grotegerd et al., 2013

    BD

    HC=10,MDD=10,UD=10fMRI(情绪识别任务)

    SVM

    LOOCV

    额下回、内侧额上回、眶部额上回、背外侧前额叶、杏仁核

    90

    Yu et al., 2013

    MDD

    HC=38,MDD=19

    rsfMRI

    SVM

    LOOCV

    内侧前额叶、楔前叶、扣带回、海马旁回、丘脑、颞下回、小脑

    80.90

    Almeida et al., 2013复发UD,BDBD=18,复发UD=18

    ASL

    SVM

    LOOCV

    膝下前扣带回

    81.00

    Miho et al., 2013MDD,SC

    BD=43,SC=41

    DTI,sMRI

    SVM

    不详

    丘脑、膝下前扣带回、脑岛、胼胝体和侧脑室

    88.00

    Lord et al., 2012

    MDD

    HC=22,MDD=21

    rsfMRI

    SVM

    CV

    岛盖部额下回、三角部额下回、脑岛、枕上回

    99.00

    Fang et al., 2012

    MDD

    HC=26,MDD=22

    DTI

    SVM

    LOOCV

    眶额叶、基底神经节、丘脑、海马和脑岛、顶上回、后扣带回、楔前叶、枕叶和颞下回

    91.70

    Mwangi et al., 2012

    MDD

    HC=32,MDD=30

    sMRI

    SVM,RVM

    LOOCV

    背外侧前额叶、内侧额叶、眶额叶、颞叶、脑岛、小脑

    90.30

    Zeng et al., 2012

    MDD

    HC=29,MDD=24

    rsfMRI

    SVM

    LOOCV

    前扣带回、内侧前额叶、基底神经节、颞叶、小脑

    94.30

    Gong et al., 2011

    MDD

    HC=23,TRD=23,非TRD=23

    sMRI

    SVM

    LOOCV

    额上回、额中回、辅助运动区、中央后回、颞中回、颞下回、缘上回、角回、梭状回

    58.7~84.6

    Hahn et al., 2011

    UD

    HC=30,UD=30

    fMRI(情绪识别任务,货币激励延迟任务)

    SVM,GPC

    LOOCV

    梭状回、尾状核、额叶

    83.00

    Mouraomiranda et al., 2011

    UD

    HC=19,UD=19

    fMRI(面部情感识别任务)

    SVM

    LOOCV

    楔前叶、顶下缘角回、前额下回、额中回、额下回、前扣带回、颞中回、枕中回、脑岛

    63~65.5

    Fu et al., 2008

    UD

    HC=19,UD=19

    fMRI(负性情绪识别任务)

    SVM

    LOOCV

    额上回、额中回、颞中回、顶下缘角回、舌回、海马、杏仁核、丘脑

    86.00

    Marquand et al., 2008

    MDD

    HC=20,UD=20

    fMRI(口头记忆任务)

    SVM

    LOOCV

    额上回、额中回、额下回、中央前回、颞上回、颞中回、顶上回、梭状回、枕叶、海马旁回、尾状核、小脑

    68.00

    注:fMRI:功能磁共振成像(functional magnetic resonance imaging);rsfMRI:静息态功能磁共振成像(resting-state functional magnetic resonance imaging);sMRI:结构磁共振成像(structural magnetic resonance imaging);DTI:弥散张量成像(diffusion tensor imaging);ASL:动脉自旋标记(arterial spin labelling);FA:各向异性分数(fractional anisotropy);TRD:难治性抑郁症(treatment-refractory depression);PD:惊恐障碍(panic disorder);AG:广场恐怖症(agoraphobia);HC:健康对照组(healthy control);BD:躁郁症(bipolar disorder);UD:单相抑郁(unipolar disorder);MDD:重度抑郁症(major depressive disorder);SC:精神分裂症(schizophrenia);GAD:广泛性焦虑障碍(generalized anxiety disorder);LLD:老年抑郁(late-life depression);ECT:电休克疗法(electroconvulsive therapy);SVM:支持向量机(support vector machine);RVM:相关向量机(relevance vector machine);ANN:人工神经网络(artificial neural network);GPC:高斯过程分类(gaussian process classifiers);LDA:最大熵线性判别分析(linear discriminant analysis);RFE:递归特征消除(recursive feature elimination);LOOCV:留一法交叉验证(leave-one-out cross-validation);CV:交叉验证法(cross validation);10-fold CV:10折交叉验证(10-fold cross-validation);5-fold CV:5折交叉验证(five-fold cross-validation);“leave 2 out”CV:留二法交叉验证(“leave 2 out”cross-validation);dDMN:背侧默认网络(dorsal default mode network);aSN:前突显网络(anterior salience network);FPNs:额顶叶网络(frontoparietal networks).

    表 2 基于精神影像和机器学习的抑郁症预防生物学标志物研究汇总

    Table 2 Studies of preventive biomarkers for depression based on psychoradiology and machine learning

    参考资料疾病种类样本容量模式分类器交叉验证法特征脑区准确率/%

    Ramasubbu et al., 2016

    MDD

    HC=19,MDD(轻微)=12,MDD(严重)=18,MDD(非常严重)=15rsfMRI,fMRI(情绪识别任务),sMRI

    SVM

    5-fold CV

    前额叶、脑岛和前扣带回

    58.00

    Sacchet et al., 2015

    MDD,BD

    HC= 61,BD=40, MDD=57,MDD(缓解)=35

    sMRI

    SVM

    不详

    尾状核、腹侧间脑

    59.50~62.70

    Foland-Ross et al., 2015

    MDD

    HC=15,MDD=18

    sMRI

    SVM

    10-fold CV

    内侧眶额叶、中央前回、前扣带回、脑岛

    70.00

    Hajek et al., 2015

    BD

    HC = 45,MDD = 45

    sMRI

    SVM,GPC

    "leave 2 out" CV

    前额叶区双侧白质束、扣带回、颞中回、楔前叶和枕叶

    72.22~70.37

    Lueken et al., 2015

    PD,AG(伴UD)

    PD/AG(不伴UD) = 33,PD/AG(伴UD)组=26

    fMRI(恐惧条件反射任务)

    随机欠采样树(random undersampling tree)

    LOOCV

    脑岛、背外侧前额叶

    73.00

    Macmaster et al., 2013

    UD

    HC=18,UD = 19

    rsfMRI

    Fisher逐步判别分析

    LOOCV

    眶额叶、前额叶、中央后回、颞下回、脑岛

    91.90

    注:fMRI:功能磁共振成像(functional magnetic resonance imaging);rsfMRI:静息态功能磁共振成像(resting-state functional magnetic resonance imaging);sMRI:结构磁共振成像(structural magnetic resonance imaging);PD:惊恐障碍(panic disorder);AG:广场恐怖症(Agoraphobia);HC:健康对照组(healthy control);BD:躁郁症(bipolar disorder);UD:单相抑郁(unipolar disorder);MDD:重度抑郁症(major depressive disorder);SVM:支持向量机(support vector machine);GPC:高斯过程分类(gaussian process classifiers);LOOCV:留一法交叉验证(leave-one-out cross-validation);CV:交叉验证法(cross validation);10-fold CV:10折交叉验证(10-fold cross-validation);5-fold CV:5折交叉验证(five-fold cross-validation);“leave 2 out”CV:留二法交叉验证(“leave 2 out”cross-validation).

    表 3 基于精神影像和机器学习的抑郁症治疗生物学标志物研究汇总

    Table 3 Studies of prognosis biomarkers for depression based on psychoradiology and machine learning

    参考资料疾病种类样本容量模式分类器交叉验证法特征脑区准确率/%
    Schnyer et al., 2017

    MDD

    HC=25,MDD=25

    DTI

    SVM

    LOOCV

    海马、扣带回、胼胝体、上纵束、上额枕束

    74.00

    Guo et al., 2017

    MDD

    HC=28,MDD=38

    rsfMRI

    SVM

    不详

    眶部额上回、颞中回、丘脑、舌回、楔叶、后扣带回

    97.54

    Redlich et al., 2016

    MDD

    HC=21,ECT与抗抑郁药物联合治疗组=23,抗抑郁药物治疗组=23

    sMRI,rsfMRI与DTI

    SVM,GPC

    LOOCV

    膝下扣带回

    78.30

    Wade et al., 2016

    MDD

    HC=33,MDD=53

    sMRI

    SVM

    LOOCV

    尾状核、苍白球后内侧

    89.00

    Patel et al., 2015

    LLD

    老年HC=35,LLD=33

    sMRI,rsfMRI与DTI

    SVM

    LOOCV

    背内侧前额叶、扣带回、颞叶、颞极、角回

    89.47

    van Waarde et al., 2015

    MDD

    MDD治疗缓解=25, MDD治疗未缓解=20

    fMRI

    SVM

    LOOCV

    背外侧前额叶、眶额叶和后扣带回

    84.50

    Liu et al., 2012

    MDD

    HC=17,MDD治疗未缓解=18,MDD治疗缓解=17

    sMRI, rsfMRI与DTI

    SVM

    LOOCV

    内侧额叶、额中回、前扣带回、中央前回、缘上回、楔前叶、侧舌回、枕中回、枕下回、颞中回

    82.90

    Gong et al., 2011

    MDD

    HC=23,TRD=23,非TRD=23

    sMRI

    SVM

    LOOCV

    额上回、额中回、额下回、中央后回、颞中回、顶上回、缘上回、角回、梭状回、豆状核

    58.70~84.60

    Mouraomiranda et al., 2011

    MDD

    HC=19,MDD=19

    fMRI(面部表情识别任务)

    SVM

    LOOCV

    额中回、额下回、前扣带回、楔前叶、顶下缘角回、颞中回、枕中回、脑岛

    63.00~65.50

    Nouretdinov et al., 2011

    UD

    HC=19,UD=19

    fMRI(悲伤情绪分级任务)

    SVM

    LOOCV

    前扣带回、后扣带回、眶额叶

    83.30

    Costafreda et al., 2009

    MDD

    HC=37,MDD=37

    sMRI, rsfMRI, DTI

    SVM

    LOOCV

    眶额叶、额中回、前扣带回、后扣带回、枕叶、海马旁回

    88.90

    Marquand et al., 2008

    UD

    HC = 20, UD = 20

    fMRI(口头工作记忆任务)

    SVM

    LOOCV

    额上回、额中回、前扣带回、颞上回、颞中回、颞下回、楔前叶、 舌回

    69.00

    注:fMRI:功能磁共振成像(functional magnetic resonance imaging);rsfMRI:静息态功能磁共振成像(resting-state functional magnetic resonance imaging);sMRI:结构磁共振成像(structural magnetic resonance imaging);DTI:弥散张量成像(diffusion tensor imaging);TRD:难治性抑郁症(treatment-refractory depression);PD:惊恐障碍(panic disorder);AG:广场恐怖症(agoraphobia);HC:健康对照组(healthy control);BD:躁郁症(bipolar disorder);UD:单相抑郁(unipolar disorder);MDD:重度抑郁症(major depressive disorder);GAD:广泛性焦虑障碍(generalized anxiety disorder);LLD:老年抑郁(late-life depression);ECT:电休克疗法(electroconvulsive therapy);SVM:支持向量机(support vector machine);GPC:高斯过程分类(gaussian process classifiers);LOOCV:留一法交叉验证(leave-one-out cross-validation);CV:交叉验证法(cross validation);10-fold CV:10折交叉验证(10-fold cross-validation);5-fold CV:5折交叉验证(five-fold cross-validation);“leave 2 out”CV:留二法交叉验证(“leave 2 out”cross-validation).

  • 3.1 诊断生物学标志物

    良好的诊断标志物应该与疾病共同存在并在一段时间内比较稳定,并可用于疾病诊断、判断疾病分期和亚[2]. 上文已述,在抑郁症的发病过程中很多脑区的结构和功能都发生了改变,其预测疾病的特异性及灵敏度如何尚需进一步研究. 目前相关研究的成像技术主要集中在sMRI、fMRI和DTI等,因此下面我们对以这三种成像技术为主的相关研究进行归纳总结和分析(表1).

  • 3.1.1 基于sMRI的诊断生物学标志物

    sMRI凭借其优越的软组织对比度、较高的空间分辨率和无限制的重复测量等优势已经成为诊断抑郁症等精神疾病很有潜力的工[26]. 目前通过sMRI研究得到的异常脑区主要集中在中脑边缘多巴胺系统,包括丘脑、尾状核、壳核、伏核、海马和杏仁核,以及额叶-皮层下结构中存在的功能障碍相关的功能回[27,28]. 这些脑区能否作为客观的诊断标志物用于临床还有待具体研究来评估其分类准确性.

    Mwangi[29]收集了30例重度抑郁症(major depressive disorder,MDD)患者和32例对照组的加权成像数据,并通过两独立样本t检验预先确定前额叶中灰质减少的区域作为子集,从而有效地去除了提供冗余信息的其他脑区. 以此成像数据作为SVM和相关向量机(relevance vector machine,RVM)的学习数据来源,对病人和健康对照组进行区分,得到了较高的分类准确性(90.30%)[29]. 另一项研究对40例双向抑郁(bipolar disorders,BD)患者、57例MDD患者、35例MMD缓解者和61例健康人群的T1加权成像数据进行自动灰质分割,再利用多元方差分析识别出具有显著组间差异的神经区域主要集中在左右侧尾状核和腹侧间脑,通过线性SVM联合递归特征消除方法(recursive feature elimination,RFE)对这些区域的体积逐个进行分类,准确度仅维持在62.44%~62.76%[30]. 同样是分析T1加权的磁共振数据,两项研究的准确性存在较大差异,可能是由于降维和分类的方法不同. 这些结果均表明机器学习技术可以对来自个体受试者的脑部扫描数据做出相对准确的预测,在整体准确性进一步提高的情况下,机器学习技术有望为临床选取诊断标志物提供有力的信息支持. 此外,机器学习加权因子也可以反映基于脑结构异常的MDD疾病严重程[29].

  • 3.1.2 基于fMRI的诊断生物学标志物

    相较于任务态fMRI,静息态磁共振成像(rsfMRI)以其设计简单、不易引起行为模式的选择和对任务执行能力要求较低的优势,在脑结构功能损伤程度更严重的患者和儿科人群显得更为实[31].

    最近,基于rsfMRI的研究已经发现了脑功能网络的离散畸变可能是抑郁症发展的原因,功能连接异常主要集中在情感和认知脑区域(前扣带回、杏仁核、苍白球和内侧丘脑),暗示皮层-边缘情绪调节回路异常同[32]. Zeng[33]分析24例抑郁症患者和相匹配的29例健康被试静息状态下的功能连接模式,将扣带回分成两个亚区,将亚区和所有其他体素子区域之间的功能连接作为分类特征,开发出一种基于最大间隔聚类(maximum margin clustering,MMC)的无监督机器学习方法,在缺乏临床信息的情况下精确(94.3%)识别出了MDD患者. 之后,他们还检测了后扣带回、前外侧额叶以及辅助运动区皮层的分类能力,均不足80%,从而提示我们膝下前扣带回(subgenual anterior cingulate cortex,sACC)在MDD病理生理过程中起主要作用.

