Detecting MRI-Invisible Prostate Cancers Using a Weakly Supervised Deep Learning Model

Table 3

Precision (detection rate) of systematic biopsy and WSUNet.

PI-RADS	Systematic biopsy	WSUNet	Ratio	Biopsy saving ratio (number)	value
All (1-5)
Training set	0.047 [333/7083]	0.090 [272/3040]	1.904	0.475 (3361)	<0.005
Testing set	0.058 [118/2025]	0.111 [94/846]	1.913	0.476 (963)	<0.005
≥3
Training set	0.047 [323/6902]	0.088 [263/2993]	1.870	0.489 (3226)	<0.005
Testing set	0.059 [117/1954]	0.112 [94/843]	1.893	0.463 (905)	<0.005

Note. Data are percentages, with numbers of participants in brackets. Biopsy saving ratio (number) refers to the proportion (number) of biopsy cores that can be reduced by the proposed model when the same number of positive cores is detected. . . . Precision refers to the proportion of true positive results among all positive results, which might reflect specific cancer detection rates in retrospective conditions. PI-RADS refers to the maximum PI-RADS for all outlined lesions in one examination. Since most of the patient’s PI-RAD greater than 3, so this part of the data has been presented alone.