Automatic Prognosis System Using Recurrent Neural Network for Prediction of Alzheimer Disease

Abstract

Alzheimer’s disease (AD) is a type of brain malfunction mainly found in old-aged people; it is irreversible and develops over time. Usually, people with AD experience initial memory loss, ultimately leading to their declined cognitive abilities like poor decision making and problem-solving abilities. However, there is no specific technique available which can predict the early detection of AD. The neuro-imaging technology that was recently adopted for the prediction of AD is based on few biomarkers, thus it is time taking. Therefore, it is essential to bring about an automated technique for AD prediction. The early prediction of AD can be pivotal in the survival of patients and will prove helpful for the doctors. In this work, we propose an automated prognosis system based on Machine Learning (ML) techniques for the prediction of AD. In the proposed model, the Neuro Psychological Measures and Magnetic Resonance Imaging (MRI) biomarkers (feature vector) are computed and passed to Recurrent Neural Network (RNN). In RNN, we use Long Short-Term Memory (LSTM). This model will predict the biomarkers (feature vector) of patients after 6, 12, 18, 24 & 36 months and will give the predicted values after these time intervals. These predicted biomarkers will go through fully connected neural network layers. The NN layers will then predict that these RNN predicted biomarkers belong to an AD patient or a Mild Cognitive Impairment (MCI) patient. If RNN predicted biomarkers belong to AD, then the patient is converted from MCI to AD. The developed methodology has been tested on a publicly available data set ADNI and achieved an accuracy of 88.24% which is better than the other state of the art techniques.