A multi-institutional validation study of predicting prostate cancer-specific mortality with AI-based grading

Purpose and goal:

Treatment decisions for prostate cancer, one of the world´s most common cancers, are primarily based on subjective histopathological grading according to Gleason. With a correct grading, treatment can be adapted to actual need. However, the subjective assessment is uncertain, with large variation between pathologists. We have trained an AI model on patient outcomes showing superior performance. We will now validate the model in prospective and retrospective studies to ensure its reliability.

Expected results and effects:

We will create an AI-system that provides more reliable prostate cancer aggressiveness grades than the present state of the art, Gleason grades. The system will be robust and trustworthy for samples from a variety of different clinics. This will enable improved diagnosis and treatment selection for the increasing number of prostate cancer patients around the world, not the least in India. Our work will advance understanding of how AI algorithms can be made sufficiently reliable for clinical use.

Approach and implementation:

Our method will be validated on material from the Regional Cancer Center in Kerala, India. From archived paraffin blocks of prostate cancer, H&E sections are acquired and assessed for tumor presence and our prostate aggressiveness index is computed. Treatment and follow-up details are obtained from patient case sheets. Histomorphometric algorithms for cancer aggressiveness grading will also be explored. The project results will be documented in top quality international scientific publications.