1　Materials and methods

A schematic diagram of the experimental setup is shown in Figure 1a. A semiconductor laser (λ= 690 nm, 100 mW, Changchun New Industries Optoelectronics Technology, China), is expanded by a beam expander (BE), and then illuminates the specimen. The diffuse reflective light is relayed by a microscope (Olympus SZ-CTV) onto an 8-bit CMOS camera (acA2000-340 km, Basler). The camera has a frame size of 2 048 × 1 088 pixels, with a 5.5 μm× 5.5 μm pixel size and a 42 fps(frames/second) frame rate. The microscope lens magnification can be adjusted within the range 0.75-2.5. The entire experimental setup is fixed on a vibration-free optical platform to reduce the effects of external vibration on the imaging.

The acquired signal can be considered as fragments of a Doppler signal, so that it is a superposition of signals with different frequencies. The dynamic integrated time influence the intensity and frequency distribution of the acquired signal. Therefore, a series of exposure time gradients were set in this method. As shown in Figure 1b, under each exposure time, raw speckle images of 512 frames were acquired and processed to obtain a blood flow image. In total, 512N frames images were continuously acquired to reconstruct N blood flow images. For the nth exposure time, the raw signal of one camera pixel can be expressed as Int :

where  I0  is the background signal, which is the so-called stationary signal, and ∑i=1nIicos 2πfit∏Tn2t-t0 is the dynamic signal generated by moving scattering particles. ∏Tn2t-t0  is a block-pulse function defined a  rect2t-t0/Tn*combvt , where t0 is the initial time point, n is the gradient number of exposure time, Tn  is the nth exposure time, v is the sampling rate of our camera, and "*" denotes convolution. The captured signal can be considered to be generated as a result of the camera integrating the number of received photons. As the exposure time increases, I0  and the mean value of  Int increase. In order to analyze the effect of dynamic integrated time on the stationary and dynamic signals, fast Fourier transform was performed to convert the raw speckle pattern recorded at each pixel from the time domain to the frequency domain:

Fig. 1 Schematic of the HDOA method

NOTE: (a) Schematic of the experimental setup. (b) The operating principle of the HDOA method.

Where  isu is a frequency-domain stationary signal and  idu±fi is a frequency-domain dynamic signal. In the frequency domain, isu  increases as the exposure time increases. This results from the increasing stationary signal contribution. Besides, dynamic signal also plays an important role. Longer exposure time blurs the dynamic signal, especially for the one corresponding to high-velocity flow, and transforms it to low-frequency signal. For each exposure time, the imaging parameter, MDn , is defined as the ratio of the average intensity of the frequency-domain dynamic signal to that of the stationary signal.

Where f0  and f are the filter range and camera sampling rate, respectively. MDn  of each-level vessels vary as the exposure time increases from T1 to  Tn . As shown in Figure 1b, MDn  of small vessels increases and that of big vessels decreases as the exposure time increases. In the common biomedical condition, the interference signal frequency is typically in the range of  20 Hz–10 kHz ^[22]. For a short exposure time, MDn  remains almost same with increasing flow velocity. Because the exposure time (e.g., 0.1 ms) is short enough to capture the instantaneous high-frequency Doppler signal. However, for a long exposure time, MDn  decreases rapidly as the flow velocity increases. For example, an exposure time of 5 ms is used in laser speckle contrast analysis imaging. Considering both the dynamic integrated effect and absorption effect, a short exposure time is more suitable for big vessels with high velocity and a long exposure time is more appropriate for small vessels with low red blood cells concentration and absorption coefficient. To integrate the superior high-dynamic information, a series of images with multi-exposure time were fused into a HDOA image. The imaging parameter, high-dynamic modulation depth (HDMD), is defined as:

First, the configuration of blood flows are extracted from the background. Second, the highest MDnx,y  of the same blood flows and the lowest pixel value MDnx,y  of the same background location in multi-exposure images are selected to reconstruct the HDOA image.

2　Experiment and results

The dependence between the dynamic integrated time and the imaging quality of each-level vessels was verified by a phantom experiment with designed flow velocity and scattering particles concentration. Two capillary tubes with the inner diameter of 0.5 mm were fixed in 2% agarose gel for index matching. As shown in Figure 2a, 0.8 g/L and 1.6 g/L TiO₂ solutions were injected into tube A and tube B, by a double-channel syringe pump (JZB-2800, JYM), with the velocity of 2.8 mm/s and 8.4 mm/s, respectively. Tube A was used to simulate a capillary with a low RBC concentration and flow velocity, and tube B corresponds to a big vessel with a high RBC concentration and flow velocity. Figure 2b shows the relationship between the MD and the exposure time for tube A and tube B. The exposure time increases from 120 μs to 700 μs with same gradient of 58 μs. The MD results from the average value of the area indicated by the dashed-line box in Figure 2a. For tube A, MD is directly proportional to the exposure time, and for tube B, the relationship is inverse. Two factors should be considered for this result. The first is the absorption effect and the second is the dynamic integrated effect. For tube A, the flow velocity is sufficiently low that a long exposure time almost does not blur the dynamic signal. Meanwhile, long exposure time increases the dynamic signal amplitude, which is too weak to be acquired for a low exposure time. Similarly, for tube B, MD considerably decreases owing to the high dynamic integrated effect shown for a high flow velocity. Correspondingly, for different level vessels, a big vessel has high RBCs concentration and flow velocity, and microvessel has low RBCs concentration and flow velocity. Therefore, we assume that the MD of vessels decreases as the exposure time increases, and the MD of microvessels increases as the exposure time increases.

