Objective: This study aimed to develop a light-weighted classification network for hepatocellular carcinoma (HCC) and non-HCC malignancies based on the automatic analysis of brightness change in contrast-enhanced ultrasound (CEUS). Methods: This retrospective study comprised 131 patients diagnosed with HCC and 30 patients with non-HCC malignancies. We used the YOLOX network to detect the tumor region of interest on B-mode ultrasound and CEUS images. A custom-developed algorithm extracted brightness change curves in the tumor and adjacent liver parenchyma regions from CEUS images. We also developed one-dimensional convolutional neural networks (1D-ResNet, 1D-ConvNeXt and 1D-CNN), and machine-learning methods such as support vector machine, ensemble learning, K-nearest neighbor, and decision tree, to analyze brightness change curves and classify HCC and non-HCC malignancies. Results: Area under the receiver operating characteristic curve (AUC) of these machine-learning methods were 0.70, 0.56, 0.63, and 0.72 respectively. Meanwhile the 1D-ResNet, 1D-ConvNeXt and 1D-CNN demonstrated AUCs of 0.72, 0.82 and 0.84 for HCC and non-HCC classification based on brightness change curves. Conclusion: The 1D-CNN model can differentiate between patients with HCC and non-HCC malignancies at an accuracy that surpass those of machine learning and other deep learning methods. This paper provides a user-friendly and cost-efficient computer-aided diagnostic solution to aid radiologists in clinical decision-making of HCC.