Machine learning for predicting breast-conserving surgery candidates after neoadjuvant chemotherapy based on DCE-MRI

Area of Science:

• Radiology
• Oncology
• Medical Imaging
• Machine Learning

Background:

• Neoadjuvant chemotherapy (NAC) is a common treatment for breast cancer.
• Predicting response to NAC is crucial for treatment planning and surgical decisions.
• Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) provides valuable information on tumor changes after NAC.

Purpose of the Study:

• To investigate a machine learning method for predicting breast-conserving surgery (BCS) candidates.
• To utilize pre- and post-NAC DCE-MRI data for prediction.
• To identify radiomics features that correlate with treatment response and surgical outcomes.

Main Methods:

• Retrospective study of 75 patients who underwent NAC and breast surgery.
• Extraction of 3,390 radiomics features from pre- and post-NAC DCE-MRIs.
• Development of logistic regression (LR) models to classify patients into response groups (pathologic complete response vs. residual disease).
• Evaluation of model performance using area under the curve (AUC), accuracy, sensitivity, and specificity.

Main Results:

• Radiomics features showed significant correlation with breast cancer shrinkage post-NAC.
• A combination model using pre- and post-NAC DCE-MRI achieved an AUC of 0.82.
• Pre-NAC and post-NAC models yielded AUCs of 0.64 and 0.70, respectively.
• The combined model identified 15 key features, including wavelet-based and Laplacian-of-Gauss (LoG) features.

Conclusions:

• Machine learning models integrating pre- and post-NAC DCE-MRI data can effectively predict BCS candidates.
• AI-based decision support can enhance clinical safety and treatment planning.
• High-order radiomics features (LoG, wavelet-based) are critical for accurate prediction.
• Features derived from both pre- and post-NAC DCE-MRI offer superior predictive performance.