Analiza automată a semnalelor fiziologice pentru diagnosticarea și monitorizarea bolilor neurologice / Automatic analysis of physiological signals for the diagnosis and monitoring of neurological disorders

Neurological diseases place a great burden on the worldwide healthcare system. This thesis proposes and explores several methods for the automatic analysis of physiological recordings with the aim of improving the monitoring and diagnosis of neurological disorders. The focus of this work is on the characterization of epilepsy and of several neurodegenerative disease, out of which the most prevalent, Parkinson’s and Alzheimer’s disease. Epilepsy and it's clinical symptom, epileptic seizures are typically diagnosed through the manual analysis of hours of electroencephalographic (EEG) recordings. Here, three methods of extracting features out of the EEG signal are proposed with the end goal of providing an automated identification of seizure segments. Unsupervised feature extraction showed the most promise in terms of interpatient model performance variability. Sleep disorders are commonly encountered in multiple neurodegenerative diseases and the automatic analysis of sleep stages could help in their characterization. Here, proposals for improving state of the art sleep scoring are explored. Using a single channel EEG as input showed a significantly high performance in detecting sleep stages, while factor analysis enhanced the detection performance using deep neural networks. A method for detecting a gait disturbance in Parkinson’s disease is proposed using a deep network model on raw accelerometer data. Automatic identification of Alzheimer’s patients from healthy participants by applying transcranial magnetic stimulation while simultaneously recording EEG is explored. The method proposed not only classifies patients with high performance, but also provides insights on the effect of the disease on the brain. The results obtained with all the proposed methods show improvements or strong alternatives to the state of the art automatic analysis applied to epilepsy and neurodegenerative disorders.