Fig. 1. Automatic identification and risk assessment process of vulnerable plaque. (a) Plaque identification; (b) risk assessment; (c) model testing

Fig. 2. Vulnerable plaque detection network. (a) Feature extraction module; (b) region extraction module; (c) secondary detection module; (d) A-Scan classification module

Fig. 3. Network structure of DeepLabV3+

Fig. 4. Calculation process of fiber cap thickness of vulnerable plaque. (a) Crop result after removing catheter noise by combining vulnerable plaque location information; (b) alignment result of vascular lumen (upper surface); (c) result of noise removal by 5×5 Gaussian filter; (d) calculating segmentation brightness threshold and delineating pixel points exceeding brightness threshold as fiber cap region with fiber cap boundary shown by white line

Fig. 5. Network structure for predicting superficial plaque microcalcification and macrophage infiltration

Fig. 6. Random cyclic shift. (a) Implementation process when area marked by solid line is cropped and spliced to left; (b) two cyclic shift instances, with initial image on left and result of cyclic shift on right

Fig. 7. Positioning and risk assessment results (green mark is actual border, blue mark is Baseline prediction result, and red mark is final prediction result). (a) Polar map positioning; (b) cartesian map positioning; (c) DeepLabV3+ segmentation result; (d) angle of lesion accumulation; (e) confusion matrix of macrophage infiltration; (f) confusion matrix of superficial microcalcification