BR-Mix3DNet: Predicting Breast Neoadjuvant Chemotherapy Response With a Hybrid MLP-Mixer and 3D CNN Model

Neoadjuvant chemotherapy (NAC) is the standard treatment for locally advanced and inflammatory breast cancer, as well as early-stage triple-negative breast cancer (TNBC) and human epidermal growth factor receptor 2 (HER2)-overexpressing breast cancer. The primary objective of NAC is to reduce the tumor size, enabling breast-conserving surgery. However, patient responses to NAC vary widely, making accurate prediction of responsiveness crucial to avoid ineffective treatments and associated toxicities. This study evaluated the diagnostic performance of a deep learning hybrid model in predicting NAC responsiveness in patients with TNBC using pretreatment MRI data. Two independent datasets were used: the CHONNAM dataset for training and the DUKE dataset for validation and testing. Preprocessing steps included region of interest extraction, N4 bias field correction, intensity normalization, and ensuring MRI shape uniformity. The BR-Mix3DNet model, which integrates MLP Mixer layers with 3D convolutional neural network (CNN) layers, was used for feature extraction from MRI images. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC), Youden Index, sensitivity, and specificity. The BR-Mix3DNet demonstrated high diagnostic accuracy, achieving a mean AUROC of 0.89 and a Youden Index of 0.68, outperforming other hybrid models and standalone 3D CNN models. The integration of MLP Mixer and 3D CNN layers significantly enhanced predictive performance, offering promise for improving clinical decision-making and optimizing treatment strategies for patients with TNBC.