Prediction of endometrial cancer recurrence using deep learning analysis of histopathology slides

Andrew Polio; Katherine Boecking; Vincent Wagner; David Bender; Michael Goodheart; Jesus Gonzalez-Bosquet

doi:10.1016/j.ygyno.2025.04.435

Objectives Endometrial cancer is the most common gynecologic cancer in the United States, with a rising incidence and mortality. Despite optimal surgical and adjuvant treatment, approximately 15–20 % of all patients will recur. Studies have identified multiple clinical and pathologic factors associated with recurrence, but models of disease relapse remain inconsistent. To better select patients for adjuvant therapy, it is important to accurately predict patients at risk for recurrence. The objective of this study was to train, validate and test models of endometrioid endometrial cancer (EEC) recurrence using deep learning (DL) analytics of histopathologic slides from endometrial cancer tumors. Methods Paraffin-embedded hematoxylin and eosin (H&E) slides from recurrent and nonrecurrent EEC samples were obtained from The Cancer Genome Atlas (TCGA) database. These were stratified into low risk (grades 1, 2, stage I; recurrent n = 16, nonrecurrent n = 150) and high risk (grade 3, stage II, III, IV; recurrent n = 44, nonrecurrent n = 196) samples. The slides were processed with solid tumor associative modeling in pathology (STAMP) using preprocessing self-supervised learning pre-trained histology feature extractor and transformer architecture. Predictive models of EEC recurrence were trained and cross-validated with 5 k-fold with DL architecture. Testing was done in 15 % of samples not used to train models. The performance of the models was assessed using the area under the receiver operating characteristic curve (AUC), the area under the precision-recall curve (AUPRC) and their 95 % confidence intervals (CI). Results Prediction models of low-risk EEC recurrence validated with 5 k-fold cross-validation and tested in independent EEC had performance, measured by the AUPRC of 0.53 (95 % CI 0.27–0.75) and AUC of 0.89 (95 % CI 0.81–0.96). A prediction model of low-risk EEC recurrence, built only with clinical data, had an AUC of 0.56 (95 % CI 0.46–0.67). Prediction models of high-risk EEC recurrence had an AUPRC of 0.53 (95 % CI 0.32–0.87) and AUC of 0.76 (95 % CI 0.53–0.94), while prediction models with clinical data had and AUC of 0.81 (95 % CI 0.67–0.95). Conclusions Prediction models of endometrial cancer recurrence based on histopathology slides had improved performance over clinical models for low-risk disease but performed similarly for high-risk disease. Integration of these models with multimodal transformer architecture could improve performance. Further studies are required to determine clinical utility.

Prediction of endometrial cancer recurrence using deep learning analysis of histopathology slides

View Online

Abstract

Details

Metrics