MEDICAL IMAGING USING MACHINE LEARNING TECHNIQUES

Abstract

"We present an application of deep Convolutional Neural Networks (CNN) for theclassification of mammogram as normal, benign and malignant tumor. The images used in this study have been used from the mini-MIAS dataset of mammograms. The proposed system has been implemented in three stages: (a) preprocessing of the dataset which includes resizing and rotation of the original mammogram (b) feature extraction by using a CNN autoencoder (c) classification task using the previously extracted features by training of a deep CNN classifier. In this research, the goal of the system is to distinguish between three classes of mammograms: benign, malignant or without tumor. In this thesis, we implemented an approach using self-learned features which are extracted from the utoencoder. These features are provided to a CNN classifier for classification. Our approach uses deep CNN for feature extraction and classification task. This approach is compared with the methods in which hand crafted features and different classifiers are used. Our approach provides better results than these methods. Hence, our approach do not require any feature engineering."