Fig.1

Workflow of the methodology in this paper"

Fig.2

Basic segmentation process for outline recognition models"

Fig.3

The structure of the U-Net model for timing IVUS segmentation"

Table 1

Number of labeled images for various plaques"

Fig.4

Various plaques in IVUS images (a) Fibrotic lesions; (b) Lipid lesions within fibrotic lesions; (c) Areas of ultrasound attenuation; (d) Calcified lesions"

Table 2

Average Dice coefficients of various pixel labeling based segmentation models in the test set"

Fig.5

Differences in segmentation results for different segmentation methods (a) Single-frame input U-Net prediction results; (b) Time-series IVUS segmentation U-Net model prediction results. Where dark blue is the vascular mesentery, light blue is the guidewire, and the green part in Fig. (a) is the incorrectly segmented fibres."

Fig.6

The segmentation results of combining time series information and using one single frame input (a) 4 consecutive frames of IVUS raw images; (b) Raw nnU-Net segmentation results; (c) Time-series IVUS segmentation U-Net model prediction results. Where green is vessel mesentery, dark blue is guidewire, grey is fibre and orange is lipid."

Table 3

Average Dice coefficients of IVUS segmentation models in the test set before and after adding the outer contour recognition model"

Fig.7

Segmentation results before and after adding the outer contour recognition model (a) Segmentation results of the mesentery of the time-series IVUS segmentation model without the addition of the outer contour recognition model; (b) Mesentery contours predicted by the outer contour recognition model; (c) Segmentation results without the addition of the outer contour recognition model; and (d) Segmentation results of the lesion after the addition of the outer contour recognition model. Where green represents calcified regions, gray is fibrous regions, and orange and tangerine represent lipid and ultrasound-attenuated regions, respectively. "

Fig.8

Difference in segmentation results of the timing IVUS segmentation model before and after adding the contour detection model in probabilistic display (a) Original IVUS image; (b) Segmentation results of the time-series IVUS segmentation model without the addition of the outer contour recognition model; (c) Segmentation results of the time-series IVUS segmentation model after the addition of the outer contour recognition model. In the probabilistic display effect, the lipids are rendered red on the basis of the original color mapping, and the color gradually deepens as the probability of lipids increases. "

Fig.9

Segmentation results of consecutive multi-frame IVUS images (a) Original IVUS image of 4 consecutive frames; (b) Segmentation results without using the outer contour recognition model; (c) Segmentation results after adding the outer contour recognition. Where green represents the calcified region, gray is the fibrous region, and orange and tangerine represent the lipid and ultrasound-attenuated region, respectively."

Fig.10

Segmentation results under different display methods (a) Results of classified segmentation of pixels based on maximum probability; (b) Visualization of confusing lesions rendered based on output probabilities. The lipids in the image are all rendered in red, and the color gradually deepens with increasing lipid probability in the probabilistic display effect."