Time-to-Event Machine Learning Model Incorporating MRI-Derived Intratumoral Heterogeneity Score for Predicting...

Area of Science:

• Oncology
• Radiology
• Machine Learning

Background:

• Postoperative recurrence is a significant challenge in invasive breast cancer (IBC).
• Conventional methods may not fully capture tumor heterogeneity or recurrence patterns.
• Predicting recurrence is crucial for effective patient management.

Purpose of the Study:

• To evaluate the prognostic value of a magnetic resonance imaging (MRI)-derived intratumoral heterogeneity (ITH) score.
• To incorporate the ITH score into time-to-event machine learning (ML) models for predicting postoperative recurrence in IBC.
• To assess the performance of ML models in predicting recurrence-free survival (RFS).

Main Methods:

• Retrospective dual-center study of 428 IBC patients.
• Calculation of ITH scores from contrast-enhanced T1-weighted MRI texture features.
• Development and validation of time-to-event ML models, including random survival forest (RSF).
• Performance assessment using C-index, ROC curves, calibration curves, and SHAP analysis.

Main Results:

• The RSF model demonstrated strong predictive performance (C-index 0.826 in validation, 0.814 in testing).
• The MRI-derived ITH score was identified as the most critical predictor of RFS.
• Higher ITH scores correlated with shorter RFS and increased tumor proliferation markers.

Conclusions:

• Integrating an MRI-derived ITH score with RSF models offers a noninvasive, interpretable framework for IBC recurrence prediction.
• This approach can aid in individualized risk stratification and treatment planning.
• Further prospective validation is needed for routine clinical use.