Neurological Disorders Detection using Longitudinal Neuroimaging A One Class Model Approach

Abstract

The brain is the most important organ in the human body, and many neurological disorders go undiagnosed until they become highly dangerous. Neurological disorders are diseases associated with the central nervous system. These diseases have an increased mortality rate of 16.8% annually, and these statistics suggest that neurological disorders are more likely to cause more disability than other human disorders. The research addresses problems of two prominent neurodevelopmental (Autism Spectrum Disorder(ASD) and Attention Deficit Hyperactivity Disorder (ADHD)) and neurodegenerative diseases(Alzheimer s Disease (AD) and Parkinson s Disease (PD)) diagnosis from longitudinal structural Magnetic Resonance Imaging (sMRI) brain scans. Neuroimaging is an essential tool used for both diagnosis and treatment. However, interpreting neuroimages and making diagnosis or treatment recommendations require specially trained medical specialists. Reading neuroimages is currently a labour-intensive, time-consuming, expensive and error-prone process. It would be more desirable to have a computer-aided system that can automatically make diagnosis and treatment recommendations. Recent advances in Machine Learning (ML) have great potential to significantly enhance the neuroimaging pipeline, supporting clinical decision-making and Computer-Aided Diagnosis (CAD). Early detection has proven critical to give patients the best chance of recovery and survival. Advanced CAD systems are expected to have high sensitivities and low error rates. The thesis initially presents baseline results on detecting neurological disorders using longitudinal sMRI samples on the ML framework. It proposes approaches to generate synthetic medical images using Generative Adversarial Networks (GANs). Building a reliable automated disease diagnosis model is challenging when working with a small dataset and few annotated samples. GAN models are proposed to solve such issues and generate brain sMRI images. Methods to improve the performance of 2D Convolutional Neural..