Multi-input deep learning architecture for predicting breast tumor response to chemotherapy using quantitative MR...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Oncology
• Breast Cancer Research

Background:

• Neoadjuvant chemotherapy (NAC) is crucial for reducing tumor size before surgery in breast cancer patients.
• Predicting response to NAC is vital for optimizing treatment, minimizing toxicity, and avoiding delays.
• Deep learning models, particularly Convolutional Neural Networks (CNNs), show potential in analyzing medical images for treatment response prediction.

Purpose of the Study:

• To develop and present a novel deep learning (DL) model for predicting breast cancer response to NAC.
• To utilize multiple dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) inputs for enhanced prediction accuracy.
• To distinguish between pathological complete response (pCR) and non-pCR patients.

Main Methods:

• A DL model was trained and validated using 723 axial slices from 42 breast cancer patients who received NAC.
• The model was further validated on 14 external cases using pre- and post-chemotherapy DCE-MRI data.
• Performance was evaluated using Area Under the Receiver Operating Characteristic Curve (AUC), accuracy, sensitivity, and specificity.

Main Results:

• The multi-input DL architecture achieved an AUC of 0.91 in predicting pCR using combined pre- and post-NAC images.
• Feature map visualization indicated that peripheral regions of non-pCR tumors were most significant.
• The proposed method demonstrated superior performance compared to previous approaches.

Conclusions:

• The developed CNN model accurately classifies pCR and non-pCR patients using DCE-MR images acquired before and after the first chemotherapy cycle.
• Despite a limited training dataset, the model shows substantial accuracy.
• Further evaluation with larger datasets is recommended for clinical implementation.