Histopathology-based Artificial Intelligence Algorithms for the Prediction of Prostate Cancer Metastasis After...

Area of Science:

• Oncology
• Artificial Intelligence
• Pathology

Background:

• Multimodal AI algorithms have predicted prostate cancer (PCa) metastasis using combined histopathology and clinical-pathologic parameters.
• This study evaluated purely histopathology-based AI algorithms for predicting lethal PCa risk in surgically treated cohorts.

Purpose of the Study:

• To assess the efficacy of histopathology-based AI algorithms in predicting lethal prostate cancer (PCa) metastasis.
• To compare the performance of AI algorithms with genomic classifiers and standard clinical risk tools.

Main Methods:

• Utilized whole slide images (WSIs) and tissue microarrays (TMAs) from radical prostatectomy (RP) and needle biopsy samples across five PCa cohorts.
• Developed a concatenated feature-based classification system using histopathologic data to generate an AI risk score for metastasis.

Main Results:

• AI risk scores from prostatectomy WSIs demonstrated performance comparable to genomic classifiers (C-index: 0.81-0.85 vs. 0.72-0.80).
• A modified TMA AI score achieved a C-index of 0.71 in a large nationwide study and 0.74 in a needle biopsy cohort.
• Combining the TMA AI score with the Cancer of the Prostate Risk Assessment (CAPRA) score improved predictive performance (C-index: 0.83).

Conclusions:

• Histopathology-based AI algorithms applied to small tumor tissue samples can predict the risk of lethal PCa.
• These AI algorithms perform comparably to established genomic classifiers.
• Combining AI algorithms with clinicopathologic variables enhances predictive performance for lethal PCa.