Table 3.

Review of the papers applying deep learning for coronary vessel segmentation.

References	Modality	Data source	No. of subj.	ML model	Input	DSC	Add’l perf. metrics
Dong et al.71	CCTA	Private	338	Di-Vnet	3D patches	0.902	Prec=0.921, Recall=0.97
Gao et al.72	XCA	Private	130	GBDT	2D images	-	F1=0.874, Sen= 0.902, Spec=0.992
Wolterink et al.73	CCTA	Publicly available	18	GCN	2D images	0.74	MSD=0.25mm
Li et al.74	CCTA	Private	243	2D U-Net with 3DNet	2D images	0.771 ± 0.021	AUC=0.737
Song et al.75	CCTA	Private	68	3D FFR U-Net	3D patches	0.816	Prec=0.77, Recall=0.87
Zeng et al.76	CCTA	Publicly available	1000	3D-UNet	3D volumes	0.82	-

DSC: Dice similarity coefficient; Prec: Precision; Sen: Sensitivity; Spec: Specificity; GBDT: Gradient boosting decision tree; CCTA:Coronary computed tomographic angiography; XCA: X-ray coronary angiography; GCN: Graph convolutional networks; MSD: Mean surface distance; AUC: Area under the receiver operating characteristic curve; ROI: Region of interest; 3D FFR U-Net: 3D feature fusion and rectification U-Net