    除了对静息状态网络功能连接横断面研究,还有研究涉及网络随时间长期变化的自相关性. 赫斯特指数(Hurst exponent,H)作为时间动力学“长期记忆”(long-term memory)的量度,可以很好地描述自发脑电活动的无标度特[34]. Wei[35]为了检测20例MDD患者和20例健康被试静息状态网络的“长期记忆”,选取12个时间过程的H指数作为分类特征,利用基于线性核函数的SVM成功以90%的准确率完成MDD患者与健康人群的分类. H指数在额叶-顶叶和默认网络明显降低,表明这些区域的功能状态受患者长期负性情绪的影响而不再稳定,为我们探索MDD的发病机制提供了新思路.

    目前,基于rsfMRI的机器学习研究相较于sMRI展示出了较高并且稳定的(90%左右)分类能力. 这提示我们,相较于结构所反映出的脑区异常,脑网络内部或之间的功能连接异常是否更接近抑郁症发病的病理生理学机制,从而更有望成为抑郁症诊断的客观标志[32]. 另外,Zeng和Wei[33,35]的研究分别向我们引进的新的分类方法和特征,两者的组合是否能产生识别能力更高的模型值得进一步探索.

  • 3.1.3 基于DTI的诊断生物学标志物

    DTI是一种非侵入性地追踪脑白质纤维束和脑认知功能变化的工具,可以用于脑部白质纤维束的完整性,更有利于反映抑郁症患者脑部系统水平的多维度障[36]. 目前,基于DTI的神经影像学研究发现MDD患者的额叶-边缘网络和额叶-顶叶网络存在功能障碍,继而可能会产生消极情绪症[37].

    Fang[38]利用基于DTI的白质纤维束成像技术(WM tractography)提取了22例首发未用药患者和26例健康被试的全脑解剖网络,并通过两样本t检验、局部线性嵌入(local linear embedding,LLE)联合SVM机器学习方法分析其全脑解剖连接模式,最终使用留一法验证其分类准确性高达91.7%. 结果发现组间差异最明显的网络集中在皮层-边缘系统网络,特别是额叶-边缘网络,患者在此处的连接性明显增加. 大多数基于DTI的研究主要关注体素内各向异性的分析,而此研究根据全脑的区域解剖连接获得了更高的准确性,这提示白质纤维束成像技术的辅助诊断潜[39,40]. Qin[37]根据29名抑郁症患者和30位健康被试者的DTI数据构建了白质网络并确定了他们的网络中枢节点(central nodes)作为分类特征,在降低特征维数后,使用具有径向基和函数(radial basis function,RBF)的非线性SVM从健康者中识别出抑郁症患者,准确率为83.05%. 结果显示,抑郁症患者中网络中枢节点普遍存在丢失,主要集中分布于额叶-顶叶网络中,另外颞中回也存在部分各向异性(fractional anisotropy,FA)降低. 虽然该项研究的准确率不如前者,但其对于中枢节点丢失的结论可以解释患者普遍存在的网络完整性和功能异常,提示我们以中枢节点作为额外的特征用于分类.

    Fang和Qin等得到的异常网络与之前的精神影像学结论一致,但由于目前基于DTI的机器学习研究较少,样本量较低,要使用DTI确定疾病的诊断标志物,还需要进一步研究.

  • 3.1.4 诊断生物学标志物的研究小结

    通过我们对诊断生物学标志物类研究(表1)的归纳整合发现,具有诊断价值的脑区主要集中在楔前叶、扣带回、顶下缘角回、脑岛、丘脑以及海马等. 其中,扣带回和楔前叶作为诊断标志物研究的热点脑区,展现出的诊断分类能力在不同研究中具有较高的异质性,可能与不同研究所纳入的样本例数、模态、分类器、交叉验证方法等不同有关. 总体来说,功能MRI研究的分类能力普遍高于结构MRI研究,但DTI作为一种特殊的脑结构研究技术,展现出了良好的应用前景,需要更多的研究进一步探索脑白质纤维束的分类能力和客观标志物.

  • 3.2 预防生物学标志物

    预防生物学标志物是指那些能够增加发病风险的脑部异常特征,它们存在于疾病发生前,往往通过研究MDD患者的后代以及临界状态的高危人群来发[41]. 这些标志物的研究有助于对抑郁症高危人群的及时发现和预防性治疗,有效降低疾病发病率. 目前神经影像学中面临的重要挑战就是利用机器学习方法检验这些具有预测能力的影像生物学标志物的准确性(表2).

  • 3.2.1 基于sMRI的预防生物学标志物

    目前认为与抑郁症风险因素相关的异常脑区分布较为分散,包括前额叶、中央前回、扣带回、脑岛以及尾状核等,其增加疾病发生风险的机制还未澄[42].

    Folandross[42]试图将基线条件下的皮层厚度作为预测未来抑郁症发病情况的分类特征,为此他们选取33名从未罹患抑郁症状的青少年进行为期5年的随访,使用SVM来测试基线条件下的皮层厚度是否能够可靠地区分将来发生抑郁症状的青少年与未患任何轴Ⅰ障碍的健康人群. 使用交叉验证评估分类器性能,其总体准确度约为70%,并发现右眶额内侧灰质厚度减少、双侧脑岛灰质厚度增加与抑郁症易患性相关. 该结论与以往的研究存在较多矛[43,44],可能与其样本量和随访时间有关. Hajek[45]招募了45名BP先证者的亲属作为BP的易感人群并配以45名健康者作为对照,分别提取被试sMRI中的白质和灰质体积作为分类标准,并选用线性核SVM和GPC作为分类方法,结果表明两种分类方法的准确度相似(SVM:68.9%;GPC:65.6%),其中贡献大的区域包括额叶下中部、颞叶下中部和楔前叶的白质. 由于被研究者的疾病反复发作、合并症以及使用药物情况均会掩盖与风险相关的脑部变化,导致目前预测准确性较低,与Hajek类似的研究对于BD易感人群的分类准确率普遍低于70%[46,47].

    目前sMRI作为预防生物学标志物普遍存在支持向量机分类性能还不够高、特征脑区未达成共识等问题,因此相关研究成果还无法应用于临床,未来研究的一个重要方向是阐明这些灰质结构的异常改变导致发病风险增加的机制.

  • 3.2.2 基于fMRI的预防生物学标志物

    很多研究报道抑郁高危人群存在背外侧前额叶、内侧前额叶和眶额叶自发活动的局部一致性(regional homogeneity,ReHo)异[48]和任务状态下脑区激活减[49]. 前额叶是认知执行和情绪调节障碍这些病理生理学过程中的主要受累区域,其作为预测抑郁症的生物学标志物有待进一步验证.

    Ma[50]使用rsfMRI检测19名阈下抑郁的高危患者和18名健康人群自发活动ReHo的变化,通过ROC曲线和Fisher逐步判别分析,分别评估ReHo指数在区分高危患者与健康人群的敏感性和特异性. 结果显示,高危患者在右眶额叶、左背外侧前额叶、左中额叶和颞下回旋显示较低的ReHo,而双侧脑岛和右背外侧前额叶的ReHo更高,预测准确度为91.9%. 除该研究外,ReHo指数目前已经成功应用于许多神经精神疾病研[51,52],其所具有的探索性分析能力非常有助于我们深入分析知之甚少的静息态大脑活动. 抑郁症常作为广场恐怖(agoraphobia,AG)的并发症而发[53],为探究机器学习在预测个体水平上AG患者并发抑郁症状的潜力,Lueken[54]选取59例AG患者(抑郁共病状态者26例)作为被试进行fMRI成像,使用随机欠采样集成分类器在留一法交叉验证框架中预测抑郁状态. 并发抑郁症者在处理安全信号过程中脑区活动明显减低,以背外侧前额叶和脑岛为主,以此作为预测指标,准确率为73%. 所得到的准确性虽然不能应用于临床,但Lueken等的发现提示我们,患者的伴随疾病往往会给精神影像学研究的结论带来偏倚,从而导致不同研究结论的异质性.

    目前,基于高危患者的预防标志开发均存在一个共同的问题,即我们无法预测易感者的脑区异常也会出现在后期的临床抑郁症中. 另外,鉴于这些研究均提到了前额叶在预测抑郁状态中的作用,未来可以进一步围绕这一脑区开展纵向研究来验证其预测能力.

  • 3.2.3 基于DTI的预防生物学标志物

    脑白质纤维束异常及各个脑区结构功能的改变可以解释抑郁症发生发展的病理生理机制. Jia[55]利用DTI研究52例MDD患者(其中36例有自杀未遂史)的部分各向异性(fractional anisotropy,FA)与自杀之间的关系,他们发现内囊左侧前肢FA降低与患者的自杀倾向有关,其他研究还发现患者临床症状的变化与FA存在线性关系. 未来的工作应开展队列研究处理纵向结果,最好是利用机器学习的建模方法识别发病前或早期的易感者DTI指标的区别性变化.

  • 3.2.4 预防生物学标志物的研究小结

    通过我们的归纳整理(表2)发现,具有预防价值的脑区主要集中在楔前叶、中央后回、背外侧前额叶、眶额叶、颞中回等. 与诊断生物学标志物类似,楔前叶在预测疾病发生发展方面也具有较高的分类价值. 相比于诊断和接下来要提及的治疗生物学标志物,关于预防标志物的研究仍然处于初步探索阶段:目前基于结构/功能的单模态MRI预测能力普遍不高,可能与早期生物学标志物的微小异常不易鉴别有关,联合多模态更有助于探索具有预测病情发展能力的标志物;相较于结构MRI的研究结果,功能MRI的研究结果更有助于达成共识;另外,开展纵向研究验证预防标志物的稳定性与进展情况,可以为预测疾病提供有价值的依据.

  • 3.3 治疗生物学标志物

    基于脑结构和功能的治疗生物学标志物可以预测患者对各种治疗的反应情况,从而帮助医生为患者选择有效的治疗手[16]. 大量的回顾性研究主要通过比较应答者与非应答者的脑影像组间差异来发现与治疗反应有关的特征标[56]. 然而,在将此类研究结果应用于临床之前,需要使用机器学习等数据挖掘技术来验证其预测治疗反应的准确性(表3).

  • 3.3.1 基于sMRI的治疗生物学标志物

    目前的研究认为以扣带回和额叶为主的灰质密度与患者治疗反应有[57,58]. Costafreda[57]分别选取18名接受药物治疗(9例缓解/无效)和12名接受认知行为疗法(cognitive behavioural therapy,CBT)的患者(6例缓解/无效),通过方差分析(analysis of variance,ANOVA)对其sMRI数据进行特征选择,以训练SVM. 临床症状缓解较好的患者表现为右前/左后扣带回、左额叶回和右枕后回灰质密度增高,以此来预测患者对药物治疗反应情况的准确性为88.9%,而对CBT的预测能力较低. 由于抑郁症患者的结构障碍相比于其他疾病往往较轻,使得基于sMRI的结构异常对CBT预测能力不佳,可以改为分析特异性更高的fMRI图像来进一步探[59].

    Nouretdinov[58]利用正形变换预测器(transductive conformal predictor,TCP)作为机器学习方法在前后扣带也得到了同样的结论,同时发现前额叶灰质密度增加反而会预示治疗后的残余症状. TCP应用于精神病学分类的优点在于它们可以衡量分类器预测的准确程度,从而帮助我们规避机器学习实际应用中出错的风[60]. Gong[61]共纳入了61名抑郁症患者和42例健康被试,采用线性SVM来检查治疗前sMRI数据的预后价值,发现白质体积也显示出了预测药物治疗反应的能力,但辨别力不如灰质(白质:65.22%;灰质:69.57%),其中右侧额中回、左侧旁中央小叶和右侧颞中回的体积增加暗示好转. 这些结果表明,灰质和白质密度在研究患者对药物反应中均具有预后能力,为使用结构脑影像预测药物治疗反应提供了初步证据.

  • 3.3.2 基于fMRI的治疗生物学标志物

    相较于正常人群,抑郁症患者在处理悲伤面孔和言语任务时全脑激活模式会表现出明显差[62,63]. 在这一理论基础上,Mourão-Miranda等定义了单人正常脑激活模式的神经影像数据边界,随即纳入19例患者,利用核函数与单类支持向量机(one class support vector machine,OC-SVM)[64]结合的模式识别与分析方法完成对其fMRI影像数据的分类. 该项研究表明,越偏离该边界的患者则越不容易对药物治疗产生反[8],预测准确率为65.5%(敏感度:79%,特异度:52%). 这一研究准确性较低的原因在于OC-SVM不具有特异性识别某种疾病的能力,但这一方法可以量化单个人脑激活模式与正常模式分布边界的偏差,接下来的研究可以用这一框架来定义不同疾病的大脑活动模式.

    Marquand[63]选取参与纵向治疗研究的20例患者接受言语记忆任务并进行功能磁共振影像扫描,把经过主要成分分析进行降维的数据用于训练SVM,最终发现前扣带回活动性增加提示良好的临床反应,准确率为67%(敏感度:56%,特异度:78%). 这一结果进一步验证了任务状态下扣带回的功能活动与治疗反应相关的重要意[65].

    相较于基于悲伤面孔和言语任务的fMRI成像在诊断生物学标志物研究中所取得的临床进[66,67],同类治疗生物学标志物目前仅表现出了统计学意义,但应用于临床仍需要准确性更高的研究结果. 针对单一模态预测准确性较低的问题,Patel[68]采用多模态磁共振联合方法,以较高的准确率预测了患者的治疗反应.

  • 3.3.3 基于DTI的治疗生物学标志物

    基于脑区域和全脑网络的多模态MRI的检测可以更好地预测的治疗反应. Patel[68]在对33例老年抑郁症患者进行多模态扫描后(FLAIR、DTI、rsfMRI等),利用来自前突显网络(anterior salience network,aSN)的DTI纤维束(track)数量和背侧默认网络(dorsal default mode network,dDMN)的rsfMRI功能连接性指数构建交替决策树模型. 基于该模型,他们发现aSN和dDMN区域白质完整性更低(即该区域的结构连接更少)以及dDMN功能连通性更低的个体更有可能成为治疗反应阳性者,预测准确率为89.47%. 该项研究不仅展示了多模态影像数据优良的预测能力,而且引进了一种可以替代传统SVM的学习方法——具有嵌入特征选择能力的决策树与逻辑回归可用于研究高维数据世界中的非线性关系,更加有助于发现预测治疗反应的特征标[69].

  • 3.3.4 治疗生物学标志物的研究小结

    通过我们的归纳整理(表3)发现,具有预测治疗反应价值的脑区主要集中在楔前叶、扣带回、顶下缘角回、额中回、枕中回、枕下回、舌回等. 其中值得注意的是,楔前叶、扣带回不仅仅是诊断和预防生物学标志物,也是治疗生物学标志物的热点脑区,表明楔前叶、扣带回是抑郁症客观生物学标志物中重要的候选靶点脑区. 与诊断和预防生物学标志物不同的是,当使用结构MRI作为预测治疗反应的生物学标志物时,更容易获得较高的预测准确值. 个性化治疗方案需要比目前研究更高的准确度,尝试把sMRI和fMRI结合的影像数据用于同一套SVM方法可能会在抑郁症中实现更高的预后准确性.

