Fully Automated Segmentation of Lower Extremity Deep Vein Thrombosis Using Convolutional Neural Network

Table 2

Comparisons of segmentation performance between the proposed CNN model and the other models.

Algorithm	DSC		Precision		Recall
	Mean ± SD	Range	Mean ± SD	Range	Mean ± SD	Range
Original; U-Net [21]	0.66±0.15; (p < 0.001)	0~0.84	0.71±0.17	0.05~0.88	0.65±0.17	0.04~0.92
Segnet [27]	0.55±0.20; (p < 0.001 )	0.01~0.81	0.64±0.25	0.01~0.93	0.51±0.19	0.02~0.89
GCN [28]	0.57±0.22; (p < 0.001 )	0.02~0.83	0.77±0.28	0~0.92	0.48±0.22	0.02~0.78
Proposed method	0.74±0.17	0~0.92	0.75±0.16	0~0.88	0.72±0.15	0~0.99

Notes: DSC, dice similarity coefficient. The p values are obtained by using two-sided paired Wilcoxon signed-rank tests.