Predicting Post Neoadjuvant Axillary Response Using a Novel Convolutional Neural Network Algorithm.

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Conventional radiographic response assessment after neoadjuvant chemotherapy (NAC) is insufficient for predicting axillary pathologic complete response (pCR).
• Accurate prediction of axillary pCR is crucial for guiding surgical management and improving patient outcomes in breast cancer treatment.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) algorithm for predicting axillary pCR using pre-treatment breast MRI data.
• To assess the performance of the CNN model in distinguishing between patients who achieve and do not achieve axillary pCR post-NAC.

Main Methods:

• Retrospective analysis of 127 breast cancer patients who completed NAC and underwent surgery with pathology data.
• Development of a 10-layer CNN model utilizing 3D segmentation of tumors from pre-NAC T1 postcontrast MRI images.
• Implementation of dropout, augmentation, and L2 regularization to prevent model overfitting.

Main Results:

• The CNN algorithm achieved an overall accuracy of 83% (95% CI ±5) in predicting axillary pCR.
• The model demonstrated high sensitivity (93%, 95% CI ±6) and specificity (77%, 95% CI ±4) for detecting residual axillary disease.
• The area under the ROC curve was 0.93 (95% CI ±0.04), indicating strong predictive performance.

Conclusions:

• CNN architecture is a feasible and effective tool for predicting post-NAC axillary pCR in breast cancer patients.
• The developed AI model shows promise in improving the accuracy of axillary response assessment compared to conventional methods.
• Further validation with larger datasets is recommended to enhance the predictive accuracy of the CNN model.