Table 5:

Patient-level evaluation on cohorts C1-test and C3-test (Total number of subjects N = 70, including 55 men with cancer from C1-test and 15 men with normal MRI from C3-test. Column “Final rank” represents rank of model based on sum of individual ranks (detailed in Supplementary material table S3). CorrSigNIA outperforms all other models in patient-level evaluation.

Model	Sensitivity	Specificity	Precision	NPV	F1 score	Accuracy	Final Rank
SPCNet	0.71	0.80	0.93	0.43	0.80	0.73	3
U-Net	0.87	0.00	0.76	0.00	0.81	0.69	13
Branched U-Net	0.87	0.13	0.79	0.22	0.83	0.71	9
CorrSigNIA	0.78	0.87	0.96	0.52	0.86	0.80	1
SPCNet + CR-3-S	0.73	0.53	0.85	0.35	0.78	0.69	12
SPCNet + CR-123-I	0.71	0.80	0.93	0.43	0.80	0.73	3
SPCNet + CR-123-S	0.67	0.87	0.95	0.42	0.79	0.71	7
SPCNet + CR-4-I	0.67	0.80	0.93	0.40	0.78	0.70	10
SPCNet + CR-4-S	0.78	0.73	0.91	0.48	0.84	0.77	2
U-Net + CR-3-I	0.91	0.07	0.78	0.17	0.84	0.73	8
U-Net + CR-123-I	0.89	0.07	0.78	0.14	0.83	0.71	10
BrU-Net + CR-3-I	0.91	0.00	0.77	0.00	0.83	0.71	5
BrU-Net + CR-123-I	0.78	0.53	0.86	0.40	0.82	0.73	5