An Innovative Deep Learning Framework Leveraging Transfer Learning for Breast Cancer Classification

Abstract

One of the most prevalent medical conditions along with one of the main causes of fatalities remains breast cancer among women worldwide. Early detection and accurate classification are vital to improving results of treatment and rates of patient survival. In this study, we propose a new deep learning framework that employs an attention-based fusion of multiple convolutional neural network (CNN) architectures, specifically DenseNet121, InceptionV3, and EfficientNetB7, for the diagnosis and categorization of breast cancer in histopathological biopsy images. Experimental results in benchmark breast cancer datasets that show how well our suggested model performs, attaining accuracy of 99. 83%, precision is 99.73%, recall of 99.7 3 %, and an F1 score is 99.7 3 %. The proposed model outperforms the individual models with an accuracy of DenseNet121 (98. 76%), InceptionV3 (98. 47%) and EfficientNetB7 (98. 82%) highlighting the effectiveness of our fusion strategy.