Table 4:

Detecting clinically significant cancer: Lesion-level evaluation on cohorts C1-test and cohort C2.

Evaluation on cohort C1-test
Patients with clinically significant cancer with lesion volume ≥ 250 mm3 N = 34, number of clinically significant cancerous lesions = 37
Model	ROC AUC	PR AUC	Sensitivity	Specificity	Precision	NPV	F1 Score	Dice	Accuracy	Agg cancer overlap	Final Rank
SPCNet	0.83±0.30	0.77±0.37	0.68±0.47	0.76±0.29	0.47±0.41	0.88±0.21	0.52±0.41	0.29±0.25	0.75±0.24	0.18±0.28	4
U-Net	0.84±0.32	0.81±0.34	0.88±0.34	0.33±0.31	0.35±0.28	0.65±0.46	0.46±0.27	0.30±0.20	0.46±0.28	0.30±0.33	7
Branched U-Net	0.81 ± 0.32	0.75±0.38	0.84±0.36	0.39±0.37	0.34±0.28	0.62±0.46	0.45±0.28	0.27±0.18	0.51±0.27	0.20±0.26	12
CorrSigNIA	0.82±0.31	0.74±0.38	0.79±0.40	0.72±0.33	0.53±0.40	0.88±0.25	0.59±0.39	0.32±0.24	0.73±0.27	0.20±0.29	2
SPCNet + CR-3-S	0.79±0.31	0.70±0.39	0.66±0.47	0.79±0.22	0.45±0.39	0.91±0.13	0.51±0.40	0.28±0.23	0.77±0.18	0.14±0.26	8
SPCNet + CR-123-I	0.83±0.32	0.76±0.37	0.71±0.46	0.76±0.29	0.50±0.41	0.89±0.21	0.55±0.41	0.31±0.23	0.76±0.23	0.18±0.28	1
SPCNet + CR-123-S	0.77±0.34	0.73±0.39	0.60±0.48	0.81±0.25	0.44±0.41	0.86±0.20	0.48±0.42	0.29±0.24	0.77±0.21	0.17±0.27	9
SPCNet + CR-4-I	0.82±0.33	0.78±0.36	0.63±0.47	0.80±0.30	0.47±0.43	0.85±0.25	0.51±0.43	0.29±0.23	0.76±0.25	0.16±0.26	6
SPCNet + CR-4-S	0.81±0.32	0.76±0.37	0.72±0.44	0.72±0.32	0.50±0.41	0.85±0.29	0.55±0.39	0.31±0.23	0.72±0.27	0.21±0.29	3
U-Net + CR-3-I	0.82±0.32	0.75±0.36	0.93±0.25	0.24±0.28	0.31±0.19	0.48±0.49	0.44±0.20	0.28±0.18	0.41±0.23	0.36±0.29	10
U-Net + CR-123-I	0.83±0.29	0.75±0.37	0.90±0.29	0.41±0.37	0.39±0.29	0.66±0.46	0.50±0.27	0.31±0.21	0.53±0.29	0.46±0.34	5
BrU-Net + CR-3-I	0.80±0.36	0.77±0.37	0.91±0.28	0.09±0.20	0.26±0.17	0.22±0.41	0.39±0.19	0.26±0.16	0.29±0.18	0.53±0.30	13
BrU-Net + CR-123-I	0.73±0.41	0.70±0.40	0.72±0.44	0.56±0.38	0.41±0.37	0.70±0.39	0.49±0.37	0.27±0.23	0.60±0.31	0.42±0.36	11
Evaluation on cohort C2
Patients with clinically significant cancer N = 90, number of clinically significant cancerous lesions = 110
Model	ROC AUC	PR AUC	Sensitivity	Specificity	Precision	NPV	F1 Score	Dice	Accuracy	Agg cancer overlap	Final Rank
SPCNet	0.79±0.34	0.79±0.34	0.58±0.47	0.74±0.37	0.48±0.42	0.71±0.34	0.49±0.41	0.29±0.25	0.70±0.25	0.14±0.24	7
U-Net	0.83±0.30	0.81±0.32	0.81±0.37	0.58±0.39	0.53±0.34	0.68±0.41	0.60±0.32	0.38±0.23	0.67±0.26	0.26±0.28	3
Branched U-Net	0.84± 0.31	0.82±0.33	0.94±0.22	0.26±0.35	0.42±0.25	0.41±0.48	0.55±0.23	0.35±0.19	0.48±0.25	0.24±0.24	4
CorrSigNIA	0.86±0.26	0.83±0.31	0.67±0.43	0.74±0.38	0.55±0.41	0.72±0.36	0.57±0.39	0.35±0.25	0.74±0.25	0.24±0.31	1
SPCNet + CR-123-I	0.81±0.30	0.79±0.33	0.65±0.45	0.80±0.34	0.54±0.42	0.77±0.31	0.56±0.41	0.36±0.21	0.76±0.21	0.21±0.29	2
U-Net + CR-3-I	0.55±0.44	0.59±0.40	0.46±0.47	0.62±0.40	0.34±0.39	0.61±0.36	0.36±0.39	0.22±0.24	0.58±0.29	0.15±0.24	9
U-Net + CR-123-I	0.72±0.37	0.70±0.37	0.61±0.46	0.70±0.37	0.48±0.42	0.70±0.35	0.50±0.41	0.28±0.25	0.67±0.28	0.26±0.31	6
BrU-Net + CR-3-I	0.86±0.31	0.85±0.30	0.93±0.24	0.12±0.23	0.35±0.17	0.25±0.42	0.49±0.18	0.35±0.19	0.39±0.19	0.62±0.30	5
BrU-Net + CR-123-I	0.67±0.39	0.67±0.38	0.49±0.47	0.80±0.33	0.44±0.45	0.73±0.30	0.44±0.43	0.25±0.26	0.71±0.25	0.23±0.30	8

Mean and standard deviation values for each metric in each cohort is reported. Column “Final rank” represents rank of model based on sum of individual ranks (details in Supplementary material table S2). CorrSigNIA ranks second in cohort C1-test, closely following another CorrSigNIA variant, whereas it ranks first in cohort C2.