Mammogram image classification system using deep learning for breast cancer diagnosis

Abstract

Mammography is the common screening method of breast cancer, a deadly disease among women in the world with a high mortality rate. Breast cancer is the uncontrollable cell growth as a tumor that may be cancerous or non-cancerous. This study employs a Deep Convolution Neural Network (DCNN) with Transfer Learning (TL) that utilizes mammogram image samples for breast cancer diagnosis. The contrast enhancement of the suspicious image tumor is done by using the Contrast Limited Adaptive Histogram Equalization (CLAHE) technique on Region of Interest (ROI). The VGG-16 network model is utilized with reduced convolution layers and max-pooling layers providing more feature datasets for efficient mammogram classification into either benign or malignant cancer for early diagnosis for its appropriate treatment. The performance of the VGG16 model is compared with the VGG-19 net. Results show that VGG-16 architecture provides promising results than VGG-19 on Mammographic Image Analysis Society (MIAS) database images with 82.5% accuracy.