New Explainable Deep CNN Design for Classifying Breast Tumor Response Over Neoadjuvant Chemotherapy

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Artificial Intelligence in Medicine

Background:

• Neoadjuvant Chemotherapy (NAC) is crucial for reducing breast tumor size before surgery.
• Current clinical protocols lack early prediction of chemotherapy response, leading to potential toxicity and treatment delays.
• Computational analysis of Dynamic Contrast-Enhanced Magnetic Resonance Images (DCE-MRI) shows promise in predicting patient response.

Purpose of the Study:

• To develop an explainable Deep Learning (DL) model for predicting breast cancer response to chemotherapy.
• To utilize multiple MRI inputs for enhanced prediction accuracy.
• To improve patient outcomes by enabling early prediction of treatment effectiveness.

Main Methods:

• A cohort of 42 breast cancer patients was used for training and validation.
• An external dataset of 14 subjects was used for independent validation.
• A multi-input Deep Convolutional Neural Network (CNN) model was developed and assessed using AUC and accuracy.

Main Results:

• The DL model achieved an AUC of 0.93 in predicting chemotherapy response on an external validation dataset.
• Feature visualization indicated that non-responding tumors had key features in peripheral regions.
• The model demonstrated higher efficiency compared to existing methods.

Conclusions:

• The developed multi-input CNN model accurately predicts breast cancer response to chemotherapy even with limited data.
• Visualizations of extracted features aid in understanding tumor response.
• Further evaluation with larger datasets is recommended for clinical integration.