OncoLens AI: When a Machine Learns to Read a Biopsy and Explains What It Sees

A Deep Learning System for Explainable Prostate Cancer Detection Using SAM-2 Segmentation

Abstract

Prostate cancer is the second most common cancer that is the diagnosed in the world but in the Pakistan its more dangerous due to the lack of the knowledge and the less access to the health care professionals. Oncolens Ai is the deep learning model that aims to address this gap present in the healthcare. The Oncolens ai currently applies the SAM-2 for the analysis of the biopsy histopathology images sourced from the Radboud Institute and the performing pixel-level tissue segmentation to identify the cancer regions. Initially the model is the trained on the 120 biopsy images out of the 5000. This segmentation is coupled with an Explainable Ai layer built on the GRAD-CAM which generates color-coded visual heatmaps that reveal the exactly which tissue features drove each prediction enables the pathologist to verify the system’s output. Oncolens Ai is not designed to replace the clinical expertise. It is designed to help the healthcare professionals and the always available diagnostic assistant in the hands of every physician. This article presents the projects methodology, early results, active challenges, and the full scope of what AI will mean for the healthcare systems when in the production.