Intelligent Alzheimer's Detector Using Deep Learning

Abstract

Researchers of the era are constantly striving to achieve accurate, precise algorithms incorporated in highly affordable models, trying to assist the medical practitioners in solving complex medical problems(Alzheimer's). Deep Learning is state-of-the-art learning algorithm in classification and exceptionally efficacious in extracting high level features from multi-dimensional data. In this system we use Convolutional Neural Network particularly for classification of fMRI clinical data on stages of Alzheimer's disease brain from Diseased(AD), Early Mild Cognitive Impairment(EMCI) to those who are normal, having healthy brains. Usage of MRI data has already been done[1] for binary classification. We aim to generalize the classifier into categorizing the images into three different distinct classes. The deep learning pipeline involved critical steps of correctly preprocessing 4D fMRI images i.e. 3D images varying with time. The preprocessing steps involved, proved to be critical in having distinct 2D image slices of Normal, EMCI and AD for better accuracy in understanding the most discriminative features of the fMRI brain scans. Transfer Learning is the concept involved wherein we utilize pretrained complex deep models for classification of images. Advantage of this learning over constructing a new convolutional network is that the knowledge gained during training of ImageNet Dataset fastens the learning process in addition to increased accuracy. We have adopted Inception Resnet V2 model and hope to achieve a competitive accuracy. The aim of the project is to create an Alzheimer's Detector ousting the accuracy of modern radiologists so as to reduce the effort and money of consulting a Radiologist.