Mid-Treatment Delta MRI Radiomics Enhances Sensitivity and Risk Stratification for Pathological Complete Response in...

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Machine Learning in Medicine

Background:

• Accurate prediction of pathological complete response (pCR) to neoadjuvant chemotherapy in breast cancer is crucial for treatment optimization.
• Current methods for predicting pCR often face limitations in accuracy and early prediction capabilities.

Purpose of the Study:

• To develop and validate mid-treatment magnetic resonance imaging (MRI) delta radiomics models for predicting pCR in breast cancer patients.
• To assess the performance of combined clinical and delta radiomics models compared to clinical features alone.

Main Methods:

• A multicenter retrospective study involving 242 breast cancer patients.
• Extraction of delta radiomics features from mid-treatment vs. pre-treatment T1-weighted contrast-enhanced MRI sequences.
• Development of six models using Random Forest, combining clinical and delta radiomics features, validated internally and externally.

Main Results:

• The combined clinical-delta radiomics model achieved an external validation AUC of 0.747, showing comparable performance to the clinical-only model (AUC=0.777).
• The combined model significantly improved sensitivity (90.6%) compared to the clinical-only model (65.6%) for pCR prediction.
• Effective risk stratification was demonstrated, with a 51%-point separation in pCR rates between predicted responder and non-responder groups.

Conclusions:

• Mid-treatment delta MRI radiomics, when combined with clinical features, provides high sensitivity and effective risk stratification for pCR prediction in breast cancer.
• While not significantly improving AUC over clinical features alone, the enhanced sensitivity and risk stratification suggest potential clinical utility.
• Larger prospective validation studies are warranted to confirm the clinical applicability of these findings.