Prediction of Response to Chemotherapy of Breast Cancer Tumors based on Deep Learning

Neoadjuvant chemotherapy (NACT) has been defined as a widely treatment approach administered before surgery for women with breast cancer to minimize tumor size and improve outcomes. After NACT, pathological complete response (pCR) indicates the absence of residual tumor in the breast. To enhance the long-term survival outcome and to avoid eventual toxicities by NACT, the prediction of pCR using routine breast imaging is an important step to determine the patient treatment. In this work, we applied deep learning models such as Resnet50 and VGG19 to predict pCR from pretreatment dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) scans. The data was obtained using the public database I-SPY1 TRIAL, which is accessible from The Cancer Imaging Archive (TCIA) and encloses 222 patients with breast cancer disease. The dataset was split into 20% for tests and 80% for training. To improve generalization of the model, we also applied data augmentation methods in the training phase as rotating and flipping. Experimental results obtained showed that Resnet50 model outperforms VGG19 in terms of accuracy; where an accuracy of 92.22% and 90.76% are obtained, respectively with data augmentation and axial orientation.