Deep Learning Combined with Radiologist’s Intervention Achieves Accurate Segmentation of Hepatocellular Carcinoma in Dual-Phase Magnetic Resonance Images

Table 1

Results of DCNN-F, DFN-F, DFN-R, and the segmentation by the radiologist with moderate experience. Comparisons of performance were performed by using paired -test.

Model	DSC			Precision			Recall			value
		Median	Range		Median	Range		Median	Range
DCNN-F		0.72	0.01~0.90		0.78	0.01~0.91		0.75	0.01~0.89	0.03
DFN-F		0.77	0.08~0.96		0.72	0.01~0.96		0.91	0.01~0.98	0.26
DFN-R		0.88	0.42~0.96		0.87	0.21~0.99		0.91	0.48~0.99	0.69
Radiologist with moderate experience		0.83	0.45~0.96		0.94	0.34~0.99		0.75	0.47~0.99	—

A value < 0.05 indicates that DSC results of one model are significantly different from those by the radiologist with moderate experience. Notes: DCNN: deep convolutional neural networks; DFN: deep fusion network; DSC: Dice similarity coefficient; std: standard deviation.