Table 1.

Comparison of performance between three deep learning model for vertebral body segmentation. R2U-net showed the highest performance in the four conditional probability values, and the values are shown in bold.

	U-net (± SD)	ResU-net (± SD)	R2U-net (± SD)
Sensitivity	0.930 (± 0.116)	0.925 (± 0.065)	0.934 (± 0.086)
Specificity	0.997 (± 0.001)	0.997 (± 0.003)	0.998 (± 0.002)
Accuracy	0.987 (± 0.006)	0.987 (± 0.005)	0.987 (± 0.005)
DSC	0.903 (± 0.090)	0.920 (± 0.068)	0.923 (± 0.073)

SD standard deviation, DSC dice similarity coefficient.