CNN-based Deep Transfer Learning Approach for Detecting Breast Cancer in Mammogram Images

Abstract

Researchers are motivated to investigate deep learning (DL) techniques for mammogram images because of the limitations of conventional systems-based computer-aided detection (CAD) for mammography, the extreme significance of early breast cancer identification, and the high impact of false diagnosis on patients. This study develops a model for breast cancer detection from mammogram images that employ Convolutional Neural Network (CNN) based Transfer Learning (TL). The developed structure is comprised of several stages: breast region extraction is performed to extract the Region of Interest (ROI) from the background and artifacts, a Gaussian filter is employed for noise reduction, and data augmentation is carried out to increase the size of the original images for better learning of CNN. Then, a pre-trained model-based CNN using an augmented dataset is presented. Deep features are extracted from CNN and they are trained based on using TL. A mini-MIAS dataset was conducted as the testing ground for this experiment, and the best accuracy achieved was 95.71%. The developed framework performs significantly better than other current methods when compared to them.