Prediction of Chemotherapy Response in Locally Advanced Breast Cancer Patients at Pre-Treatment Using CT Textural...

Area of Science:

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

Background:

• Neoadjuvant chemotherapy (NAC) is a standard treatment for locally advanced breast cancer (LABC).
• Accurate prediction of NAC response is vital for personalizing treatment strategies and improving patient outcomes.
• Current methods for predicting NAC response in LABC patients have limitations.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting tumor response to NAC in LABC patients.
• To utilize textural computed tomography (CT) features and their wavelet coefficients for predictive modeling.
• To identify optimal feature selection methods for enhancing predictive accuracy.

Main Methods:

• A dataset of 117 LABC patients undergoing NAC was analyzed.
• 851 textural biomarkers and wavelet coefficients were extracted from CT images.
• Machine learning classifiers were trained using original image features, wavelet features, or a combination.
• Feature selection techniques including mRMR were employed to identify top predictive features.

Main Results:

• The study achieved predictive accuracies of up to 77% for hold-out data and 75% for leave-one-out cross-validation.
• The K-Nearest Neighbors (KNN) classifier with the top 5 features selected by mRMR demonstrated the best performance.
• A combination of original textural and wavelet features yielded superior predictive accuracy for NAC response.

Conclusions:

• Machine learning models integrating CT textural and wavelet features can accurately predict NAC response in LABC patients.
• This predictive capability allows for pre-treatment assessment of treatment outcomes.
• The developed model can serve as a recommender system to guide treatment modifications for LABC patients.