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

Dong, Liang; Lu, Wei; Lu, Xuzhou; Leng, Xiaochang; Xiang, Jianping; Li, Changling

doi:10.1186/s12938-023-01171-2

Table 3 Means and standard deviations of the dice similarity coefficient (DSC), intersection over union (IoU) and Hausdorff distance (HD) of five segmentation models evaluated on seven different IVUS categories subsets of the testing cohort

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

	Models	Lumen			EEM
	Models	DSC	IoU	HD (mm)	DSC	IoU	HD (mm)
Calcified plaque (651 images)	R	0.961 ± 0.026	0.925 ± 0.045	0.202 ± 0.188	0.963 ± 0.030	0.930 ± 0.051	0.257 ± 0.228
	D	0.955 ± 0.032	0.915 ± 0.053	0.227 ± 0.230	0.962 ± 0.027	0.929 ± 0.047	0.259 ± 0.215
	S	0.947 ± 0.032	0.902 ± 0.054	0.303 ± 0.210	0.951 ± 0.034	0.908 ± 0.058	0.379 ± 0.233
	U	0.950 ± 0.032	0.906 ± 0.054	0.294 ± 0.294	0.950 ± 0.040	0.907 ± 0.067	0.414 ± 0.373
	C	0.960 ± 0.024	0.924 ± 0.043	0.222 ± 0.208	0.964 ± 0.028	0.932 ± 0.049	0.262 ± 0.246
Bifurcation (145 image ± s)	R	0.953 ± 0.044	0.913 ± 0.071	0.274 ± 0.325	0.969 ± 0.031	0.942 ± 0.053	0.267 ± 0.403
	D	0.947 ± 0.047	0.903 ± 0.075	0.318 ± 0.413	0.968 ± 0.030	0.939 ± 0.051	0.281 ± 0.407
	S	0.941 ± 0.041	0.891 ± 0.067	0.393 ± 0.328	0.955 ± 0.038	0.916 ± 0.063	0.401 ± 0.311
	U	0.933 ± 0.049	0.879 ± 0.080	0.473 ± 0.518	0.953 ± 0.045	0.913 ± 0.073	0.585 ± 0.724
	C	0.955 ± 0.041	0.917 ± 0.068	0.294 ± 0.387	0.969 ± 0.035	0.943 ± 0.058	0.271 ± 0.410
Adjacent vessels (197 images)	R	0.962 ± 0.027	0.928 ± 0.046	0.161 ± 0.112	0.971 ± 0.031	0.945 ± 0.054	0.160 ± 0.181
	D	0.956 ± 0.028	0.917 ± 0.048	0.187 ± 0.129	0.963 ± 0.049	0.932 ± 0.077	0.220 ± 0.255
	S	0.944 ± 0.043	0.897 ± 0.067	0.277 ± 0.165	0.940 ± 0.065	0.892 ± 0.101	0.398 ± 0.312
	U	0.947 ± 0.037	0.901 ± 0.061	0.297 ± 0.342	0.928 ± 0.086	0.876 ± 0.129	0.552 ± 0.605
	C	0.962 ± 0.024	0.928 ± 0.042	0.168 ± 0.109	0.973 ± 0.024	0.949 ± 0.042	0.194 ± 0.298
Stent (60 images)	R	0.957 ± 0.020	0.918 ± 0.036	0.241 ± 0.165	0.968 ± 0.020	0.939 ± 0.036	0.229 ± 0.147
	D	0.948 ± 0.022	0.902 ± 0.039	0.270 ± 0.150	0.968 ± 0.019	0.939 ± 0.034	0.234 ± 0.151
	S	0.936 ± 0.036	0.881 ± 0.061	0.381 ± 0.208	0.956 ± 0.024	0.917 ± 0.043	0.380 ± 0.215
	U	0.945 ± 0.027	0.897 ± 0.047	0.33 ± 0.188	0.951 ± 0.038	0.908 ± 0.066	0.407 ± 0.295
	C	0.956 ± 0.018	0.916 ± 0.032	0.251 ± 0.159	0.968 ± 0.017	0.939 ± 0.032	0.244 ± 0.141
Guidewire artifacts (945 images)	R	0.953 ± 0.047	0.913 ± 0.720	0.252 ± 0.243	0.976 ± 0.022	0.954 ± 0.039	0.183 ± 0.203
	D	0.949 ± 0.045	0.906 ± 0.071	0.268 ± 0.248	0.975 ± 0.020 ±	0.952 ± 0.036	0.186 ± 0.192
	S	0.943 ± 0.041	0.895 ± 0.066	0.334 ± 0.220	0.965 ± 0.032	0.933 ± 0.053	0.306 ± 0.220
	U	0.942 ± 0.044	0.894 ± 0.069	0.346 ± 0.301	0.967 ± 0.029	0.937 ± 0.049	0.313 ± 0.338
	C	0.954 ± 0.043	0.914 ± 0.067	0.273 ± 0.264	0.977 ± 0.021	0.955 ± 0.038	0.192 ± 0.231
None (422 images)	R	0.966 ± 0.019	0.934 ± 0.035	0.180 ± 0.125	0.976 ± 0.025	0.954 ± 0.043	0.138 ± 0.136
	D	0.961 ± 0.023	0.926 ± 0.041	0.202 ± 0.146	0.970 ± 0.041	0.945 ± 0.066	0.177 ± 0.202
	S	0.953 ± 0.033	0.912 ± 0.053	0.265 ± 0.162	0.957 ± 0.052	0.922 ± 0.083	0.298 ± 0.233
	U	0.956 ± 0.023	0.917 ± 0.041	0.260 ± 0.206	0.950 ± 0.069	0.912 ± 0.107	0.331 ± 0.362
	C	0.964 ± 0.021	0.932 ± 0.038	0.202 ± 0.154	0.977 ± 0.023	0.955 ± 0.040	0.160 ± 0.204
Lipid fibrous plaques (1109 images)	R	0.954 ± 0.046	0.914 ± 0.070	0.244 ± 0.243	0.980 ± 0.018	0.961 ± 0.032	0.144 ± 0.175
	D	0.947 ± 0.048	0.902 ± 0.074	0.268 ± 0.245	0.978 ± 0.017	0.958 ± 0.031	0.155 ± 0.162
	S	0.939 ± 0.054	0.889 ± 0.081	0.354 ± 0.277	0.968 ± 0.030	0.940 ± 0.050	0.278 ± 0.222
	U	0.940 ± 0.046	0.890 ± 0.073	0.338 ± 0.284	0.970 ± 0.034	0.943 ± 0.054	0.262 ± 0.272
	C	0.954 ± 0.043	0.915 ± 0.068	0.260 ± 0.250	0.980 ± 0.019	0.961 ± 0.033	0.148 ± 0.143

R Res-UNet, D DeepLab v3 plus, S Swin-UNet, U UNeXt, C CENet
The bold values indicate the optimal values for different models at the current categories and metrics

Back to article page

ISSN: 1475-925X

Contact us

Submission enquiries: journalsubmissions@springernature.com

BioMedical Engineering OnLine

Contact us