. 2022 Nov 29;3(12):100642. doi: 10.1016/j.patter.2022.100642

Table 3.

External validation set performance of algorithms in prostate cancer detection

Study	External dataset (B: benign, M: malignant)	AUROC (95% CI)
Campanella et al.⁸	12,727 core needle biopsy slides (B = 314, M = 12,413)	0.932 (0.911–0.952)^a
Pantanowitz et al.¹³	355 parts with multiple slides (B = 225, M = 130)	0.991 (0.979–1.000)^b
Ström et al.¹⁴	330 core needle biopsy slides (B = 108, M = 222)	0.986 (0.972–0.996)
Bulten et al.¹⁵	245 tissue microarray cores (B = 10, M = 235)	0.988 (0.984–1.000)
Current study	280 core needle biopsy parts (B = 179, M = 81)	0.992 (0.985–0.997)

AUROC, area under receiver operating characteristics curve; CI, confidence interval. See also Table S2.

CI was obtained on provided predictions in the paper using the percentile bootstrap method.²²

Model was fine-tuned on 44 parts from the external site. Average of four slides per part.