A novel CNN algorithm for pathological complete response prediction using an I-SPY TRIAL breast MRI database.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Neoadjuvant chemotherapy (NAC) is a critical treatment for breast cancer, but predicting patient response remains challenging.
• Accurate prediction of NAC response can guide personalized treatment strategies and improve patient outcomes.
• Breast MRI is a valuable tool for monitoring tumor changes during NAC.

Purpose of the Study:

• To develop and validate a convolutional neural network (CNN) algorithm for predicting neoadjuvant chemotherapy (NAC) response in breast cancer patients.
• To assess the feasibility of using a multi-institution dataset for training and evaluating the CNN model.
• To determine the diagnostic accuracy of the CNN in distinguishing between pathological complete response (pCR) and non-pCR.

Main Methods:

• Utilized the I-SPY TRIAL breast MRI dataset comprising 131 patients from 9 institutions.
• Employed 3D segmentation of first post-contrast MRI images.
• Developed a CNN with 12 convolutional layers, residual connections, dropout, and L2 normalization, trained using the Adam optimizer and evaluated with 5-fold cross-validation.

Main Results:

• The CNN model achieved a diagnostic accuracy of 72.5% (SD ± 8.4) in classifying patients with pCR versus non-pCR.
• Sensitivity was 65.5% (SD ± 28.1) and specificity was 78.9% (SD ± 15.2).
• The area under the ROC curve (AUC) was 0.72 (SD ± 0.08).

Conclusions:

• It is feasible to apply a CNN algorithm to predict NAC response in breast cancer patients using a multi-institution dataset.
• The developed CNN shows potential as a tool for predicting treatment response, aiding in personalized medicine approaches.
• Further validation and refinement of the CNN model may enhance its clinical utility in oncology.