A Radiomic-Based Machine Learning Model Predicts Endometrial Cancer Recurrence Using Preoperative CT Radiomic...

Area of Science:

• Medical Imaging
• Oncology
• Machine Learning

Background:

• Current endometrial cancer (EC) prognostic models do not incorporate pre-operative CT imaging.
• Predicting recurrence is crucial for tailoring treatment and improving outcomes in EC patients.

Purpose of the Study:

• To investigate the utility of radiomic features from pre-surgical CT scans for predicting disease-free survival (DFS) in endometrial cancer (EC) patients.
• To develop and evaluate machine learning (ML) models for forecasting EC recurrence risk.

Main Methods:

• Radiomic features were extracted from contrast-enhanced CT (CE-CT) scans of 81 EC patients.
• A 10-fold cross-validation with a 6:4 training/test split was used, incorporating data augmentation and balancing.
• Three ML models (LASSO-Cox, CoxBoost, RFsrc) were developed for DFS prediction after feature reduction.

Main Results:

• ML models achieved high performance in both training (AUCs 0.92-0.93) and test sets (AUCs 0.86-0.90).
• Sensitivities ranged from 0.89-1.00 and specificities from 0.73-0.90 across models and datasets.
• High recurrence risk predicted by ML models correlated with significantly worse DFS (p < 0.001).

Conclusions:

• Radiomics shows significant potential for predicting endometrial cancer recurrence.
• The developed ML models demonstrate a promising role in forecasting EC patient outcomes.
• Further validation studies are warranted to confirm these findings in larger cohorts.