Development of a web application for processing Neuroimaging data in the Cloud. Application to Brain Connectivity

• Autor(a) principal: Ferreira, Pedro Miguel Botica
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/22038
• topic: Convolutional Neural Networks; Cloud Computing; Alcohol Use Disorder; Machine Learning; Brain Connectivity; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Médica

Neuropsychiatric disorders, or mental disorders, have long been known to be a major cause of burden to society and it is estimated that one in every four people worldwide will be affected by one of these conditions during their lifetime. The diagnosis of these conditions is based on a set of subjective criteria and on the experience of physicians and is therefore highly prone to error. Alcohol Use Disorder (AUD) is one such disorder with particularly devastating consequences to both individual and society, representing a total of 5.1% of the global burden of disease and injury. As image classification methods improve, reaching near-human capabilities, and research on brain physiology continues to advance and allow us to better understand brain structure and function through novel methods such as Brain Connectivity analysis, ingenious approaches to medical diagnosis can be envisioned. Furthermore, as new technologies allow the world to be more connected and less dependent on physical machinery, there is an interest in bringing this vision to both healthcare and biomedical research, through technologies such as Cloud computing. This work focuses on the creation of an intuitive Cloud-based application which uses the image classification algorithm Convolutional Neural Network (CNN). The application would then be used to classify Electroencephalography data to diagnose AUD, in particular using Brain Connectivity metrics. The created application was successfully developed according to the objectives, proving to be simple to operate but effective in the use of the CNN algorithm. However, due to the environment used, it showed high processing times which hamper the training of CNN classifiers. Classification results, while not conclusive, show indication that the employed metrics and methodology may be of use in the context of neuropsychiatric disorder diagnosis both in a research and clinical context in the future. Finally, discussion and analysis of these results were performed so as to drive forward the research into this methodology.