An Integrated Deep Learning Framework for Automated Breast Cancer Identification

Abstract

Breast cancer is a significant disease burden in the world and corrects and timely classification is essential to improve patient outcome and treatment strategies. Current diagnostic methods are generally subjective, and leave room for subjectivity and delays in treatment. For this purpose, we used the BRACS dataset which had 1000 images of two classes, Benign and Malignant. Applying augmentation techniques, we were able to get 6520 images, resized to 227x227x3. The models used in this work for classification of breast cancer are the VGG19 model with accuracy of 96.7% and the MobileNetV2 model with accuracy of 91%. This is true for VGG19 which had the best performance among all models. The results of the study underscore the value of deep learning and the use of augmented datasets in breast cancer classification to enable accurate detection and decision making in automated diagnostic systems and improve patient care.