  • 4 基于深度学习的精神疾病客观生物学标志物研究进展

  • 4.1 深度学习在精神疾病客观生物学标志物研究中的应用

    近年来,深度学习方法在语言识别、自然图像分类和文本挖掘等各个表征学习和分类的领域取得了显著的进步. 虽然这一技术在精神影像学领域尚处于初始阶段,但在精神分裂[70,71]、自闭[72,73]、阿尔茨海默[74,75,76,77]等疾病辅助诊断中已经显示出了有希望的结果(表4). 相较于机器学习,深度学习表现出了更高的分类准确率,更加有可能在寻找基于图像的精神病学生物学标志物方面取得根本性进展. 与机器学习联合精神影像学的临床应用相类似,应用深度学习联合sMRI、fMRI、DTI等的研究同样基于疾病早期预[72,78]、精确诊[79]以及临床治疗反应预[80]这三方面来探寻潜在的影像生物学标志物.

    表4 深度学习在抑郁症等精神疾病客观标志物研究中的应用

    Table 4 The application of deep learning in studies of prognosis biomarkers for mental illness including depression

    参考资料疾病种类样本容量数据类型深度学习方法准确率/%
    Choi & Jin, 2018

    AD

    HC=182,AD=139,MCI=171

    PET[角回、颞叶、后扣带回的FDG摄取量]

    CNNs

    84.00

    Islam & Zhang, 2017

    AD

    AD=416

    sMRI

    CNNs

    73.75

    Suk et al., 2016

    AD

    HC=31,SC=31

    rsfMRI

    DAE

    72.58~81.08

    Suk et al., 2016

    AD/MCI

    HC=52,AD=51,MCI=99

    sMRI[GM体积]与PET,3种CSF标志物:Aβ42、t-tau、p-tau

    MTL

    90.27

    Han et al., 2015

    ADHD,AD

    不详

    fMRI

    HCSAE

    65.00~80.00
    Deshpande et al., 2017

    ADHD

    HC=744 ADHD(不注意型)= 173,ADHD(混合型)=260

    rsfMRI

    FCCANN

    90.00~95.00
    Kuang & He, 2015

    ADHD

    HC=95,SC=115

    rsfMRI

    DBN

    48.90~72.70

    Hazlett et al., 2017

    ASD

    HC=117,ASD高风险组=248,ASD = 70

    sMRI[全脑体积:GM +WM、CSF、大脑、小脑和侧脑室]

    DAN

    94.00

    Heinsfeld et al., 2018

    ASD

    HC=530,SC=505

    rsfMRI

    DNN

    70.00

    Acharya et al., 2018

    MDD

    HC=15,MDD=15

    EEG

    CNNs

    95.49

    Lin et al., 2018

    MDD

    MDD=455

    遗传和临床行为数据

    MFNNs

    80.60

    Orabi et al., 2018

    MDD,PTSD

    HC=572, MDD= 327, PTSD=246

    非结构化文本数据

    CNNs

    87.90

    Geraci et al., 2017

    MDD

    MDD=366(存在共病=735, 不存在共病=126)

    EMRs分析

    NLP,DNN

    70.37

    Tran & Kavuluru, 2017

    焦虑、抑郁、惊恐障碍、ADHD等11种精神障碍患者

    986例精神障碍患者(ADHD 41%, 焦虑68%, BD 33%, 痴呆27%, 进食障碍31%, 悲伤27%, OCD/OCSD 34%, 惊恐障碍47%, 精神错乱25%, PTSD 38%)

    病史文字描述

    DNN

    63.14

    Kim et al., 2016

    SC

    HC=50,SC=50

    rsfMRI

    DNN

    85.80

    Pinaya et al., 2016

    SC

    HC=83,SC=143,首发SC=32

    sMRI[GM厚度和体积]

    DBN

    73.60

    Ulloa et al., 2015

    SC

    HC=191,SC=198

    sMRI[GM 体积]

    MLP

    75.00

    注:PET:正电子发射型计算机断层显像(positron emission computed tomography);fMRI:功能磁共振成像(functional magnetic resonance imaging);rsfMRI:静息态功能磁共振成像(resting-state functional magnetic resonance imaging);sMRI:结构磁共振成像(structural magnetic resonance imaging);EEG:脑电图(electroencephalogram);SNPs:单核苷酸多态性(single nucleotide polymorphisms);EMRs:电子医疗记录(electronic medical records);AD:阿尔茨海默病(Alzheimer's disease);MCI:轻度认知障碍(mild cognitive impairment);ADHD:多动症(attention deficit hyperactivity disorder);ASD:自闭症(autistic-spectrum disorder);PTSD:创伤后应激障碍(post-traumatic stress disorder);SC:精神分裂症(schizophrenia);OCD:强迫障碍(obsessive compulsive disorders);OCSD:强迫症(obsessive compulsive spectrum disorder);BD:双相障碍(bipolar disorder);CNNs:深度卷积神经网络(convolutional neural networks);DAE:去噪自动编码(denoising autoencoders);DAN:深度自动编码网络(deep autoencoder networks);HCSAE:层叠卷积稀疏自动编码器(hierarchical convolutional sparse auto-encoder);FCC ANN:全连通级联人工神经网络(fully connected cascade artificial neural network);DBN:深度置信网络(deep belief network);DNN:深度神经网络(Deep Neural Network);DTL:深度转化学习(deep transformation learning);MFNNs:多层前馈神经网络(multilayer feedforward neural networks);MLP:多层感知器(multilayer perceptron);MTL:多任务学习(multi-task learning);NLP:自然语言处理(natural language processing);GM:灰质体积(gray matter);WM:白质体积(white matter);CSF:脑脊液体积(cerebrospinal fluid).

    在深度学习识别脑影像的实验中,研究数量最多的仍然是疾病诊断. 人工神经网络(artificial neural network,ANN)在使用fMRI数据的疾病分类中很流[70]. ANN通过模仿大脑神经突触的结构对信息处理的过程,构建一个由大量简单原件相互连接而成的复杂网络,实现了复杂的逻辑操作和非线性关系判[81]. Deshpande[82]从ADHD-200计划中选取744例健康儿童和433例多动症患儿(attention deficit hyperactivity disorder,ADHD)作为研究对象,并构建了全连通级联人工神经网络模型(fully connected cascade artificial neural network,FCC ANN)用于执行被试fMRI数据分类. 其在ADHD亚型之间分类准确率高达95%,并提出了ADHD相关的病理生理学见解:左侧眶额叶和小脑区之间的连通性减少是最具辨别力的分类特[79]. 与SVM相比,该模型展示了更强大的分类能力,同时Deshpande等还提出在ADHD中以有向网络作为分类特征有助于获得更高的准确率. 传统多层神经网络基于标准的反向传播算法,容易出现过度拟合的情[22]. DNN突破了这一局限性,相较于SVM和其他传统模型实现了更强大的分类性能. Kim[83]分别纳入了50例精神分裂症(schizophrenia,SZ)患者和50例健康人并获得了他们的fMRI数据,将其用于训练经L1范数正则化的DNN模型,分类犯错率为14.2%(SVM:22.3%). 相较于正常人,SC患者的小脑和皮层下区域的功能连接性明显降[70]. Berman[84]的研究也表明,抑郁症也是以异常功能连接模式为特征的精神疾病,同样适合于DNN分类器.

    深度学习辅助发病预测的研究目前较少,但其表现出的预测准确性普遍高于传统的机器学习方[85]. 深度CNNs凭借其应用的简易性在图像辨别领域展现了推广潜力. Choi[85]纳入139名阿尔茨海默病患者、171名轻度认知障碍者和182名健康人,并将深度CNNs应用18氟标记的荧光脱氧葡萄糖18F-fluorodeoxyglucose,FDG)PET图像,联合fMRI来预测轻度认知缺陷(mild cognitive impairment,MCI)患者加重发展为阿尔茨海默病的可能性,预测认知衰退的准确率为84%. CNNs可以分层方式迭代地过滤出无特征信息,联合多模态信息可以很大程度上提高预测准确[86]. Hazlett[72]发现婴儿早期大脑皮层面积多度生长可以作为自闭症(autistic disorder,ASD)早期诊断的生物学标志物,他们纳入了6~12月的106例家族高风险和42例低风险的婴儿,并设计了一种基于CNNS的算法,利用婴儿大脑皮层面积预测未来24个月是否会发生ASD,获得了81%的敏感性和88%的特异性,那些后来被诊断为ASD的婴儿左额上回、中央后回、顶叶脑回的皮层增长速度更快. 另外,Hazlett[72]还提出目前基于高风险人群的疾病预测可能会夸大预测准确性,更客观的数据应该从普通人群中获得. 目前深度学习结合脑影像应用于治疗反应预测的研究只有一例,但其存在样本量过小(70例)、缺少正则化(regularization)过程、过拟合严重等问题,导致其最终准确率仅为57%,低于SVM[80],该研究存在的方法缺陷可能会掩盖深度学习的分类能力,需要我们谨慎看待.

    总体而言,通过从高维精神影像数据中提取隐藏模式,深度学习是一种非常有前景的揭开精神疾病发病机理的工[87]. 然而,这一方法尚未应用于抑郁症的精神影像学研究领域,其可能原因和解决方法会在下文分析. 其他研究针对于与抑郁症类似的其他脑部功能连接异常疾病,为我们提供了初步思路,鉴于fMRI是以上研究中的主要成像方法,在样本量足够的情况下,可以率先应用于DNN和CNNs分类器中.

  • 4.2 深度学习辅助诊断抑郁症的研究现状

    目前,使用深度学习分析抑郁症患者影像数据的研究较少,更多的研究利用多模态混合深度学习框架综合分析文字、图像和视频等信息检测具有抑郁症迹象的用[88,89,90,91,92,93,94]表4). Acharya[88]利用15例抑郁症患者和15例健康人的脑电图(electroencephalogram,EEG)信号对CNNs模型进行训练,成功完成了疾病分类,左脑和右脑EEG信号识别准确率分别为93.5%和96.0%. EEG以其便捷、经济、同时能够动态监测患者一段时间内脑活动变化的优势展现了在抑郁症中的应用前[88]. 但该项研究的样本量过低,难以保证研究结果不存在较大偏倚,应用于临床前还需要进一步扩大数据集进行验证. Orabi[91]从Twitter社交媒体平台提取了327例抑郁症患者、246例创伤后应激障碍者和572例健康人群的非结构化文本数据,选用在自然语言处理领域最流行的两种深度学习方法CNNs来检测具有抑郁症倾向的用户,最高准确率达87.9%,这也是除了图像识别领域外,CNNs在自然语言处理领域的一大进展. 以上研究取得的惊人成果都为多模态抑郁症标志物的开发提供了新思路.

    深度学习尚未在抑郁症影像诊断方面广泛应用的原因可能与其需要的更大样本量有关,在数千个超参数组合的背景下,需要大量数据来增加细节以避免高维度和过拟合造成的实验失[95]. 目前多数深度学习研究选用MCI与AD患者作为研究对象,可能是因为ADNI数据库的使用便捷,该数据库涵盖数以千计MCI与AD患者的影像资源,并且免费开[96]. 另外,深度学习模型训练耗费的计算资源和时间也导致了其不能成为常用的分类方[70]. 不过,就目前深度学习应用于以上精神疾病的效果来看,在检测神经成像数据以及临床/认知信息类数据之前的抽象关系时,深度学习表现出了较传统SVM更高的分类性[97]. 抑郁症所具有的病因及发病机制不清、心理和社会因素多方面参与的特点使其各种模态之间的关系可能发生在更深层的概念层面,其数据特点非常适合深度学[98].

    未来的工作可致力于降低时间和资源复杂度,并将核函数矩阵与聚类算法或深度学习相结合,从而获得更复杂的新数据来弥补样本[98]. 另外,将深度学习解剖分割技术作为辅助手段应用于抑郁症诊断中也值得进一步研[99,100]. Ghafoorian[99]使用含有解剖位置信息的CNNS网络成功分割除了大脑sMRI和FLAIR图像白质高信号(white matter hyperintensities,WMH)数据集,相较于正常人,WMH在抑郁症患者中更容易出[101,102],这一研究提示我们根据更准确的WMH进行容量评估,可以更好地与身体表现和临床认知相关联,从而帮助对疾病的诊断.

  • 5 总结与展望

    针对抑郁症疾病客观生物学标志物这一临床难题和科研热点问题,结合机器学习和深度学习为代表的人工智能技术,本文从诊断生物学标志物、预防生物学标志物和治疗生物学标志物等三大临床实践环节总结和评述近年来抑郁症精神影像生物学标志物研究的最新研究成果. 研究表明,近年来机器学习及深度学习技术在抑郁症以及其他神经精神疾病的辅助诊断、预防与治疗领域得到了广泛的应用. 基于机器学习的抑郁症精神影像生物学标志物研究提出了大量可能作为诊断及治疗反应预测的异常脑区:其中具有诊断价值的脑区主要集中在楔前叶、扣带回、顶下缘角回、脑岛、丘脑以及海马等;具有预防价值的脑区主要集中在楔前叶、中央后回、背外侧前额叶、眶额叶、颞中回等;具有预测治疗反应价值的脑区主要集中在楔前叶、扣带回、顶下缘角回、额中回、枕中回、枕下回、舌回等. 值得关注的是,扣带回作为研究者们普遍关注的脑区,在关于诊断和治疗的研究中均显示出了高度的预测准确性(扣带回参与计算的脑区群在诊断和治疗反应方面的准确率分别在65.00%~91.70%和64.00%~89.47%之间),楔前叶也是目前的热点脑区,并在诊断、预防、治疗三个方面均取得了较高的准确性(楔前叶参与计算的脑区群在诊断、预防、治疗反应方面的准确率分别在65.00%~91.70%、68.90%~79.00%、64.00%~89.47%之间). 这表明扣带回和楔前叶可能是抑郁症客观生物学标志物中最重要的候选靶点脑区. 另外,我们的研究发现皮层区域的研究热度普遍高于中脑边缘系统或多巴胺系统相关的脑区,后者是以往研究主要关注的脑异常集中区域. 以杏仁核为例,仅有3篇诊断、1篇预防、1篇治疗方面的文章提及此脑区(其参与计算的脑区群在诊断、预防、治疗反应方面的准确率分别为56.68%~90.00%、79.00%、49.99%). 综上,我们不难看出,尽管脑区的选择会影响最终的分类准确率,但也存在应用同一脑区的不同研究得出的准确率存在异质性的问题,这可能与不同研究选用的方法有关.

    生物学标志物的开发已经为抑郁症的预防、诊断和治疗研究开辟了一条新的途径,许多潜在的生物学标志物已经被评估应用于MDD临床决策的有效性,主要包括:脑影像生物学标志物、遗传和蛋白质组学标志、体液中的激素和神经递质、以及近年兴起的非结构化文本数据(如过去史、家族史)[103]. 这当中,基于MRI的精神影像生物学标志物是唯一可以直接评估大脑功能、结构、相关代谢和疾病靶点的研究标志,其高分辨、非侵入性、时空连续性等优点也进一步确定了重要的临床价[3]. 随着影像学技术的发展,特别是功能磁共振成像和DTI的广泛应用,有望打破抑郁症目前依靠量表主观诊断的局面,为抑郁症的病因、发病机制、病程转归提供了强有力的研究手段. 据此,基于MRI的精神影像被认为是最有可能成为潜在的抑郁症客观生物学标志物.

