Prediction of Breast Cancer Response to Neoadjuvant Therapy with Machine Learning: A Clinical, MRI-Qualitative, and...

Area of Science:

• Oncology
• Radiology
• Biomedical Engineering

Background:

• Pathological complete response (pCR) is a key endpoint for breast cancer neoadjuvant systemic therapy (NAST).
• Predicting pCR and recurrence-free survival (RFS) is crucial for optimizing treatment strategies.
• Machine learning (ML) offers potential for developing novel predictive biomarkers.

Purpose of the Study:

• To develop and evaluate ML-based biomarkers for predicting pCR and RFS in breast cancer patients.
• To assess the utility of integrating clinical, morphological, and radiomics data for prediction.
• To identify optimal ML models for diverse breast cancer subtypes.

Main Methods:

• Retrospective analysis of 235 non-metastatic breast cancer patients treated with NAST.
• Development of ML models using clinical data, pre-treatment MRI morphological features, and radiomics.
• Implementation of a customized ML pipeline involving feature selection and classification.

Main Results:

• ML models incorporating radiomics data demonstrated superior prediction of pCR (AUC 0.72).
• Optimal prediction performance was observed in triple-negative breast cancer (AUC 0.80) and HER2-positive subgroups (AUC 0.65).
• Radiomics features significantly enhanced predictive capabilities compared to clinical data alone.

Conclusions:

• ML models integrating clinical, morphological, and radiomics data from pre-treatment MRI can effectively predict pCR in NAST-treated breast cancer.
• These models show particular promise for predicting outcomes in triple-negative and HER2-positive breast cancer subgroups.
• The findings support the use of ML-driven radiomics for personalized neoadjuvant therapy selection.