Predicting prostate cancer grade reclassification on active surveillance using a deep learning-based grading...

Area of Science:

• Oncology
• Artificial Intelligence
• Pathology

Background:

• Deep learning (DL) algorithms for prostate cancer (PCa) Grade Group (GG) determination on biopsy slides lack clinical outcome validation.
• Accurate grading is crucial for managing PCa, especially in active surveillance (AS) protocols.

Purpose of the Study:

• To validate the utility of a DL-based algorithm (AIRAProstate) for regrading prostate biopsies in independent PCa active surveillance cohorts.
• To assess the association of DL-based regrading with clinical outcomes like grade reclassification during AS.

Main Methods:

• Utilized the AIRAProstate DL algorithm to regrade initial prostate biopsies from two independent PCa AS cohorts.
• Analyzed the association between AIRAProstate-based upgrading (to GG≥2) and subsequent grade reclassification on AS.
• Compared DL-based upgrading with contemporary uropathologist reviews in one cohort.

Main Results:

• In the first cohort (n=138, initial GG1), AIRAProstate upgrading was significantly associated with grade reclassification on AS (OR=3.3, P=0.04), unlike uropathologist reviews.
• In the validation cohort (n=169, all initial GG1), AIRAProstate upgrading also predicted grade reclassification on AS (HR=1.7, P=0.03).

Conclusions:

• The DL-based AIRAProstate algorithm demonstrates significant utility in predicting grade reclassification for prostate cancer patients on active surveillance.
• This AI tool shows promise for improving risk stratification and clinical decision-making in PCa management.