    尽管精神影像学标志物展现了广阔的应用前景,但面对抑郁症复杂的病理生理改变和病因异质特性,只应用单一生物学标志物无法达到足够的灵敏度和特异[104]. 将影像学联合其他指标(生化、基因、家族史等)共同用于辅助决策,从而开发一套基于多种模态的抑郁症综合预测模型,有望对抑郁症病程转归和预后做出更加客观的评估. 组合反映不同病理生理改变的几种生物学标志物,能够尽可能避免疾病的异质性所带来的漏诊与误诊风险,从而提高整体诊断准确率. 非结构化的影像、生理以及临床数据组成的多维数据集为选择合适的数据处理方法带来了挑战. 相较于机器学习,深度学习具有较深层的学习架构,其数据自驱动的学习方式也使得其更加适合于抑郁症的辅助决策. 计算机视觉领域的发展使得深度学习在医学图像分析中有了更加广泛的应用,如图像分[105]、图像配[106]、图像融[107]以及图像注[108]等,这些技术的有机结合有助于实现深度学习对复杂影像数据的高精度提取与准确分类. 总之,深度学习可以使特殊工程的负担从医学专家转移到计算机,从而允许非医学专家介入,为抑郁症的客观生物学标志物研究注入了新的活力. 因此,基于MRI的精神影像和人工智能相结合是目前抑郁症客观生物学标志物研究中最有可能率先取得突破性进展的方法和途径. 而充分发挥深度学习的技术优势,构建“以影像学标志物为主,联合其他客观标志物”的多模态决策模型是将来抑郁症等精神疾病客观生物学标志物研究的主要发展趋势.

    虽然现有的基于精神影像和人工智能的抑郁症客观生物学标志物研究有了初步研究成果,但该新兴领域的研究和临床应用还面临一定的挑战. 其主要原因是目前绝大多数抑郁症客观标志物的研究仍采用了有监督的机器学习技术,深度学习研究仍然处于初步探索阶段. 另外,很少有研究证实人工智能算法在现实抑郁症患者中的临床应用价值,说明其部署于实际临床工作中可能会带来更多的挑战. 目前相关研究数量过少,研究质量参差不齐,客观标志物研究结果难以得到共识,这些都造成了目前的研究成果难以应用到临床患者中. 想要开发出更加具有实用价值的深度学习决策模型,需要更多优质的大样本、多中心、多模态抑郁症客观标志物研究,但模型的开发存在以下阻碍:模型的训练需要大量有/无标记的数据;要强大的计算性能和图形处理器(GPU);训练时间往往是传统机器学习方法的几倍甚至十几[109]. 另外,收集、存储和分享患者隐私数据涉及的伦理问题和道德规范也需要引起重视. 为早日突破局限,我们提出几点建议作为参考:针对于抑郁症研究的小样本问题,一方面,为了用有限的数据集创建样本数量更大的集合,将未标记的数据合并可能是有帮助[110],同时,利用在计算机视觉领域常用的增强技术对数据进行变换(如旋转、剪切、缩放)也同样可以增加样本[111]. 另一方面,通过多中心协作,并在这些协作中使用相同的纳入标准和跨站点扫描协议来收集数据. 增加样本量的另一种方法是通过多点数据共享计划,最具代表性的便是国际神经影像数据共享计划和OpenfMRI计[112]. 针对复杂模型训练时间较长的问题,我们在此处推荐CNNs与迁移学习方法. CNNs已经在神经影像识别中显示出了惊人的结[113],与具有同样层数的标准前馈神经网络相比,CNNs的连接和参数要少得多,因此更容易训练;同时,其稀疏连接和权值共享的特点使得体素水平的网络运行成为可能. 迁移学习即通过来自相似医学领域的数据集以有监督方式预训练CNNs网络,把网络的全部参数保存下来,应用于目标的小数据[114]. 目前已经成功应用于肠[115]、CT肺栓[116]、病理切[117]等医学图像中,在很多任务中有出色的表现,这一方法可以在精神影像中进一步探索. 另外,研究人员可以考虑采用简易数据进行实时智能分析的可能性. 这种分析技术与监督技术相结合,可以为临床医生评估和干预提供更加敏感的个性化信[118]. 例如,Netflix(个性化推荐系统[119]可以为不同用户推荐合适的算法完成个体用户服务,从而应用于个性化的心理健康评估和干预. 最后,在当研究条件受限于样本量、性能与时间的情况下,不断优化的机器学习方法仍然不失为一种选择. 许多机器学习技术仍需要访问训练数据集,这需要研究人员和临床医生之间加强合[27]. MDD未来生物学标志物研究的主要目标即进一步确定和改进适合的生物学标志物,与临床评估工具(如DSM-V)一样可用于临床环境以诊断MDD及其亚型,或进行治疗反应的预测.

    近年来,随着精神影像学技术相关的定性和定量方法的发展,为抑郁症脑科学研究注入了源源不断的活力,并产生了逐渐庞大的影像数据集,标志着抑郁症的脑影像研究也初步步入了大数据时代. 与此同时,以精神分裂症等为代表的其他精神疾病同样具有病因及发病机制不清,心理、社会、遗传多方面因素参与的特点,抑郁症的客观标志物研究方法对其同样适用. 总之,这一领域今后的研究热点将会集中在对影像学数据进行有效的挖掘与解读. 与传统的机器学习方法相比,深度学习能够对非常复杂的数据模型进行建模,在一定程度上提高计算机辅助诊断系统的准确率、灵敏度与特异度. 随着影像数据的多点共享计划取得突破,深度学习在影像分析中的巨大潜能有待更深入的挖掘. 同时在实践中,尝试将更加多元化的医疗数据应用于数据挖掘,推进医院与人工智能公司的多学科交叉融合,是指引我们走出科研摸索期的重要方向.

    Tel: 86-28-85423817, E-mail: tlchen@scu.edu.cn

  • 参 考 文 献

    • 1

      American Psychiatric Association. Dsm-5 Task Force. Diagnostic and statistical manual of mental disorders : DSM-5. 5th ed. Washington, D.C. 2013: xliv, 947 p

    • 2

      Gottesman, Ii, Gould T D. The endophenotype concept in psychiatry: etymology and strategic intentions. Am J Psychiatry, 2003, 160(4): 636-645

    • 3

      Li L J. Functional magnetic resonance imaging diagnostic marker for depression and its clinical application prospects. Chin J Psychiatry, 2015, 48(2) : 65-67

    • 4

      Dilsizian S E, Siegel E L. Artificial intelligence in medicine and cardiac imaging: harnessing big data and advanced computing to provide personalized medical diagnosis and treatment. Curr Cardiol Rep, 2014, 16(1): 441

    • 5

      Jentsch M C, Van Buel E M, Bosker F J, et al. Biomarker approaches in major depressive disorder evaluated in the context of current hypotheses. Biomark Med, 2015, 9(3): 277-297

    • 6

      Lui S, Zhou X J, Sweeney J A, et al. Psychoradiology: the frontier of neuroimaging in psychiatry. Radiology, 2016, 281(2): 357-372

    • 7

      Lin W. Principles of magnetic resonance imaging: a signal processing perspective [Book Review]. IEEE Signal Process Mag, 2000, 19(5): 129-130

    • 8

      Le-Bihan D, Mangin J C, Clark C, et al. Diffusion tensor imaging: concepts and applications. J Magn Reson Imaging, 2010, 13(4): 534-546

    • 9

      Deyoe E A, Bandettini P, Neitz J, et al. Functional magnetic resonance imaging (FMRI) of the human brain. J Neurosci Methods, 1994, 54(2): 171-187

    • 10

      Rive M M, Van R G, Veltman D J, et al. Neural correlates of dysfunctional emotion regulation in major depressive disorder. a systematic review of neuroimaging studies. Neurosci Biobehav Rev, 2013, 37(10): 2529-2553

    • 11

      Marchetti I, Koster E H, Sonuga-Barke E J, et al. The default mode network and recurrent depression: a neurobiological model of cognitive risk factors. Neuropsychol Rev, 2012, 22(3): 229-251

    • 12

      Chen M C, Burley H W, Gotlib I H. Reduced sleep quality in healthy girls at risk for depression. J Sleep Res, 2012, 21(1): 68-72

    • 13

      Joormann J, Cooney R E, Henry M L, et al. Neural correlates of automatic mood regulation in girls at high risk for depression. J Abnorm Psychol, 2012, 121(1): 61-72

    • 14

      Drevets W C. Orbitofrontal cortex function and structure in depression. Ann N Y Acad Sci, 2007, 1121(1): 499-527

    • 15

      Drevets W C, Price J L, Furey M L. Brain structural and functional abnormalities in mood disorders: implications for neurocircuitry models of depression. Brain Struct Funct, 2008, 213(1-2): 93-118

    • 16

      Fonseka T M, Macqueen G M, Kennedy S H. Neuroimaging biomarkers as predictors of treatment outcome in Major Depressive Disorder. J Affect Disord, 2017, 233(6): 2049

    • 17

      Fu C H, Costafreda S G. Neuroimaging-based biomarkers in psychiatry: clinical opportunities of a paradigm shift. Canadian Journal of Psychiatry Revue Canadienne De Psychiatrie, 2013, 58(9): 499

    • 18

      Koza J R, Bennett F H, Andre D, et al. Automated design of both the topology and sizing of analog electrical circuits using genetic programming. Dordrecht: Springer Netherlands, 1996: 151-170

    • 19

      Wilson S M, Ogar J M, Laluz V, et al. Automated MRI-based classification of primary progressive aphasia variants. Neuroimage, 2009, 47(4): 1558-1567

    • 20

      Fung G, Deng Y, Zhao Q, et al. Distinguishing bipolar and major depressive disorders by brain structural morphometry: a pilot study. BMC Psychiatry, 2015, 15(1): 298-298

    • 21

      Zhang J, Zong C. Deep neural networks in machine translation: an overview. IEEE Intelligent Systems, 2015, 30(5): 16-25

    • 22

      Hinton G E. Deep belief networks. Scholarpedia, 2009, 4(6): 5947

    • 23

      Lecun Y, Bengio Y, Hinton G. Deep learning. Nature, 2015, 521(7553): 436-444

    • 24

      Shen D, Wu G, Suk H I. Deep learning in medical image analysis. Annu Rev Biomed Eng, 2017, 19(1): 221-248

    • 25

      Georgevici A I, Terblanche M. Neural networks and deep learning: a brief introduction. Intensive Care Med, 2019, 45(5): 712-714

    • 26

      Symms M, Jager H R, Schmierer K, et al. A review of structural magnetic resonance neuroimaging. J Neurol Neurosurg Psychiatry, 2004, 75(9): 1235-1244

    • 27

      Bora E, Fornito A, Pantelis C, et al. Gray matter abnormalities in Major Depressive Disorder: a meta-analysis of voxel based morphometry studies. J Affect Disord, 2012, 138(1-2): 9-18

    • 28

      Chen T, Kendrick K M, Wang J, et al. Anomalous single-subject based morphological cortical networks in drug-naive, first-episode major depressive disorder. Hum Brain Mapp, 2017, 38(5): 2482-2494

    • 29

      Mwangi B, Ebmeier K P, Matthews K, et al. Multi-centre diagnostic classification of individual structural neuroimaging scans from patients with major depressive disorder. Brain, 2012, 135(5): 1508-152

    • 30

      Sacchet M D, Livermore E E, Iglesias J E, et al. Subcortical volumes differentiate major depressive disorder, bipolar disorder, and remitted major depressive disorder. J Psychiatr Res, 2015, 68: 91-98

    • 31

      Dosenbach N U F, Schlaggar B L. Prediction of individual brain maturity using fMRI. Science, 2010, 329(5997): 1358-1361

    • 32

      Guo H, Cheng C, Cao X, et al. Resting-state functional connectivity abnormalities in first-onset unmedicated depression. Neural Regen Res, 2014, 9(2): 153-163

    • 33

      Zeng L L, Shen H, Liu L, et al. Identifying major depression using whole-brain functional connectivity: a multivariate pattern analysis. Brain A Journal of Neurology, 2012, 135(Pt 5): 1498

    • 34

      Maxim V, Sendur L, Fadili J, et al. Fractional Gaussian noise, functional MRI and Alzheimer's disease. Neuroimage, 2005, 25(1): 141-158

    • 35

      Wei M, Qin J, Yan R, et al. Identifying major depressive disorder using Hurst exponent of resting-state brain networks. Psychiatry Res, 2013, 214(3): 306-312

    • 36

      Versace A, Almeida J R, Hassel S, et al. Elevated left and reduced right orbitomedial prefrontal fractional anisotropy in adults with bipolar disorder revealed by tract-based spatial statistics. Arch Gen Psychiatry, 2008, 65(9): 1041

    • 37

      Qin J, Wei M, Liu H, et al. Abnormal hubs of white matter networks in the frontal-parieto circuit contribute to depression discrimination via pattern classification. Magn Reson Imaging, 2014, 32(10): 1314-1320

    • 38

      Fang P, Zeng L L, Shen H, et al. Increased cortical-limbic anatomical network connectivity in major depression revealed by diffusion tensor imaging. Plos One, 2012, 7(9): e45972

    • 39

      Mclaughlin N C, Paul R H, Grieve S M, et al. Diffusion tensor imaging of the corpus callosum: a cross-sectional study across the lifespan. International Journal of Developmental Neuroscience the Official Journal of the International Society for Developmental Neuroscience, 2007, 25(4): 215-221

    • 40

      Versace A, Almeida J R, Quevedo K, et al. Right orbitofrontal corticolimbic and left corticocortical white matter connectivity differentiate bipolar and unipolar depression. Biol Psychiatry, 2010, 68(6): 560-567

    • 41

      Jacobs R H, Orr J L, Gowins J R, et al. Biomarkers of intergenerational risk for depression: a review of mechanisms in longitudinal high-risk (LHR) studies. J Affect Disord, 2015, 175: 494-506

    • 42

      Folandross L C, Sacchet M D, Prasad G, et al. Cortical thickness predicts the first onset of major depression in adolescence. Int J Dev Neurosci, 2015, 46: 125-131

    • 43

      Belden A C, Barch D M, Oakberg T J, et al. Anterior insula volume and guilt: neurobehavioral markers of recurrence after early childhood major depressive disorder. Jama Psychiatry, 2014, 72(1): 40

    • 44

      Koolschijn P C, Van Haren N E, Lensvelt-Mulders G J, et al. Brain volume abnormalities in major depressive disorder: a meta-analysis of magnetic resonance imaging studies. Neuroimage, 2009, 47(11): S152-S152

    • 45

      Hajek T, Cooke C, Kopecek M, et al. Using structural MRI to identify individuals at genetic risk for bipolar disorders: a 2-cohort, machine learning study. Journal of Psychiatry & Neuroscience Jpn, 2015, 40(2): 140142

    • 46

      Schnack H G, Nieuwenhuis M, Haren N E M V, et al. Can structural MRI aid in clinical classification? A machine learning study in two independent samples of patients with schizophrenia, bipolar disorder and healthy subjects. Neuroimage, 2014, 84(1): 299-306

    • 47

      Rocharego V, Jogia J, Marquand A F, et al. Examination of the predictive value of structural magnetic resonance scans in bipolar disorder: a pattern classification approach. Psychol Med, 2014, 44(3): 519-532

    • 48

      Cuijpers P, Beekman A, Smit F, et al. Predicting the onset of major depressive disorder and dysthymia in older adults with subthreshold depression: a community based study. Int J Geriatr Psychiatr, 2010, 21(9): 811-818

    • 49

      Kim B, Kim M-K, Yoo E, et al. Comparison of panic disorder with and without comorbid major depression by using brain structural magnetic resonance imaging. Prog Neuropsychopharmacol Biol Psychiatry, 2013, 43: 188-196

