A CNN–RNN–LSTM Based Amalgamation for Alzheimer’s Disease Detection

Abstract

Alzheimer’s disease is a fatal brain condition that causes irreversible brain damage and gradually depletes memory of an individual. The basic idea of the presented work in this paper is to determine the level of dementia accurately such that the patients can receive medication before their condition worsens. The results produced by existing models like Support Vector Machine (SVM), Convolutional Neural Network (CNN) are not satisfactory enough for large datasets. This paper suggests a solution based on Amalgamation of deep learning models. The Magnetic Resonance Imaging (MRI) scan data from the Open Access Series of Imaging Studies (OASIS) Brain, has been used for experimental purposes. The first phase focuses on implementing the primary models i.e., CNN, Recurrent Neural Networks (RNN), and Long Short Term Memory (LSTM). The second phase implements ensemble technique to combine these three models using a weighted average approach. An approach called Bagging is applied in all the three models to decrease the variance and the three Bagged models are combined using the ensemble method. The results show that the ensemble of CNN, RNN, and LSTM achieved an accuracy of 89.75% whereas the ensemble after Bagging the primary models achieved an accuracy of 92.22%. On comparison, the ensemble of Bagged models achieved better results than ensemble of primary models. The comparison is done on the basis of accuracy metric along with sensitivity, specificity and precision.