. Author manuscript; available in PMC: 2023 Sep 18.

Published in final edited form as: Abdom Radiol (NY). 2022 Jan 31;47(4):1425–1434. doi: 10.1007/s00261-022-03419-2

Table 5.

Lesion-level detection performance

Split	Model	Sensitivity (95% CI)	PPV (95% CI)	Mean FPs/exam (range)	Detection rate by lesion-level ISUP Grade Groups
Split	Model	Sensitivity (95% CI)	PPV (95% CI)	Mean FPs/exam (range)	1	2	3	4	5

Validation	UNet	0.709 (0.60,0.81)	0.355 (0.27,0.45)	1.41 (0–6)	14/22	27/32	8/12	9/14	3/6
	AHNet	0.744 (0.66,0.83)	0.478 (0.38,0.58)	0.87 (0–5)	15/22	25/32	9/12	11/14	4/6
Test	UNet	0.728 (0.62,0.83)	0.285 (0.21,0.36)	1.90 (0–7)	11/20	22/30	10/12	10/13	6/6
	AHNet	0.630 (0.53–0.73)	0.319 (0.24,0.41)	1.40 (0–6)	12/20	18/30	7/12	9/13	5/6

95% Confidence Intervals produced from bootstrap analysis on the patient-level

ISUP International Society of Urological Pathology