    • 50

      Ma Z, Li R, Yu J, et al. Alterations in regional homogeneity of spontaneous brain activity in late-life subthreshold depression. Plos One, 2013, 8(1): e53148

    • 51

      Wu T, Long X, Zang Y, et al. Regional homogeneity changes in patients with Parkinson's disease. Hum Brain Mapp, 2010, 30(5): 1502-1510

    • 52

      Paakkia J J, Rahkob J, Longc X, et al. Alterations in regional homogeneity of resting-state brain activity in autism spectrum disorders. Brain Res, 2010, 1321(12): 169-179

    • 53

      Emmrich A, Beesdobaum K, Gloster A T, et al. Depression does not affect the treatment outcome of CBT for panic and agoraphobia: results from a multicenter randomized trial. Psychother Psychosom, 2012, 81(3): 161-172

    • 54

      Lueken U, Straube B, Yang Y, et al. Separating depressive comorbidity from panic disorder: A combined functional magnetic resonance imaging and machine learning approach. J Affect Disord, 2015, 184: 182-192

    • 55

      Jia Z, Huang X, Wu Q, et al. High-field magnetic resonance imaging of suicidality in patients with major depressive disorder. Am J Psychiatry, 2010, 167(11): 1381-1390

    • 56

      Macqueen G M. Magnetic resonance imaging and prediction of outcome in patients with major depressive disorder. Journal of Psychiatry & Neuroscience Jpn, 2009, 34(5): 343

    • 57

      Costafreda S G, Chu C, Ashburner J, et al. Prognostic and diagnostic potential of the structural neuroanatomy of depression. Plos One, 2009, 4(7): e6353

    • 58

      Nouretdinov I, Costafreda S G, Gammerman A, et al. Machine learning classification with confidence: application of transductive conformal predictors to MRI-based diagnostic and prognostic markers in depression. NeuroImage, 2011, 56(2): 809-813

    • 59

      Davatzikos C, Shen D, Gur R C, et al. Whole-brain morphometric study of schizophrenia revealing a spatially complex set of focal abnormalities. Arch Gen Psychiatry, 2005, 62(11): 1218

    • 60

      Gammerman A, Nouretdinov I, Burford B, et al. Clinical mass spectrometry proteomic diagnosis by conformal predictors. Stat Appl Genet Mol Biol, 2008, 7(2): Article13

    • 61

      Gong Q, Wu Q, Scarpazza C, et al. Prognostic prediction of therapeutic response in depression using high-field MR imaging. Neuroimage, 2011, 55(4): 1497-1503

    • 62

      Fu C H, Williams S C, Cleare A J, et al. Neural responses to sad faci al expressions in major depression following cognitive behavioral therapy. Biol Psychiatry, 2008, 64(6): 505-512

    • 63

      Marquand A F, Mourao-Miranda J, Brammer M J, et al. Neuroanatomy of verbal working memory as a diagnostic biomarker for depression. Neuroreport, 2008, 19(15): 1507-1511

    • 64

      Scholkopf B, Platt J C, Shawe-Taylor J, et al. Estimating the support of a high-dimensional distribution. Neural computation, 2001, 13(7): 1443-1471

    • 65

      Hoy K E, Thomson R H, Cherk M H, et al. Effect of magnetic seizure therapy on regional brain glucose metabolism in major depression. Psychol Med, 2013, 211(2): 169-175

    • 66

      Shimizu Y, Yoshimoto J, Toki S, et al. Toward probabilistic diagnosis and understanding of depression based on functional MRI data analysis with logistic group LASSO. Plos One, 2015, 10(5): e0123524

    • 67

      Rosa M J, Portugal L, Hahn T, et al. Sparse network-based models for patient classification using fMRI. Neuroimage, 2015, 105: 493-506

    • 68

      Patel M J, Andreescu C, Price J C, et al. Machine learning approaches for integrating clinical and imaging features in late‐life depression classification and response prediction. Int J Geriatr Psychiatry, 2015, 30(10): 1056

    • 69

      Yuan G X, Chang K W, Hsieh C J, et al. A comparison of optimization methods and software for large-scale L1-regularized linear classification. J Mach Learn Res, 2010, 11(2): 3183-3234

    • 70

      Pinaya W H, Gadelha A, Doyle O M, et al. Using deep belief network modelling to characterize differences in brain morphometry in schizophrenia. Sci Rep, 2016, 6(1): 38897

    • 71

      Ulloa A, Plis S M, Erhardt E B, et al. Synthetic structural magnetic resonance image generator improves deep learning prediction of schizophrenia. Proceedings of the International Workshop on Machine Learning for Signal Processing. 2015: 1-6

    • 72

      Hazlett H C, Gu H, Munsell B C, et al. Early brain development in infants at high risk for autism spectrum disorder. Nature, 2017, 542(7641): 348

    • 73

      Heinsfeld A S, Franco A R, Craddock R C, et al. Identification of autism spectrum disorder using deep learning and the ABIDE dataset. Neuroimage Clinical, 2018, 17(C): 16-23

    • 74

      Suk H I, Wee C Y, Lee S W, et al. State-space model with deep learning for functional dynamics estimation in resting-state fMRI. Neuroimage, 2016, 129: 292-307

    • 75

      Islam J, Zhang Y. A novel deep learning based multi-class classification method for alzheimer’s disease detection using brain MRI data. Cham. Springer International Publishing, 2017: 213-222

    • 76

      Suk H I, Lee S W, Shen D. Deep sparse multi-task learning for feature selection in Alzheimer's disease diagnosis. Brain Struct Funct, 2016, 221(5): 2569-2587

    • 77

      Han X, Zhong Y, He L, et al. The unsupervised hierarchical convolutional sparse auto-encoder for neuroimaging data classification. cham. Springer International Publishing, 2015: 156-166

    • 78

      Choi H, Jin K H. Predicting cognitive decline with deep learning of brain metabolism and amyloid imaging. Behav Brain Res, 2018, 344: 103-109

    • 79

      Deshpande G, Wang P, Rangaprakash D, et al. Fully connected cascade artificial neural network architecture for attention deficit hyperactivity disorder classification from functional magnetic resonance imaging data. IEEE Transactions on Cybernetics, 2015, 45(12): 2668-2679

    • 80

      Munsell B C, Wee C-Y, Keller S S, et al. Evaluation of machine learning algorithms for treatment outcome prediction in patients with epilepsy based on structural connectome data. Neuroimage, 2015, 118: 219-230

    • 81

      Du Y, Fu Z, Calhoun V D. Classification and prediction of brain disorders using functional connectivity: promising but challenging. Front Neurosci, 2018, 12: 525

    • 82

      Deshpande G, Li Z, Santhanam P, et al. Recursive cluster elimination based support vector machine for disease state prediction using resting state functional and effective brain connectivity. PLoS One, 2010, 5(12): e14277

    • 83

      Kim J, Calhoun V D, Shim E, et al. Deep neural network with weight sparsity control and pre-training extracts hierarchical features and enhances classification performance: Evidence from whole-brain resting-state functional connectivity patterns of schizophrenia. Neuroimage, 2015, 124(Pt A): 127-146

    • 84

      Berman M G, Misic B, Buschkuehl M, et al. Does resting-state connectivity reflect depressive rumination? A tale of two analyses. Neuroimage, 2014, 103: 267-279

    • 85

      Choi H, Jin K H. Predicting cognitive decline with deep learning of brain metabolism and amyloid imaging. Behav Brain Res, 2018, 344: 103-109

    • 86

      Westman E, Muehlboeck J S, Simmons A. Combining MRI and CSF measures for classification of Alzheimer's disease and prediction of mild cognitive impairment conversion. Alzheimers Dement, 2011, 7(4): e16-e16

    • 87

      Kriegeskorte N. Deep Neural Networks: a new framework for modeling biological vision and brain information processing. Annu Rev Vis Sci, 2015, 1(1): 417

    • 88

      Acharya U R, Oh S L, Hagiwara Y, et al. Automated EEG-based screening of depression using deep convolutional neural network. Comput Methods Programs Biomed, 2018, 161: S0169260718301494

    • 89

      Lin E, Kuo P H, Liu Y L, et al. A deep learning approach for predicting antidepressant response in major depression using clinical and genetic biomarkers. Front Psychiatry, 2018, 9: 290

    • 90

      Jan A, Meng H, Gaus Y F A, et al. Artificial intelligent system for automatic depression level analysis through visual and vocal expressions. IEEE Transactions on Cognitive & Developmental Systems, 2017, PP(99: 1-1

    • 91

      Orabi A H, Buddhitha P, Orabi M H, et al. Deep learning for depression detection of twitter users. Proceedings of the The Workshop on Computational Linguistics & Clinical Psychology: From Keyboard To Clinic. New Orleans: Association for Computational Linguistics, 2018: 88-97

    • 92

      Kang Y, Xiao J, Ye Y, et al. Deep transformation learning for depression diagnosis from facial images. Chinese Conference on Biometric Recognition ,2017, 13-22

    • 93

      Geraci J, Wilansky P, Luca V D, et al. Applying deep neural networks to unstructured text notes in electronic medical records for phenotyping youth depression. Evid Based Ment Health, 2017, 20(3): 83-87

    • 94

      Tran T, Kavuluru R. Predicting mental conditions based on “History of present illness” in psychiatric notes with deep neural networks. J Biomed Inf, 2017, 75: S138-S148

    • 95

      Dwyer D B, Falkai P, Koutsouleris N. Machine learning approaches for clinical psychology and psychiatry. Annu Rev Clin Psychol, 2018, 14(1): 91-118

    • 96

      Mueller S G, Weiner M W, Thal L J, et al. Ways toward an early diagnosis in alzheimer's disease: the alzheimer's disease neuroimaging initiative (ADNI). Alzheimers Dement, 2005, 1(1): 55-66

    • 97

      Plis S M, Hjelm D R, Salakhutdinov R, et al. Deep learning for neuroimaging: a validation study. Front Neurosci, 2014, 8(8): 229

    • 98

      Wintermark M, Colen R, Whitlow C T, et al. The vast potential and bright future of neuroimaging. Brit J of adiol, 2018, 91(1087): 20170505

    • 99

      Ghafoorian M, Karssemeijer N, Heskes T, et al. Location sensitive deep convolutional neural networks for segmentation of white matter hyperintensities. Sci Rep, 2017, 7(1): 5110

    • 100

      Dolz J, Desrosiers C, Ben A I. 3D fully convolutional networks for subcortical segmentation in MRI: A large-scale study. Neuroimage, 2017, 170: 456-470

    • 101

      Herrmann L, Le Masurier M, Ebmeier K. White matter hyperintensities in late life depression: a systematic review. J Neurol Neurosurg Psychiatry, 2008, 79(6): 619-624

    • 102

      Van Uden I W, Tuladhar A M, De Laat K F, et al. White matter integrity and depressive symptoms in cerebral small vessel disease: The RUN DMC Study. Am J Geriatr Psychiatry, 2015, 23(5): 525-535

    • 103

      Papakostas G I, Shelton R C, Kinrys G, et al. Assessment of a multi-assay, serum-based biological diagnostic test for major depressive disorder: a pilot and replication study. Mol Psychiatry, 2013, 18(3): 332-339

    • 104

      Jentsch M C, Buel E M, Van, Bosker F J, et al. Biomarker approaches in major depressive disorder evaluated in the context of current hypotheses. Biomark Med, 2015, 9(3): 277-297

    • 105

      Zhang W, Li R, Deng H, et al. Deep convolutional neural networks for multi-modality isointense infant brain image segmentation. Neuroimage, 2015, 108: 214-224

    • 106

      Wu G, Kim M, Wang Q, et al. Scalable high-performance image registration framework by unsupervised deep feature representations learning. IEEE Trans Biomed Eng, 2016, 63(7): 1505-1516

    • 107

      Suk H I, Lee S W, Shen D. Hierarchical feature representation and multimodal fusion with deep learning for AD/MCI diagnosis. Neuroimage, 2014, 101: 569-582

    • 108

      Shin H, Roberts K, Lu L, et al. Learning to read chest X-rays: recurrent neural cascade model for automated image annotation. proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 2497-2506

    • 109

      Greenspan H, Van Ginneken B, Summers R M. Guest editorial deep learning in medical imaging: overview and future promise of an exciting new technique. IEEE Trans Med Imaging, 2016, 35(5): 1153-1159

    • 110

      Krogel M A, Scheffer T. Multi-relational learning, text mining, and semi-supervised learning for functional genomics. Mach Learn, 2004, 57(1-2): 61-81

    • 111

      Vieira S, Pinaya W H L, Mechelli A. Using deep learning to investigate the neuroimaging correlates of psychiatric and neurological disorders: methods and applications. Neurosci Biobehav Rev, 2017, 74(Pt A): 58-75

    • 112

      Poldrack R A, Barch D M, Mitchell J P, et al. Toward open sharing of task-based fMRI data: the OpenfMRI project. Front Neuroinform, 2013, 7: 12

    • 113

      Hosseini-Asl E, Ghazal M, Mahmoud A, et al. Alzheimer's disease diagnostics by a 3D deeply supervised adaptable convolutional network. Front Biosci, 2016, 23: 584-596

    • 114

      Pan S J, Yang Q. A survey on transfer learning. IEEE Trans Knowl Data Eng, 2010, 22(10): 1345-1359

    • 115

      Tajbakhsh N, Shin J Y, Gurudu S R, et al. Convolutional neural networks for medical image analysis: fine tuning or full training? IEEE Trans Med Imaging, 2017, 35(5): 1299-1312

    • 116

      Gao M, Bagci U, Lu L, et al. Holistic classification of CT attenuation patterns for interstitial lung diseases via deep convolutional neural networks. Comput Methods Biomech Biomed Eng Imaging Vis, 2018, 6(1): 1-6

    • 117

      Xu Y, Jia Z, Ai Y, et al. Deep convolutional activation features for large scale Brain Tumor histopathology image classification and segmentation. proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, Brisbane. IEEE, 2015: 947-951

    • 118

      Linder H P, Johnson S D. The netflix recommender system: algorithms, business value, and innovation. Acm Transactions on Management Information Systems, 2015, 6(4): 1-19

    • 119

      Johansson R, Sjoberg E, Sjogren M, et al. Tailored vs. standardized internet-based cognitive behavior therapy for depression and comorbid symptoms: a randomized controlled tial. Plos One, 2012, 7(5): e36905

孙也婷

机 构:

1). 四川大学华西医院放射科华西磁共振研究中心,成都 610041

2). 四川大学华西临床医学院,成都 610041

Affiliation:

1). Huaxi MR Research Center(HMRRC), Department of Radiology, West China Hospital of Sichuan University, Chengdu 610041, China

2). West China School of Medicine, Sichuan University, Chengdu 610041, China

陈桃林

机 构:

1). 四川大学华西医院放射科华西磁共振研究中心,成都 610041

3). 四川大学公共管理学院社会学与心理学系,成都 610065

Affiliation:

1). Huaxi MR Research Center(HMRRC), Department of Radiology, West China Hospital of Sichuan University, Chengdu 610041, China

3). Department of Sociality and Psychology, School of Public Administration, Sichuan University, Chengdu 610065, China

角 色:通讯作者

Role:Corresponding author

何度

机 构:四川大学华西医院病理科,成都 610041

Affiliation:Department of Pathology, West China Hospital of Sichuan University, Chengdu 610041, China

董再全

机 构:四川大学华西心理卫生中心,成都 610041

Affiliation:Center for Educational and Health Psychology Sichuan University, Chengdu 610041, China

