Comparison of deep learning-based image segmentation methods for intravascular ultrasound on retrospective and large image cohort study

Table 2 Performance of the five segmentation models on the testing cohort for lumen and EEM segmentation with metrics including dice similarity coefficient (DSC), intersection over union (IoU) and Hausdorff distance (HD)

Models	Lumen			EEM			Parameters (M)
Models	DSC	IoU	HD (mm)	DSC	IoU	HD (mm)	Parameters (M)
Res-UNet	0.958 ± 0.037	0.921 ± 0.058	0.219 ± 0.210	0.974 ± 0.024	0.951 ± 0.043	0.178 ± 0.200	24.45
DeepLab v3 plus	0.952 ± 0.040	0.911 ± 0.064	0.243 ± 0.226	0.972 ± 0.027	0.947 ± 0.046	0.190 ± 0.192	54.11
Swin-UNet	0.944 ± 0.045	0.897 ± 0.070	0.321 ± 0.242	0.961 ± 0.037	0.927 ± 0.062	0.312 ± 0.231	41.39
UNeXt	0.946 ± 0.039	0.900 ± 0.064	0.310 ± 0.276	0.960 ± 0.046	0.926 ± 0.073	0.321 ± 0.330	1.47
CENet	0.958 ± 0.035	0.921 ± 0.057	0.237 ± 0.223	0.975 ± 0.024	0.951 ± 0.042	0.184 ± 0.197	29.00

The number of parameters representing the size of the model was also listed
The bold values indicate the optimal values for different models at the current metrics

ISSN: 1475-925X