Grading a prostate biopsy on the Gleason scale · Contested (since 2022)

In January 2022 Nature Medicine reported the PANDA challenge, where algorithms trained on 10 616 digitised prostate biopsies reached agreement of 0.862 and 0.868 with expert uropathologists on blinded American and European cohorts.

The measure is agreement, not accuracy, because pathologists disagree with each other and with themselves on this scale, which is exactly why the grade was worth automating. The challenge was retrospective and read from scanned slides, and the authors ask for prospective clinical trials before anyone is treated on a machine grade. The number matters because it decides between active surveillance, surgery and radiotherapy, and each of those has a different cost to a man's life. That conversation, held in a room with him, has not moved. Nor has the responsibility of the pathologist who signs the report.