程勃超

机 构:四川大学华西第二医院放射科,成都 610041

Affiliation:Department of Radiology, West China Second University Hospital,Sichuan University, Chengdu 610041, China

王淞

机 构:四川大学华西医院放射科华西磁共振研究中心,成都 610041

Affiliation:Huaxi MR Research Center(HMRRC), Department of Radiology, West China Hospital of Sichuan University, Chengdu 610041, China

汤万杰

机 构:四川大学华西心理卫生中心,成都 610041

Affiliation:Center for Educational and Health Psychology Sichuan University, Chengdu 610041, China

况伟宏

机 构:四川大学华西心理卫生中心,成都 610041

Affiliation:Center for Educational and Health Psychology Sichuan University, Chengdu 610041, China

龚启勇

机 构:四川大学华西医院放射科华西磁共振研究中心,成都 610041

Affiliation:Huaxi MR Research Center(HMRRC), Department of Radiology, West China Hospital of Sichuan University, Chengdu 610041, China

参考文献疾病种类样本容量模式分类器交叉验证法特征脑区准确率/%
Roberts et al., 2017

BD

HC=80,BD=49,BD高危组 = 71

rsfMRI

SVM

不详

腹外侧前额叶、内侧前额叶、额下回、脑岛、颞上回、豆状壳核

64.30

Sundermann et al., 2017

MDD

HC=180,MDD=180

rsfMRI

SVM

CV

不详

47.50~53.60
Hilbert et al., 2017GAD,MDDHC=24,GAD=19,MDD=14

sMRI

SVMs

LOOCV

额上回、额中回、额下回、前扣带回、尾状核、杏仁核

68.05

Ramasubbu et al., 2016

MDD

HC=19,MDD(轻微)=12, MDD(严重)=18,MDD(非常严重)=15

rsfMRI,fMRI(情绪识别任务),sMRI

SVM

5-fold CV

额极、额上回、额中回、额下回、眶额叶、内侧额叶、中央前回、前扣带回、脑岛

52.00~66.00

Rive et al., 2016

MDD,BD

BD=36,MDD=45

sMRI, rsfMRI(DMN, SN, FPNs)

GPC

LOOCV

内侧眶额叶、额中回、额前回、颞极、顶下缘角回、楔前叶、膝下前扣带回、海马旁回

69.10

Frangou et al., 2016BD,MDD家族史者HC=30,BD=30,MDD家族史者=30fMRI(工作记忆任务)

GPC

“leave 2 out”CV

额上回、额中回、颞叶

73.10

Rosa et al., 2015

MDD

HC=19,MDD=19

fMRI (性别区分与情感任务)

SVM

LOOCV

额上回、颞下回、颞上回、顶上回、脑岛、前扣带回、壳核、丘脑、胼胝体78.90~85.00
Sato et al., 2015

MDD

HC=21,MDD=25

fMRI(社会观念任务)

LDA

LOOCV

海马、丘脑和前脑岛

78.10

Shimizu et al., 2015

MDD

HC=31,MDD=31

fMRI(言语流畅性任务)

SVM

10-fold CV

舌回、小脑、楔前叶

90.00~95.00

Sacchet et al., 2015

MDD,BD

HC=61,BD=40,MDD=57, MDD(缓解)=35

sMRI

SVM

不详

杏仁核、伏隔核、尾状核、苍白球、壳核、丘脑、海马

59.50~62.70

Fung et al., 2015MDD,BDHC=29,MDD=19,BD=16

sMRI

SVM

10-fold CV

颞上沟、颞中回、楔前叶、顶上回、顶下回

74.30

Patel et al., 2015

LLD

老年HC=35,LLD=33sMRI,rsfMRI与DTI

SVM

LOOCV

扣带回、内侧前额叶、前脑岛、楔前叶和顶下缘角回

87.27

Johnston et al., 2015

TRD

HC=21,TRD=20

sMRI

SVM

LOOCV

尾状核、脑岛、松果体缰和脑室周围灰质

85.00

Mwangi et al., 2015

BD

HC=16,BD=16

DTI

SVM

LOOCV

颞上沟、额极、海马旁回、颞横回、颞极、前扣带回

78.12

Wu et al., 2015

UD

HC=26,UD=25

sMRI

SVM

LOOCV

丘脑和颞极

78.40

Koutsouleris et al., 2015

MDD,

SC

SC=158,

MDD=104

sMRI

SVM

CV

额下回、脑岛、颞上回、颞极、小脑、脑室周围区域、顶叶、辅助运动区域

76.00

Cao et al., 2014

MDD

HC=37,

MDD=39

rsfMRI

SVM

LOOCV

眶部额下回、缘上回、顶下缘角回、后扣带回、颞中回、颞下回

76.60

Guo et al., 2014

首发UD

HC=27,首发UD=36

rsfMRI

ANN

CV

海马、扣带回、豆状核和丘脑、顶下缘角回

90.50

Serpa et al., 2014

MDD,BD

HC-1组=33,HC-2组=38, MDD=19,BD=23

sMRI

SVM

LOOCV

不详

54.6~66.1

MacMaster et al., 2014MDD,BDHC=22,MDD=32,BD=14

sMRI

SVM

CV

不详

81.00

Chen et al., 2014

BD

HC=14,BD=9

sMRI

SVM

LOOCV

边缘叶、额叶、顶叶

57.00

Qin et al., 2014

MDD

HC=30,MDD=29

sMRI,DTI

SVM,GPC

LOOCV

背外侧额上回、颞下回、颞中回

83.05

Qiu et al., 2014

MDD

HC=32,MDD=32

sMRI

SVM

LOOCV

额中回尾部、前扣带回、颞极、颞中回、中央前回和枕外侧回

78.00

Dominik et al., 2014MDD,BDHC=22,MDD=22,BD=22fMRI(情绪识别任务)

GPC

LOOCV

杏仁核的浅表核、杏仁核基底外侧

79.60

Rondina et al., 2014

MDD

HC=30,MDD=30

fMRI(情绪识别任务)

RFE-SVM

CV

额中回、额下回、颞中回小脑、基底神经节、前扣带回、眶额叶、基底神经节和梭状回

67.00

Wei et al., 2013

MDD

HC = 20,MDD=20

rsfMRI

SVM

LOOCV

腹内侧前额叶、额顶叶、内侧额上回、额中回、角回、扣带回、楔前叶、小脑

90.00

Grotegerd et al., 2013

BD

HC=10,MDD=10,UD=10fMRI(情绪识别任务)

SVM

LOOCV

额下回、内侧额上回、眶部额上回、背外侧前额叶、杏仁核

90

Yu et al., 2013

MDD

HC=38,MDD=19

rsfMRI

SVM

LOOCV

内侧前额叶、楔前叶、扣带回、海马旁回、丘脑、颞下回、小脑

80.90

Almeida et al., 2013复发UD,BDBD=18,复发UD=18

ASL

SVM

LOOCV

膝下前扣带回

81.00

Miho et al., 2013MDD,SC

BD=43,SC=41

DTI,sMRI

SVM

不详

丘脑、膝下前扣带回、脑岛、胼胝体和侧脑室

88.00

Lord et al., 2012

MDD

HC=22,MDD=21

rsfMRI

SVM

CV

岛盖部额下回、三角部额下回、脑岛、枕上回

99.00

Fang et al., 2012

MDD

HC=26,MDD=22

DTI

SVM

LOOCV

眶额叶、基底神经节、丘脑、海马和脑岛、顶上回、后扣带回、楔前叶、枕叶和颞下回

91.70

Mwangi et al., 2012

MDD

HC=32,MDD=30

sMRI

SVM,RVM

LOOCV

背外侧前额叶、内侧额叶、眶额叶、颞叶、脑岛、小脑

90.30

Zeng et al., 2012

MDD

HC=29,MDD=24

rsfMRI

SVM

LOOCV

前扣带回、内侧前额叶、基底神经节、颞叶、小脑

94.30

Gong et al., 2011

MDD

HC=23,TRD=23,非TRD=23

sMRI

SVM

LOOCV

额上回、额中回、辅助运动区、中央后回、颞中回、颞下回、缘上回、角回、梭状回

58.7~84.6

Hahn et al., 2011

UD

HC=30,UD=30

fMRI(情绪识别任务,货币激励延迟任务)

SVM,GPC

LOOCV

梭状回、尾状核、额叶

83.00

Mouraomiranda et al., 2011

UD

HC=19,UD=19

fMRI(面部情感识别任务)

SVM

LOOCV

楔前叶、顶下缘角回、前额下回、额中回、额下回、前扣带回、颞中回、枕中回、脑岛

63~65.5

Fu et al., 2008

UD

HC=19,UD=19

fMRI(负性情绪识别任务)

SVM

LOOCV

额上回、额中回、颞中回、顶下缘角回、舌回、海马、杏仁核、丘脑

86.00

Marquand et al., 2008

MDD

HC=20,UD=20

fMRI(口头记忆任务)

SVM

LOOCV

额上回、额中回、额下回、中央前回、颞上回、颞中回、顶上回、梭状回、枕叶、海马旁回、尾状核、小脑

68.00

参考资料疾病种类样本容量模式分类器交叉验证法特征脑区准确率/%

Ramasubbu et al., 2016

MDD

HC=19,MDD(轻微)=12,MDD(严重)=18,MDD(非常严重)=15rsfMRI,fMRI(情绪识别任务),sMRI

SVM

5-fold CV

前额叶、脑岛和前扣带回

58.00

Sacchet et al., 2015

MDD,BD

HC= 61,BD=40, MDD=57,MDD(缓解)=35

sMRI

SVM

不详

尾状核、腹侧间脑

59.50~62.70

Foland-Ross et al., 2015

MDD

HC=15,MDD=18

sMRI

SVM

10-fold CV

内侧眶额叶、中央前回、前扣带回、脑岛

70.00

Hajek et al., 2015

BD

HC = 45,MDD = 45

sMRI

SVM,GPC

"leave 2 out" CV

前额叶区双侧白质束、扣带回、颞中回、楔前叶和枕叶

72.22~70.37

Lueken et al., 2015

PD,AG(伴UD)

PD/AG(不伴UD) = 33,PD/AG(伴UD)组=26

fMRI(恐惧条件反射任务)

随机欠采样树(random undersampling tree)

LOOCV

脑岛、背外侧前额叶

73.00

Macmaster et al., 2013

UD

HC=18,UD = 19

rsfMRI

Fisher逐步判别分析

LOOCV

眶额叶、前额叶、中央后回、颞下回、脑岛

91.90

参考资料疾病种类样本容量模式分类器交叉验证法特征脑区准确率/%
Schnyer et al., 2017

MDD

HC=25,MDD=25

DTI

SVM

LOOCV

海马、扣带回、胼胝体、上纵束、上额枕束

74.00

Guo et al., 2017

MDD

HC=28,MDD=38

rsfMRI

SVM

不详

眶部额上回、颞中回、丘脑、舌回、楔叶、后扣带回

97.54

Redlich et al., 2016

MDD

HC=21,ECT与抗抑郁药物联合治疗组=23,抗抑郁药物治疗组=23

sMRI,rsfMRI与DTI

SVM,GPC

LOOCV

膝下扣带回

78.30

Wade et al., 2016

MDD

HC=33,MDD=53

sMRI

SVM

LOOCV

尾状核、苍白球后内侧

89.00

Patel et al., 2015

LLD

老年HC=35,LLD=33

sMRI,rsfMRI与DTI

SVM

LOOCV

背内侧前额叶、扣带回、颞叶、颞极、角回

89.47

van Waarde et al., 2015

MDD

MDD治疗缓解=25, MDD治疗未缓解=20

fMRI

SVM

LOOCV

背外侧前额叶、眶额叶和后扣带回

84.50

Liu et al., 2012

MDD

HC=17,MDD治疗未缓解=18,MDD治疗缓解=17

sMRI, rsfMRI与DTI

SVM

LOOCV

内侧额叶、额中回、前扣带回、中央前回、缘上回、楔前叶、侧舌回、枕中回、枕下回、颞中回

82.90

Gong et al., 2011

MDD

HC=23,TRD=23,非TRD=23

sMRI

SVM

LOOCV

额上回、额中回、额下回、中央后回、颞中回、顶上回、缘上回、角回、梭状回、豆状核

58.70~84.60

Mouraomiranda et al., 2011

MDD

HC=19,MDD=19

fMRI(面部表情识别任务)

SVM

LOOCV

额中回、额下回、前扣带回、楔前叶、顶下缘角回、颞中回、枕中回、脑岛

63.00~65.50

Nouretdinov et al., 2011

UD

HC=19,UD=19

fMRI(悲伤情绪分级任务)

SVM

LOOCV

前扣带回、后扣带回、眶额叶

83.30

Costafreda et al., 2009

MDD

HC=37,MDD=37

sMRI, rsfMRI, DTI

SVM

LOOCV

眶额叶、额中回、前扣带回、后扣带回、枕叶、海马旁回

88.90

Marquand et al., 2008

UD

HC = 20, UD = 20

fMRI(口头工作记忆任务)

SVM

LOOCV

额上回、额中回、前扣带回、颞上回、颞中回、颞下回、楔前叶、 舌回

69.00

参考资料疾病种类样本容量数据类型深度学习方法准确率/%
Choi & Jin, 2018

AD

HC=182,AD=139,MCI=171

PET[角回、颞叶、后扣带回的FDG摄取量]

CNNs

84.00

Islam & Zhang, 2017

AD

AD=416

sMRI

CNNs

73.75

Suk et al., 2016

AD

HC=31,SC=31

rsfMRI

DAE

72.58~81.08

Suk et al., 2016

AD/MCI

HC=52,AD=51,MCI=99

sMRI[GM体积]与PET,3种CSF标志物:Aβ42、t-tau、p-tau

MTL

90.27

Han et al., 2015

ADHD,AD

不详

fMRI

HCSAE

65.00~80.00
Deshpande et al., 2017

ADHD

HC=744 ADHD(不注意型)= 173,ADHD(混合型)=260

rsfMRI

FCCANN

90.00~95.00
Kuang & He, 2015

ADHD

HC=95,SC=115

rsfMRI

DBN

48.90~72.70

Hazlett et al., 2017

ASD

HC=117,ASD高风险组=248,ASD = 70

sMRI[全脑体积:GM +WM、CSF、大脑、小脑和侧脑室]

DAN

94.00

Heinsfeld et al., 2018

ASD

HC=530,SC=505

rsfMRI

DNN

70.00

Acharya et al., 2018

MDD

HC=15,MDD=15

EEG

CNNs

95.49

Lin et al., 2018

MDD

MDD=455

遗传和临床行为数据

MFNNs

80.60

Orabi et al., 2018

MDD,PTSD

HC=572, MDD= 327, PTSD=246

非结构化文本数据

CNNs

87.90

Geraci et al., 2017

MDD

MDD=366(存在共病=735, 不存在共病=126)

EMRs分析

NLP,DNN

70.37

Tran & Kavuluru, 2017

焦虑、抑郁、惊恐障碍、ADHD等11种精神障碍患者

986例精神障碍患者(ADHD 41%, 焦虑68%, BD 33%, 痴呆27%, 进食障碍31%, 悲伤27%, OCD/OCSD 34%, 惊恐障碍47%, 精神错乱25%, PTSD 38%)

