Main Coronary Vessel Segmentation Using Deep Learning in Smart Medical

Table 1

Comparison of segmentation performance between deep learning models for three main vessels (LAD, LCX, and RCA).

Method

ALL

LAD

LCX

RCA

Precision

Recall

F1 score

Precision

Recall

F1 score

Precision

Recall

F1 score

Precision

Recall

F1 score

UNet

0.875 ± 0.125

0.855 ± 0.134

0.865 ± 0.130

0.870 ± 0.115

0.855 ± 0.110

0.862 ± 0.116

0.821 ± 0.160

0.801 ± 0.186

0.811 ± 0.191

0.881 ± 0.132

0.878 ± 0.146

0.879 ± 0.168

ResNet

0.901 ± 0.101

0.891 ± 0.120

0.896 ± 0.117

0.905 ± 0.121

0.881 ± 0.115

0.893 ± 0.118

0.861 ± 0.137

0.852 ± 0.123

0.856 ± 0.128

0.925 ± 0.137

0.902 ± 0.117

0.913 ± 0.126

DenseNet

0.912 ± 0.108

0.923 ± 0.125

0.917 ± 0.122

0.918 ± 0.102

0.925 ± 0.115

0.921 ± 0.108

0.878 ± 0.152

0.890 ± 0.162

0.884 ± 0.147

0.929 ± 0.128

0.927 ± 0.105

0.928 ± 0.119

ResAttNet

0.919 ± 0.118

0.924 ± 0.102

0.921 ± 0.120

0.921 ± 0.110

0.918 ± 0.120

0.919 ± 0.091

0.882 ± 0.148

0.897 ± 0.122

0.889 ± 0.130

0.933 ± 0.125

0.925 ± 0.101

0.929 ± 0.105

The highest F1 score is shown in bold. RCA, right coronary artery; LAD, left anterior descending artery; LCX, left circumflex artery.