Breast Cancer Detection using Thermal Images and Deep Learning

Abstract

Breast cancer is indeed a major cause of concern for women in India. There are several detection techniques currently available for breast cancer diagnosis, like mammography, magnetic resonance imaging, ultrasound, and currently Computerized Thermal Imaging is developing its way to enter this field in conjunction with mammography. Infrared imaging is found to be less harmful as compared to mammography and can give results many years prior to the detection of cancer by mammography. The purpose of the current study is to use Thermography and deep learning combination to provide new insight on how to make better predictions for breast cancer. In this work, the methodology and techniques used are based on a deep Convolutional Neural Network model to predict breast cancer from thermal images. Thermal images are pre-processed, segmented and classified using a deep neural network. The research concludes with the major finding that 95.8% accuracy of prediction is achieved for breast cancer based on the output spectrum using training data of 680 thermograms. The current approach demonstrated a significant improvement over the earlier published accuracy of 93.30% with 50 thermograms. Hence, we can summarize that the proposed deep convolutional Neural Network model is highly effective in the prediction of breast cancer.