病史文字描述

DNN

63.14

Kim et al., 2016

SC

HC=50,SC=50

rsfMRI

DNN

85.80

Pinaya et al., 2016

SC

HC=83,SC=143,首发SC=32

sMRI[GM厚度和体积]

DBN

73.60

Ulloa et al., 2015

SC

HC=191,SC=198

sMRI[GM 体积]

MLP

75.00

表 1 基于精神影像和机器学习的抑郁症诊断生物学标志物研究汇总

Table 1 Studies of diagnostic biomarkers for depression based on psychoradiology and machine learning

表 2 基于精神影像和机器学习的抑郁症预防生物学标志物研究汇总

Table 2 Studies of preventive biomarkers for depression based on psychoradiology and machine learning

表 3 基于精神影像和机器学习的抑郁症治疗生物学标志物研究汇总

Table 3 Studies of prognosis biomarkers for depression based on psychoradiology and machine learning

表4 深度学习在抑郁症等精神疾病客观标志物研究中的应用

Table 4 The application of deep learning in studies of prognosis biomarkers for mental illness including depression

image /

fMRI:功能磁共振成像(functional magnetic resonance imaging);rsfMRI:静息态功能磁共振成像(resting-state functional magnetic resonance imaging);sMRI:结构磁共振成像(structural magnetic resonance imaging);DTI:弥散张量成像(diffusion tensor imaging);ASL:动脉自旋标记(arterial spin labelling);FA:各向异性分数(fractional anisotropy);TRD:难治性抑郁症(treatment-refractory depression);PD:惊恐障碍(panic disorder);AG:广场恐怖症(agoraphobia);HC:健康对照组(healthy control);BD:躁郁症(bipolar disorder);UD:单相抑郁(unipolar disorder);MDD:重度抑郁症(major depressive disorder);SC:精神分裂症(schizophrenia);GAD:广泛性焦虑障碍(generalized anxiety disorder);LLD:老年抑郁(late-life depression);ECT:电休克疗法(electroconvulsive therapy);SVM:支持向量机(support vector machine);RVM:相关向量机(relevance vector machine);ANN:人工神经网络(artificial neural network);GPC:高斯过程分类(gaussian process classifiers);LDA:最大熵线性判别分析(linear discriminant analysis);RFE:递归特征消除(recursive feature elimination);LOOCV:留一法交叉验证(leave-one-out cross-validation);CV:交叉验证法(cross validation);10-fold CV:10折交叉验证(10-fold cross-validation);5-fold CV:5折交叉验证(five-fold cross-validation);“leave 2 out”CV:留二法交叉验证(“leave 2 out”cross-validation);dDMN:背侧默认网络(dorsal default mode network);aSN:前突显网络(anterior salience network);FPNs:额顶叶网络(frontoparietal networks).

fMRI:功能磁共振成像(functional magnetic resonance imaging);rsfMRI:静息态功能磁共振成像(resting-state functional magnetic resonance imaging);sMRI:结构磁共振成像(structural magnetic resonance imaging);PD:惊恐障碍(panic disorder);AG:广场恐怖症(Agoraphobia);HC:健康对照组(healthy control);BD:躁郁症(bipolar disorder);UD:单相抑郁(unipolar disorder);MDD:重度抑郁症(major depressive disorder);SVM:支持向量机(support vector machine);GPC:高斯过程分类(gaussian process classifiers);LOOCV:留一法交叉验证(leave-one-out cross-validation);CV:交叉验证法(cross validation);10-fold CV:10折交叉验证(10-fold cross-validation);5-fold CV:5折交叉验证(five-fold cross-validation);“leave 2 out”CV:留二法交叉验证(“leave 2 out”cross-validation).

fMRI:功能磁共振成像(functional magnetic resonance imaging);rsfMRI:静息态功能磁共振成像(resting-state functional magnetic resonance imaging);sMRI:结构磁共振成像(structural magnetic resonance imaging);DTI:弥散张量成像(diffusion tensor imaging);TRD:难治性抑郁症(treatment-refractory depression);PD:惊恐障碍(panic disorder);AG:广场恐怖症(agoraphobia);HC:健康对照组(healthy control);BD:躁郁症(bipolar disorder);UD:单相抑郁(unipolar disorder);MDD:重度抑郁症(major depressive disorder);GAD:广泛性焦虑障碍(generalized anxiety disorder);LLD:老年抑郁(late-life depression);ECT:电休克疗法(electroconvulsive therapy);SVM:支持向量机(support vector machine);GPC:高斯过程分类(gaussian process classifiers);LOOCV:留一法交叉验证(leave-one-out cross-validation);CV:交叉验证法(cross validation);10-fold CV:10折交叉验证(10-fold cross-validation);5-fold CV:5折交叉验证(five-fold cross-validation);“leave 2 out”CV:留二法交叉验证(“leave 2 out”cross-validation).

PET:正电子发射型计算机断层显像(positron emission computed tomography);fMRI:功能磁共振成像(functional magnetic resonance imaging);rsfMRI:静息态功能磁共振成像(resting-state functional magnetic resonance imaging);sMRI:结构磁共振成像(structural magnetic resonance imaging);EEG:脑电图(electroencephalogram);SNPs:单核苷酸多态性(single nucleotide polymorphisms);EMRs:电子医疗记录(electronic medical records);AD:阿尔茨海默病(Alzheimer's disease);MCI:轻度认知障碍(mild cognitive impairment);ADHD:多动症(attention deficit hyperactivity disorder);ASD:自闭症(autistic-spectrum disorder);PTSD:创伤后应激障碍(post-traumatic stress disorder);SC:精神分裂症(schizophrenia);OCD:强迫障碍(obsessive compulsive disorders);OCSD:强迫症(obsessive compulsive spectrum disorder);BD:双相障碍(bipolar disorder);CNNs:深度卷积神经网络(convolutional neural networks);DAE:去噪自动编码(denoising autoencoders);DAN:深度自动编码网络(deep autoencoder networks);HCSAE:层叠卷积稀疏自动编码器(hierarchical convolutional sparse auto-encoder);FCC ANN:全连通级联人工神经网络(fully connected cascade artificial neural network);DBN:深度置信网络(deep belief network);DNN:深度神经网络(Deep Neural Network);DTL:深度转化学习(deep transformation learning);MFNNs:多层前馈神经网络(multilayer feedforward neural networks);MLP:多层感知器(multilayer perceptron);MTL:多任务学习(multi-task learning);NLP:自然语言处理(natural language processing);GM:灰质体积(gray matter);WM:白质体积(white matter);CSF:脑脊液体积(cerebrospinal fluid).

  • 参 考 文 献

    • 1

      American Psychiatric Association. Dsm-5 Task Force. Diagnostic and statistical manual of mental disorders : DSM-5. 5th ed. Washington, D.C. 2013: xliv, 947 p

    • 2

      Gottesman, Ii, Gould T D. The endophenotype concept in psychiatry: etymology and strategic intentions. Am J Psychiatry, 2003, 160(4): 636-645

    • 3

      Li L J. Functional magnetic resonance imaging diagnostic marker for depression and its clinical application prospects. Chin J Psychiatry, 2015, 48(2) : 65-67

    • 4

      Dilsizian S E, Siegel E L. Artificial intelligence in medicine and cardiac imaging: harnessing big data and advanced computing to provide personalized medical diagnosis and treatment. Curr Cardiol Rep, 2014, 16(1): 441

    • 5

      Jentsch M C, Van Buel E M, Bosker F J, et al. Biomarker approaches in major depressive disorder evaluated in the context of current hypotheses. Biomark Med, 2015, 9(3): 277-297

    • 6

      Lui S, Zhou X J, Sweeney J A, et al. Psychoradiology: the frontier of neuroimaging in psychiatry. Radiology, 2016, 281(2): 357-372

    • 7

      Lin W. Principles of magnetic resonance imaging: a signal processing perspective [Book Review]. IEEE Signal Process Mag, 2000, 19(5): 129-130

    • 8

      Le-Bihan D, Mangin J C, Clark C, et al. Diffusion tensor imaging: concepts and applications. J Magn Reson Imaging, 2010, 13(4): 534-546

    • 9

      Deyoe E A, Bandettini P, Neitz J, et al. Functional magnetic resonance imaging (FMRI) of the human brain. J Neurosci Methods, 1994, 54(2): 171-187

    • 10

      Rive M M, Van R G, Veltman D J, et al. Neural correlates of dysfunctional emotion regulation in major depressive disorder. a systematic review of neuroimaging studies. Neurosci Biobehav Rev, 2013, 37(10): 2529-2553

    • 11

      Marchetti I, Koster E H, Sonuga-Barke E J, et al. The default mode network and recurrent depression: a neurobiological model of cognitive risk factors. Neuropsychol Rev, 2012, 22(3): 229-251

    • 12

      Chen M C, Burley H W, Gotlib I H. Reduced sleep quality in healthy girls at risk for depression. J Sleep Res, 2012, 21(1): 68-72

    • 13

      Joormann J, Cooney R E, Henry M L, et al. Neural correlates of automatic mood regulation in girls at high risk for depression. J Abnorm Psychol, 2012, 121(1): 61-72

    • 14

      Drevets W C. Orbitofrontal cortex function and structure in depression. Ann N Y Acad Sci, 2007, 1121(1): 499-527

    • 15

      Drevets W C, Price J L, Furey M L. Brain structural and functional abnormalities in mood disorders: implications for neurocircuitry models of depression. Brain Struct Funct, 2008, 213(1-2): 93-118

    • 16

      Fonseka T M, Macqueen G M, Kennedy S H. Neuroimaging biomarkers as predictors of treatment outcome in Major Depressive Disorder. J Affect Disord, 2017, 233(6): 2049

    • 17

      Fu C H, Costafreda S G. Neuroimaging-based biomarkers in psychiatry: clinical opportunities of a paradigm shift. Canadian Journal of Psychiatry Revue Canadienne De Psychiatrie, 2013, 58(9): 499

    • 18

      Koza J R, Bennett F H, Andre D, et al. Automated design of both the topology and sizing of analog electrical circuits using genetic programming. Dordrecht: Springer Netherlands, 1996: 151-170

    • 19

      Wilson S M, Ogar J M, Laluz V, et al. Automated MRI-based classification of primary progressive aphasia variants. Neuroimage, 2009, 47(4): 1558-1567

    • 20

      Fung G, Deng Y, Zhao Q, et al. Distinguishing bipolar and major depressive disorders by brain structural morphometry: a pilot study. BMC Psychiatry, 2015, 15(1): 298-298

    • 21

      Zhang J, Zong C. Deep neural networks in machine translation: an overview. IEEE Intelligent Systems, 2015, 30(5): 16-25

    • 22

      Hinton G E. Deep belief networks. Scholarpedia, 2009, 4(6): 5947

    • 23

      Lecun Y, Bengio Y, Hinton G. Deep learning. Nature, 2015, 521(7553): 436-444

    • 24

      Shen D, Wu G, Suk H I. Deep learning in medical image analysis. Annu Rev Biomed Eng, 2017, 19(1): 221-248

    • 25

      Georgevici A I, Terblanche M. Neural networks and deep learning: a brief introduction. Intensive Care Med, 2019, 45(5): 712-714

    • 26

      Symms M, Jager H R, Schmierer K, et al. A review of structural magnetic resonance neuroimaging. J Neurol Neurosurg Psychiatry, 2004, 75(9): 1235-1244

    • 27

      Bora E, Fornito A, Pantelis C, et al. Gray matter abnormalities in Major Depressive Disorder: a meta-analysis of voxel based morphometry studies. J Affect Disord, 2012, 138(1-2): 9-18

    • 28

      Chen T, Kendrick K M, Wang J, et al. Anomalous single-subject based morphological cortical networks in drug-naive, first-episode major depressive disorder. Hum Brain Mapp, 2017, 38(5): 2482-2494

    • 29

      Mwangi B, Ebmeier K P, Matthews K, et al. Multi-centre diagnostic classification of individual structural neuroimaging scans from patients with major depressive disorder. Brain, 2012, 135(5): 1508-152

    • 30

      Sacchet M D, Livermore E E, Iglesias J E, et al. Subcortical volumes differentiate major depressive disorder, bipolar disorder, and remitted major depressive disorder. J Psychiatr Res, 2015, 68: 91-98

    • 31

      Dosenbach N U F, Schlaggar B L. Prediction of individual brain maturity using fMRI. Science, 2010, 329(5997): 1358-1361

    • 32

      Guo H, Cheng C, Cao X, et al. Resting-state functional connectivity abnormalities in first-onset unmedicated depression. Neural Regen Res, 2014, 9(2): 153-163

    • 33

      Zeng L L, Shen H, Liu L, et al. Identifying major depression using whole-brain functional connectivity: a multivariate pattern analysis. Brain A Journal of Neurology, 2012, 135(Pt 5): 1498

    • 34

      Maxim V, Sendur L, Fadili J, et al. Fractional Gaussian noise, functional MRI and Alzheimer's disease. Neuroimage, 2005, 25(1): 141-158

    • 35

      Wei M, Qin J, Yan R, et al. Identifying major depressive disorder using Hurst exponent of resting-state brain networks. Psychiatry Res, 2013, 214(3): 306-312

    • 36

      Versace A, Almeida J R, Hassel S, et al. Elevated left and reduced right orbitomedial prefrontal fractional anisotropy in adults with bipolar disorder revealed by tract-based spatial statistics. Arch Gen Psychiatry, 2008, 65(9): 1041

    • 37

      Qin J, Wei M, Liu H, et al. Abnormal hubs of white matter networks in the frontal-parieto circuit contribute to depression discrimination via pattern classification. Magn Reson Imaging, 2014, 32(10): 1314-1320

    • 38

      Fang P, Zeng L L, Shen H, et al. Increased cortical-limbic anatomical network connectivity in major depression revealed by diffusion tensor imaging. Plos One, 2012, 7(9): e45972

    • 39

      Mclaughlin N C, Paul R H, Grieve S M, et al. Diffusion tensor imaging of the corpus callosum: a cross-sectional study across the lifespan. International Journal of Developmental Neuroscience the Official Journal of the International Society for Developmental Neuroscience, 2007, 25(4): 215-221

    • 40

      Versace A, Almeida J R, Quevedo K, et al. Right orbitofrontal corticolimbic and left corticocortical white matter connectivity differentiate bipolar and unipolar depression. Biol Psychiatry, 2010, 68(6): 560-567

    • 41

      Jacobs R H, Orr J L, Gowins J R, et al. Biomarkers of intergenerational risk for depression: a review of mechanisms in longitudinal high-risk (LHR) studies. J Affect Disord, 2015, 175: 494-506

    • 42

      Folandross L C, Sacchet M D, Prasad G, et al. Cortical thickness predicts the first onset of major depression in adolescence. Int J Dev Neurosci, 2015, 46: 125-131

    • 43

      Belden A C, Barch D M, Oakberg T J, et al. Anterior insula volume and guilt: neurobehavioral markers of recurrence after early childhood major depressive disorder. Jama Psychiatry, 2014, 72(1): 40

    • 44

      Koolschijn P C, Van Haren N E, Lensvelt-Mulders G J, et al. Brain volume abnormalities in major depressive disorder: a meta-analysis of magnetic resonance imaging studies. Neuroimage, 2009, 47(11): S152-S152

    • 45

      Hajek T, Cooke C, Kopecek M, et al. Using structural MRI to identify individuals at genetic risk for bipolar disorders: a 2-cohort, machine learning study. Journal of Psychiatry & Neuroscience Jpn, 2015, 40(2): 140142

