AI-based prostate analysis system trained without human supervision to predict patient outcome from tissue samples

Area of Science:

• Computational pathology
• Oncology
• Artificial intelligence in medicine

Background:

• Prostate cancer treatment planning relies on tumor aggressiveness grading using the Gleason scale, a subjective visual assessment of biopsies.
• Existing AI models for Gleason grading are limited by the accuracy of the 'ground truth' human grades used for training.
• There is a need for more objective and accurate methods to predict patient outcomes in prostate cancer.

Purpose of the Study:

• To develop and evaluate an AI model capable of predicting prostate cancer patient outcomes directly from histopathology images.
• To assess the AI model's performance independently of subjective human-derived Gleason grading.

Main Methods:

• Trained an AI model using image analysis of a large biobank of prostate tissue samples with known patient outcomes.
• The AI model was developed without incorporating any human knowledge of cancer grading scales.
• Evaluated the AI model's predictive performance on an independent test dataset.

Main Results:

• The AI model demonstrated a significant ability to predict patient outcomes directly from tissue image analysis.
• The model's predictive performance was comparable to, and in some cases exceeded, that of expert pathologists using conventional Gleason grading.
• This suggests a potential for AI to provide more accurate prognostic information than current subjective grading methods.

Conclusions:

• Direct AI-driven outcome prediction from prostate cancer tissue images offers a promising alternative to traditional Gleason grading.
• This approach may overcome limitations associated with subjective ground truth grading, leading to improved patient management.
• Further validation could establish this AI as a valuable tool in precision oncology for prostate cancer.