Development and validation of a machine learning-based pathomics nomogram for prognostic prediction in endometrial
1Department of Gynecology, Ganzhou People's Hospital, Ganzhou, 341000, Jiangxi, China.
Introduction:
Endometrial cancer (EC) is a common gynecologic malignancy with increasing incidence and notable molecular heterogeneity. This study aimed to develop a non-invasive, cost-effective prognostic model using pathomics features extracted from histopathological images.
Methods:
We retrospectively analyzed hematoxylin and eosin (H&E)-stained whole slide images (WSIs) and clinical data from 514 EC patients in the TCGA-UCEC cohort. Tumor regions were manually annotated, preprocessed, and tiled into patches. Using CellProfiler, 728 high-dimensional quantitative features were extracted per case. Feature selection was performed through univariate Cox regression followed by LASSO regression to build a prognostic risk score. A nomogram integrating pathomics and clinical variables was constructed and evaluated.
Results:
The final prognostic model identified a panel of pathomics features significantly associated with overall survival. Patients in the high-risk group exhibited significantly worse survival outcomes compared to those in the low-risk group (p < 0.001). In both training and validation cohorts, the model demonstrated robust predictive accuracy, with AUCs of 0.806 and 0.782 for 3-year survival, respectively. The nomogram integrating clinical parameters and the pathomics score achieved an AUC of 0.791 and showed good calibration. Multivariate analysis confirmed the independent prognostic value of the pathomics signature. Subgroup analyses revealed consistent performance across molecular subtypes and FIGO stages.
Conclusion:
This study presents a validated pathomics-based model that accurately predicts prognosis in EC patients using routine H&E slides. It provides a scalable, low-cost tool for risk stratification and personalized management in clinical practice.
