Table 4 Image group assessment by pathologist in accordance with the BCR score risk stratification with decision explanation.

From: Artificial intelligence unravels interpretable malignancy grades of prostate cancer on histology images

Pathologist

Sorting agreement with AI

Explanation

Pathologist A

Yes

Group A: Dominant Gleason pattern 3

Group B: Mixed Gleason pattern 3 and 4, cannot determine which pattern is dominant.

Group C: Mixed Gleason pattern 3 and 4, but Gleason pattern 4 is dominant.

Group D: Few Gleason pattern 3, mostly Gleason pattern 4 and/or 5.

Pathologist B

Yes

Group A has the most favorable malignancy grade (mostly Gleason pattern 3), whereas Group D has the worst differentiation grade (hard to identify Gleason pattern 3, mostly Gleason pattern 4 and 5). Cannot determine a significant difference in pattern distribution in Group B and C, but it seems to me that C has more Gleason pattern 4.

Pathologist C

Yes

There is a pattern trend in these image groups, likely driven by Gleason pattern 3. For example, Group A has mostly Gleason pattern 3 and Group D has mostly Gleason pattern 4 and 5. This trend is, however, not clearly visible in group B and C (mixed Gleason patterns 3 and 4).

Pathologist D

Partially

Group A: Dominantly favorable Gleason pattern 3

Group B: Mixed Gleason patterns 3 and 4

Group D: Dominantly Gleason pattern 4 and 5 (Mostly Gleason score at least 8)

Group C: Mostly heterogenous prostate cancer constellation with dominance of Gleason pattern 4.

Pathologist E

Yes

Group A: Dominant Gleason pattern 3. Rare occurrences of mucinous cribriform, glomeruloid patterns

Group B: Mixed Gleason pattern 3 and 4, with rare occurrence of single cells (pattern 5). When pattern 4 is present, it is mostly ill-defined, sometimes cribriform.

Group C: Dominant pattern 4, with rare occurrences of pure pattern 3. Pattern 4 is often cribriform.

Group D: Dominant patterns 4 and 5.

  1. The image groups, which were randomly assigned, required sorting by the pathologists. Pathologist D weighted the heterogeneity definition to rank group D and C, thereby resulting into the partial agreement.