Fig. 2 Phatom experiment images

NOTE: (a) MD image of the phantom. (b) Relationship between the average MD and the exposure time at different concentrations and flow speed.

To demonstrate the influence of absorption on imaging, we imaged the blood flow of a twenty-day-old Gold Pristella Tetra. The Gold Pristella Tetra was anesthetized with 5 mg/L stabilizer (MS-222). Then, it was fixed in a plastic dish filled with low melting-point agarose L.M.P (1% agar gel, temperature 24℃) to minimize the impact of its movement on the imaging. Figure 3a is a raw acquired image of the fish. Figure 3b-e shows the MD image obtained at the exposure time of 0.05, 0.25, 0.35 and 0.50 ms, respectively. Figure 3f shows the HDMD image using the HDOA imaging method. According to Lambert-Beer's law, samples of different thicknesses and absorption coefficients have different transmittances of light. We selected three regions of the Gold Pristella Tetra (P1, P2, and P3) to compare the imaging differences between the MD and the HDMD images. P1, P2, and P3 are related to fish mouse, swim bladder, and fish tail, respectively. P1 and P2 have similar thickness; however, the difference between their absorption coefficients is significant. The difference between the thicknesses of P2 and P3, is very high. For any single exposure time, these vessels in P1, P2, and P3 cannot be imaged simultaneously. However, HDOA can extract the optimal information at different absorption conditions and reconstruct all vessels in a HDMD image. To illustrate the effect of absorption on HDOA, we quantitively analyze the spatial resolution and signal noise ratio of P1, P2, and P3. Figure 3g-h shows the spatial resolution and signal-to-noise ratio (SNR) of the MD image at the exposure time of 0.05, 0.25, 0.35 and 0.50 ms and the HDMD image, respectively. The pixel values of the regions indicated by the dashed-line boxes in Figure 3b-e are used to calculate the spatial resolution and SNR. The calculation method of spatial resolution is described as follows. We select the data of a region to obtain the MD curve, where the difference between the half-height and the quarter-height of the obtained curve peak is defined as the spatial resolution. Note that in Figure 3h the HDMD image has a higher SNR and its SNR increases with the exposure time. This verifies the feasibility of the HDOA method for obtaining high-resolution blood flow imaging.

Fig. 3 High-dynamic angiography with HDOA imaging method

NOTE: (a)Raw acquired image.(b-e)MD image obtained at exposure times of 0.05, 0.25, 0.35 and 0.50 ms, respectively.(f)HDMD image. (g-h)Spatial resolution and SNR corresponding to the three regions shown in the dashed-line boxes in(b-e).

To demonstrate the influence of high dynamic effect on imaging, we present an enlarged view of the Gold Pristella Tetra and analyze different-level vessels. The magnification of the microscope is set to 1.8. Figure 4a shows the HDMD image of the Gold Pristella Tetra. In this figure, M represents the area of the microvessel, and V represents the area of the vessel. To facilitate the observation of the blood vessel distribution of the Gold Pristella Tetra at different exposure times, we present a view of the area indicated by the solid-line box(Figure 4a) at different exposure times. Figure 4b-g show the MD images obtained at exposure times of 0.024, 0.2, 0.4, 0.6, 0.8, and 1.0 ms. Arrow A and arrow B represent the area of the vessel and microvessel, respectively. As the exposure time increases, over-exposure occurs in vessel, and the clarity of microvessel improves gradually. The acquired signal of RBCs flowing with high velocities in large vessels fluctuates with high frequency. When a short exposure time is used in the experiment, the fluctuated fragments of the Doppler signal can be acquired. However, the acquired Doppler signal must be blurred as exposure time increases. In this case, the imaging parameter MD will decrease, as the integration of more signals eliminates the fluctuation of the acquired signal. For RBCs with low flow velocities, an increase in the exposure time does not lead to the blurring of the acquired signal. Instead, it enlarges the acquired signal and improve its sensitivity. Figure 4h shows the exposure time required for the imaging of vessel and microvessel in the dashed-line box in Figure 4a. From it we can found that the data of microvessels is mainly acquired under long exposure time, while the one of vessel mainly from shorter exposure time. The conclusion also coincides with the assumption made in the phantom experiment.

Fig. 4 Local high-dynamic angiography

NOTE: (a) HDMD image of Gold Pristella Tetra. (b-g) MD image of the region in the solid-line rectangle in (a) at exposure times of 0.024, 0.2, 0.4, 0.6, 0.8 and 1.0 ms, respectively. (h) The optimal exposure time required for imaging of vessel and microvessel in the dashed-line rectangle in (a). Arrow A and arrow B represent the area of the vessel and microvessel, respectively.

3　Discussion and conclusion

In conclusion, we have proposed a HDOA method, which can perform blood flow imaging for specimens with different thicknesses and absorption coefficients. The high performance of the proposed method was verified by a phantom and an in vivo biological experiments. However, the proposed HDOA method also has its limitations. For example, the biological tissue must be thin or neartransparent, and its imaging time is affected by the multiple exposure time acquisition mechanism.

References