MRI Brain Disease Detection using Enhanced Landmark based Deep Feature Learning

For brain diseases like Alzheimer's disease (AD) detection, several studies have been proposed in the last two decades by various biomarkers, but the accurate detection and retrieval is still a research problem. To overcome the challenges of state-of-art techniques of automatic brain disease detection using MRI images, novel Enhanced Landmark based Deep Feature Learning (ELDFL) proposed in this paper consist of phases like pre-processing, landmark estimation, the hybrid features learning and normalization, classification and retrieval. The raw brain MRI image is first pre-processed using noise removing filters and contrast enhancement functions followed by the landmark estimation. In the hybrid features learning phase, we automatically extract patch-based features from landmarks followed by the texture features to form the hybrid deep learning-based features. The min-max normalization technique is applied to extracted features to enhance the performances. The proposed hybrid feature, learning-based Convolution Neural Network (CNN) generates the trained model of training MRI datasets which is further used by the Long-Short Term Memory (LSTM) classifier to classify the input test MRI brain image either into AD class or Normal Control (NC) class followed by the retrieval of relevant images. The experimental results show the proposed model improved the performance of image retrieval and classification compared to the state of art method.