Automation of Wilms’ tumor segmentation by artificial intelligence

Cancer Imaging

Table 2 Dice indices and estimated time saving for the segmentation of neoplastic kidneys with manual segmentation (inter-individual variability), automatic segmentation (CNN U-Net) and semi-automatic segmentation (CNN U-Net trained with the OV²ASSION method)

Patient	Inter-individual variability (manual segmentation)	Automatic segmentation CNN U-Net	Semi-automatic segmentation CNN U-Net + OV²ASSION
Patient	Inter-individual variability (manual segmentation)	Automatic segmentation CNN U-Net	Gap 1	Gap 5	Gap 10
1	0.83	0.15	0.97	0.95	0.92
2	0.96	0.43	0.97	0.96	0.95
3	0.92	0.41	0.99	0.98	0.97
4	0.88	0.20	0.97	0.95	0.94
5	0.89	0.56	0.93	0.92	0.90
6	0.82	0.17	0.93	0.91	0.88
7	0.92	0.15	0.93	0.86	0.84
8	0.92	0.19	0.97	0.96	0.94
9	0.69	0.05	0.84	0.79	0.45
10	0.79	0.26	0.86	0.82	0.77
11	0.88	0.02	0.94	0.91	0.85
12	0.88	0.23	0.90	0.84	0.82
13	0.88	0.46	0.93	0.91	0.88
14	0.87	0.46	0.99	0.98	0.98
Average	0.87	0.27	0.94	0.91	0.86
Estimated time saving	0%	100%	33%	71%	83%

ISSN: 1470-7330