Breast Cancer Detection with Transfer Learning Technique in Convolutional Neural Networks

Abstract

Medical imaging plays a major role in disease diagnosis. It helps the physicians to look into the human body which is impossible otherwise. Computer Aided Diagnosis is gaining popularity in the detection of tumors from radiology images. Cancer is one of the most feared diseases as the survivors of the disease are very few. But when treated early one can actually increase the survival rate. In this research work MRI images of breast are used to identify and classify the tumors present in the image as either benign or malignant[1]. Convolutional Neural Network is a deep learning algorithm that is gaining lot of popularity in computer vision problems. They tend to produce promising and accurate results for image classifications as the one targeted here.  As medical data are very sparse and difficult to obtain it is not possible to create a neural network and train from scratch as it requires huge amount of data. Transfer learning technique is used to handle this issue. Three pre trained models namely, VGG19, ResNet50 and Xception are employed here for the classification process and their results were studied individually. It proved that transfer learning improves the performance of the models as compared to the existing systems. And among the three pre-trained models VGG19 and ResNet50 came up with 98% accuracy and Xception showed a slightly lower value of 96% which also seem to be a pretty good performance when compared with existing systems that do not use transfer learning techniques.