Design of Dementia Detection Techniques Using Machine Learning

Abstract

Dementia is a neurological disease emerging as a global health issue. Dementia has symptoms in which the brain encounters a condition called Apoptosis in which brain cells stop working properly, leading to cell death. Dementia occurs inside specific brain regions responsible for an individual s thinking, recalling, movement, behavior and mood. The early onset of Dementia occurs before the age of 65, but as the disease advances, the condition worsens. newline newline A significant challenge in the detection of dementia is to achieve an accurate and timely diagnosis. There is no known cure for dementia. However, it requires lab tests or imaging techniques like MRI to detect the disease. Computer-aided algorithms using neuroimaging have made prominent advances to address this challenge. Despite considerable progress in this domain, the detection of dementia is still a challenging research issue due to the lack of a robust and efficient general-purpose algorithm. This study focuses on detecting dementia through emerging technologies of machine and deep learning. newline newline In the present work, an ensemble of univariate and multivariate filter feature selection methods has been proposed using a neuropsychological MMSE dataset. The optimal feature set is further classified using Naïve Bayes, Random Forest and Support Vector Machine. This ensemble approach reveals that SVM performs best among the classifiers used. Using the MRI dataset of OASIS, a hybrid model using machine learning is proposed leveraging the benefits of Discrete Wavelet Transform, Bag of Features and Support Vector Machines for the classification of dementia. To overcome the limitations of machine learning, two deep learning models have been proposed using MRI data of OASIS and ADNI. In the first model, a superpixel-powered autoencoder has been proposed using a histogram of oriented feature extraction. It can be inferred by experimental results that the proposed model shows a remarkable performance compared to the previous studies. In the second model, the fine-