Artificial intelligence at the crossroads of histopathology: a new era in endometrial cancer precision diagnostics and prognostics

Abstract

This thesis explores the application of Artificial Intelligence, specially Deep Learning models, in computational pathology to improve diagnostics and prognostics in endometrial cancer (EC). While EC care has advanced through molecular classification, its clinical implementation remains challenging due to complexity, cost, and turnaround times. DL offers a promising alternative by leveraging routinely available H&E-stained slides to extract morphological features that correlate with molecular alterations and patient outcomes.

The thesis presents two key contributions: (1) im4MEC, a DL model predicting EC molecular subtypes from histology images, and (2) HECTOR, a prognostic model predicting distant recurrence to support adjuvant treatment planning. Both models emphasize explainability and potential for real-world integration. Additionally, the thesis benchmarks foundation DL models across eight EC-related tasks, offering technical insights into future directions.

By addressing limitations of current EC diagnostics and risk assessment, this work positions DL as a cost-effective, scalable tool for improving clinical decision-making. It ultimately advocates for a shift in data interpretation in histopathology and highlights AI's potential to enhance personalized cancer care.