AWFCNET: An Attention-Aware Deep Learning Network with Fusion Classifier for Breast Cancer Classification Using Enhanced Mammograms
Renato R. Maaliw, Mukesh Soni, Manuel P. Delos Santos, Maria Rossana D. de Veluz, Pitz Gerald G. Lagrazon, Mariebeth P. Seño, Devie R. Salvatierra - Bello, Reynaldo V. Danganan
Abstract
Breast cancer remains a significant public health concern and a leading cause of female mortality despite recent advances in healthcare. Experts agree that its early prognosis is a key to survivability. In this research, we proposed a deep learning architecture code-named AWFCNET. It comprised multiple segments of preprocessing techniques (color shifting & image enhancement), feature learning based on ResNeXt-101 convolutional network as a backbone with transfer and attention-aware mechanisms, and fusion classifier composed of three recurrent neural networks. The generalization capability of the pipeline produced 98.10% accuracy on a mammogram dataset using 10-fold cross-validation. Computational benchmarks revealed that it surpassed existing state-of-the-art approaches with provisions of visual interpretability via gradient maps. Thus, our framework could complement physicians’ expertise in rapid and dependable breast cancer diagnoses.