    • 46

      Schnack H G, Nieuwenhuis M, Haren N E M V, et al. Can structural MRI aid in clinical classification? A machine learning study in two independent samples of patients with schizophrenia, bipolar disorder and healthy subjects. Neuroimage, 2014, 84(1): 299-306

    • 47

      Rocharego V, Jogia J, Marquand A F, et al. Examination of the predictive value of structural magnetic resonance scans in bipolar disorder: a pattern classification approach. Psychol Med, 2014, 44(3): 519-532

    • 48

      Cuijpers P, Beekman A, Smit F, et al. Predicting the onset of major depressive disorder and dysthymia in older adults with subthreshold depression: a community based study. Int J Geriatr Psychiatr, 2010, 21(9): 811-818

    • 49

      Kim B, Kim M-K, Yoo E, et al. Comparison of panic disorder with and without comorbid major depression by using brain structural magnetic resonance imaging. Prog Neuropsychopharmacol Biol Psychiatry, 2013, 43: 188-196

    • 50

      Ma Z, Li R, Yu J, et al. Alterations in regional homogeneity of spontaneous brain activity in late-life subthreshold depression. Plos One, 2013, 8(1): e53148

    • 51

      Wu T, Long X, Zang Y, et al. Regional homogeneity changes in patients with Parkinson's disease. Hum Brain Mapp, 2010, 30(5): 1502-1510

    • 52

      Paakkia J J, Rahkob J, Longc X, et al. Alterations in regional homogeneity of resting-state brain activity in autism spectrum disorders. Brain Res, 2010, 1321(12): 169-179

    • 53

      Emmrich A, Beesdobaum K, Gloster A T, et al. Depression does not affect the treatment outcome of CBT for panic and agoraphobia: results from a multicenter randomized trial. Psychother Psychosom, 2012, 81(3): 161-172

    • 54

      Lueken U, Straube B, Yang Y, et al. Separating depressive comorbidity from panic disorder: A combined functional magnetic resonance imaging and machine learning approach. J Affect Disord, 2015, 184: 182-192

    • 55

      Jia Z, Huang X, Wu Q, et al. High-field magnetic resonance imaging of suicidality in patients with major depressive disorder. Am J Psychiatry, 2010, 167(11): 1381-1390

    • 56

      Macqueen G M. Magnetic resonance imaging and prediction of outcome in patients with major depressive disorder. Journal of Psychiatry & Neuroscience Jpn, 2009, 34(5): 343

    • 57

      Costafreda S G, Chu C, Ashburner J, et al. Prognostic and diagnostic potential of the structural neuroanatomy of depression. Plos One, 2009, 4(7): e6353

    • 58

      Nouretdinov I, Costafreda S G, Gammerman A, et al. Machine learning classification with confidence: application of transductive conformal predictors to MRI-based diagnostic and prognostic markers in depression. NeuroImage, 2011, 56(2): 809-813

    • 59

      Davatzikos C, Shen D, Gur R C, et al. Whole-brain morphometric study of schizophrenia revealing a spatially complex set of focal abnormalities. Arch Gen Psychiatry, 2005, 62(11): 1218

    • 60

      Gammerman A, Nouretdinov I, Burford B, et al. Clinical mass spectrometry proteomic diagnosis by conformal predictors. Stat Appl Genet Mol Biol, 2008, 7(2): Article13

    • 61

      Gong Q, Wu Q, Scarpazza C, et al. Prognostic prediction of therapeutic response in depression using high-field MR imaging. Neuroimage, 2011, 55(4): 1497-1503

    • 62

      Fu C H, Williams S C, Cleare A J, et al. Neural responses to sad faci al expressions in major depression following cognitive behavioral therapy. Biol Psychiatry, 2008, 64(6): 505-512

    • 63

      Marquand A F, Mourao-Miranda J, Brammer M J, et al. Neuroanatomy of verbal working memory as a diagnostic biomarker for depression. Neuroreport, 2008, 19(15): 1507-1511

    • 64

      Scholkopf B, Platt J C, Shawe-Taylor J, et al. Estimating the support of a high-dimensional distribution. Neural computation, 2001, 13(7): 1443-1471

    • 65

      Hoy K E, Thomson R H, Cherk M H, et al. Effect of magnetic seizure therapy on regional brain glucose metabolism in major depression. Psychol Med, 2013, 211(2): 169-175

    • 66

      Shimizu Y, Yoshimoto J, Toki S, et al. Toward probabilistic diagnosis and understanding of depression based on functional MRI data analysis with logistic group LASSO. Plos One, 2015, 10(5): e0123524

    • 67

      Rosa M J, Portugal L, Hahn T, et al. Sparse network-based models for patient classification using fMRI. Neuroimage, 2015, 105: 493-506

    • 68

      Patel M J, Andreescu C, Price J C, et al. Machine learning approaches for integrating clinical and imaging features in late‐life depression classification and response prediction. Int J Geriatr Psychiatry, 2015, 30(10): 1056

    • 69

      Yuan G X, Chang K W, Hsieh C J, et al. A comparison of optimization methods and software for large-scale L1-regularized linear classification. J Mach Learn Res, 2010, 11(2): 3183-3234

    • 70

      Pinaya W H, Gadelha A, Doyle O M, et al. Using deep belief network modelling to characterize differences in brain morphometry in schizophrenia. Sci Rep, 2016, 6(1): 38897

    • 71

      Ulloa A, Plis S M, Erhardt E B, et al. Synthetic structural magnetic resonance image generator improves deep learning prediction of schizophrenia. Proceedings of the International Workshop on Machine Learning for Signal Processing. 2015: 1-6

    • 72

      Hazlett H C, Gu H, Munsell B C, et al. Early brain development in infants at high risk for autism spectrum disorder. Nature, 2017, 542(7641): 348

    • 73

      Heinsfeld A S, Franco A R, Craddock R C, et al. Identification of autism spectrum disorder using deep learning and the ABIDE dataset. Neuroimage Clinical, 2018, 17(C): 16-23

    • 74

      Suk H I, Wee C Y, Lee S W, et al. State-space model with deep learning for functional dynamics estimation in resting-state fMRI. Neuroimage, 2016, 129: 292-307

    • 75

      Islam J, Zhang Y. A novel deep learning based multi-class classification method for alzheimer’s disease detection using brain MRI data. Cham. Springer International Publishing, 2017: 213-222

    • 76

      Suk H I, Lee S W, Shen D. Deep sparse multi-task learning for feature selection in Alzheimer's disease diagnosis. Brain Struct Funct, 2016, 221(5): 2569-2587

    • 77

      Han X, Zhong Y, He L, et al. The unsupervised hierarchical convolutional sparse auto-encoder for neuroimaging data classification. cham. Springer International Publishing, 2015: 156-166

    • 78

      Choi H, Jin K H. Predicting cognitive decline with deep learning of brain metabolism and amyloid imaging. Behav Brain Res, 2018, 344: 103-109

    • 79

      Deshpande G, Wang P, Rangaprakash D, et al. Fully connected cascade artificial neural network architecture for attention deficit hyperactivity disorder classification from functional magnetic resonance imaging data. IEEE Transactions on Cybernetics, 2015, 45(12): 2668-2679

    • 80

      Munsell B C, Wee C-Y, Keller S S, et al. Evaluation of machine learning algorithms for treatment outcome prediction in patients with epilepsy based on structural connectome data. Neuroimage, 2015, 118: 219-230

    • 81

      Du Y, Fu Z, Calhoun V D. Classification and prediction of brain disorders using functional connectivity: promising but challenging. Front Neurosci, 2018, 12: 525

    • 82

      Deshpande G, Li Z, Santhanam P, et al. Recursive cluster elimination based support vector machine for disease state prediction using resting state functional and effective brain connectivity. PLoS One, 2010, 5(12): e14277

    • 83

      Kim J, Calhoun V D, Shim E, et al. Deep neural network with weight sparsity control and pre-training extracts hierarchical features and enhances classification performance: Evidence from whole-brain resting-state functional connectivity patterns of schizophrenia. Neuroimage, 2015, 124(Pt A): 127-146

    • 84

      Berman M G, Misic B, Buschkuehl M, et al. Does resting-state connectivity reflect depressive rumination? A tale of two analyses. Neuroimage, 2014, 103: 267-279

    • 85

      Choi H, Jin K H. Predicting cognitive decline with deep learning of brain metabolism and amyloid imaging. Behav Brain Res, 2018, 344: 103-109

    • 86

      Westman E, Muehlboeck J S, Simmons A. Combining MRI and CSF measures for classification of Alzheimer's disease and prediction of mild cognitive impairment conversion. Alzheimers Dement, 2011, 7(4): e16-e16

    • 87

      Kriegeskorte N. Deep Neural Networks: a new framework for modeling biological vision and brain information processing. Annu Rev Vis Sci, 2015, 1(1): 417

    • 88

      Acharya U R, Oh S L, Hagiwara Y, et al. Automated EEG-based screening of depression using deep convolutional neural network. Comput Methods Programs Biomed, 2018, 161: S0169260718301494

    • 89

      Lin E, Kuo P H, Liu Y L, et al. A deep learning approach for predicting antidepressant response in major depression using clinical and genetic biomarkers. Front Psychiatry, 2018, 9: 290

    • 90

      Jan A, Meng H, Gaus Y F A, et al. Artificial intelligent system for automatic depression level analysis through visual and vocal expressions. IEEE Transactions on Cognitive & Developmental Systems, 2017, PP(99: 1-1

    • 91

      Orabi A H, Buddhitha P, Orabi M H, et al. Deep learning for depression detection of twitter users. Proceedings of the The Workshop on Computational Linguistics & Clinical Psychology: From Keyboard To Clinic. New Orleans: Association for Computational Linguistics, 2018: 88-97

    • 92

      Kang Y, Xiao J, Ye Y, et al. Deep transformation learning for depression diagnosis from facial images. Chinese Conference on Biometric Recognition ,2017, 13-22

    • 93

      Geraci J, Wilansky P, Luca V D, et al. Applying deep neural networks to unstructured text notes in electronic medical records for phenotyping youth depression. Evid Based Ment Health, 2017, 20(3): 83-87

    • 94

      Tran T, Kavuluru R. Predicting mental conditions based on “History of present illness” in psychiatric notes with deep neural networks. J Biomed Inf, 2017, 75: S138-S148

    • 95

      Dwyer D B, Falkai P, Koutsouleris N. Machine learning approaches for clinical psychology and psychiatry. Annu Rev Clin Psychol, 2018, 14(1): 91-118

    • 96

      Mueller S G, Weiner M W, Thal L J, et al. Ways toward an early diagnosis in alzheimer's disease: the alzheimer's disease neuroimaging initiative (ADNI). Alzheimers Dement, 2005, 1(1): 55-66

    • 97

      Plis S M, Hjelm D R, Salakhutdinov R, et al. Deep learning for neuroimaging: a validation study. Front Neurosci, 2014, 8(8): 229

    • 98

      Wintermark M, Colen R, Whitlow C T, et al. The vast potential and bright future of neuroimaging. Brit J of adiol, 2018, 91(1087): 20170505

    • 99

      Ghafoorian M, Karssemeijer N, Heskes T, et al. Location sensitive deep convolutional neural networks for segmentation of white matter hyperintensities. Sci Rep, 2017, 7(1): 5110

    • 100

      Dolz J, Desrosiers C, Ben A I. 3D fully convolutional networks for subcortical segmentation in MRI: A large-scale study. Neuroimage, 2017, 170: 456-470

    • 101

      Herrmann L, Le Masurier M, Ebmeier K. White matter hyperintensities in late life depression: a systematic review. J Neurol Neurosurg Psychiatry, 2008, 79(6): 619-624

    • 102

      Van Uden I W, Tuladhar A M, De Laat K F, et al. White matter integrity and depressive symptoms in cerebral small vessel disease: The RUN DMC Study. Am J Geriatr Psychiatry, 2015, 23(5): 525-535

    • 103

      Papakostas G I, Shelton R C, Kinrys G, et al. Assessment of a multi-assay, serum-based biological diagnostic test for major depressive disorder: a pilot and replication study. Mol Psychiatry, 2013, 18(3): 332-339

    • 104

      Jentsch M C, Buel E M, Van, Bosker F J, et al. Biomarker approaches in major depressive disorder evaluated in the context of current hypotheses. Biomark Med, 2015, 9(3): 277-297

    • 105

      Zhang W, Li R, Deng H, et al. Deep convolutional neural networks for multi-modality isointense infant brain image segmentation. Neuroimage, 2015, 108: 214-224

    • 106

      Wu G, Kim M, Wang Q, et al. Scalable high-performance image registration framework by unsupervised deep feature representations learning. IEEE Trans Biomed Eng, 2016, 63(7): 1505-1516

    • 107

      Suk H I, Lee S W, Shen D. Hierarchical feature representation and multimodal fusion with deep learning for AD/MCI diagnosis. Neuroimage, 2014, 101: 569-582

    • 108

      Shin H, Roberts K, Lu L, et al. Learning to read chest X-rays: recurrent neural cascade model for automated image annotation. proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 2497-2506

    • 109

      Greenspan H, Van Ginneken B, Summers R M. Guest editorial deep learning in medical imaging: overview and future promise of an exciting new technique. IEEE Trans Med Imaging, 2016, 35(5): 1153-1159

    • 110

      Krogel M A, Scheffer T. Multi-relational learning, text mining, and semi-supervised learning for functional genomics. Mach Learn, 2004, 57(1-2): 61-81

    • 111

      Vieira S, Pinaya W H L, Mechelli A. Using deep learning to investigate the neuroimaging correlates of psychiatric and neurological disorders: methods and applications. Neurosci Biobehav Rev, 2017, 74(Pt A): 58-75

    • 112

      Poldrack R A, Barch D M, Mitchell J P, et al. Toward open sharing of task-based fMRI data: the OpenfMRI project. Front Neuroinform, 2013, 7: 12

    • 113

      Hosseini-Asl E, Ghazal M, Mahmoud A, et al. Alzheimer's disease diagnostics by a 3D deeply supervised adaptable convolutional network. Front Biosci, 2016, 23: 584-596

    • 114

      Pan S J, Yang Q. A survey on transfer learning. IEEE Trans Knowl Data Eng, 2010, 22(10): 1345-1359

    • 115

      Tajbakhsh N, Shin J Y, Gurudu S R, et al. Convolutional neural networks for medical image analysis: fine tuning or full training? IEEE Trans Med Imaging, 2017, 35(5): 1299-1312

    • 116

      Gao M, Bagci U, Lu L, et al. Holistic classification of CT attenuation patterns for interstitial lung diseases via deep convolutional neural networks. Comput Methods Biomech Biomed Eng Imaging Vis, 2018, 6(1): 1-6

    • 117

      Xu Y, Jia Z, Ai Y, et al. Deep convolutional activation features for large scale Brain Tumor histopathology image classification and segmentation. proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, Brisbane. IEEE, 2015: 947-951

    • 118

      Linder H P, Johnson S D. The netflix recommender system: algorithms, business value, and innovation. Acm Transactions on Management Information Systems, 2015, 6(4): 1-19

    • 119

      Johansson R, Sjoberg E, Sjogren M, et al. Tailored vs. standardized internet-based cognitive behavior therapy for depression and comorbid symptoms: a randomized controlled tial. Plos One, 2012, 7(5): e36905