Table 10 The optimization sets used to select the best models for high-grade intraprostatic intraepithelial neoplasia, vessel, nerve, and inflammatory cell infiltration.

From: Critical evaluation of artificial intelligence as a digital twin of pathologists for prostate cancer pathology

Finding

Regions, n (case, n)

Magnification level

Positive, n (%)

Negative, n (regions, n; cases, n)

List of negative findings (%)

Cribriform pattern

87 (12)

10×

197 (32.9)

402 (67.1)

(197, 25)

Gleason pattern 3 (41.0)

Gleason pattern 4 without cribriform pattern (9.4)

Nerve (36.7)

Normal glandular tissues (12.9)

Ductal morphology

36 (1)

10×

59 (3%)

1906 (97%)

(496, 42)

Gleason pattern 3 (8.3)

Gleason pattern 4 (11.4)

Gleason pattern 5 (56.1)

Nerve (8.6)

Blood vessel (6.8)

Cribriform pattern (6.9)

Normal glandular tissues (1.9)

HGPIN

16 (7)

10×

204 (2.5)

8104 (97.5) (581, 35)

Benign Prostatic

Hyperplasia (5.8)

Gleason pattern 3 (11.4)

Gleason pattern 4 (16.2)

Gleason pattern 5 (66.6)

Blood vessel

96 (18)

10×

123 (6.3)

1820 (93.7)

(447,40)

Gleason pattern 3 (9.6)

Gleason pattern 4 (10.9)

Gleason pattern 5 (57.8)

Nerve (8.2)

Cribriform pattern (7.9)

Normal glandular tissues (2.2)

Ductal adenocarcinoma (3.4)

Nerve

147 (17)

10×

157 (8.0)

1812 (92.0)

(392,41)

Gleason pattern 3 (10.2)

Gleason pattern 4 (12.7)

Gleason pattern 5 (56.8)

Cribriform pattern (6.7)

Vessel (6.3)

Normal glandular tissues (3.1)

Ductal adenocarcinoma (4.2)

Inflammatory cell infiltration

109 (22)

10×

161 (7.8)

1904 (92.2)

(511, 42)

Gleason pattern 3 (10.0)

Gleason pattern 4 (11.4)

Gleason pattern 5 (54.1)

Nerve. (7.3)

Cribriform pattern (6.5)

Vessel (5.5)

Normal glandular tissues (1.7)

Ductal adenocarcinoma (3.5)

  1. These datasets were curated from the PRAD-TCGA histology image dataset. We emphasize that we added images representing Gleason pattern 5 as an unseen pathological finding. A total of 42 additional patients were considered for the optimization set. For the HGPIN, we increased the patch number to increase.