Modely stanovení vývoje Parkinsonovy choroby z řečového signálu

Abstract

This diploma thesis deals with software model development, which is capable of mapping sixteen input elements of speech signal oscillation measurements to clinical outputs. Samples of speech signal were obtained by measurements in collaboration with patients suffering from Parkinson's disease. There are provided 5875 measurement samples. The system's output determines severity of Parkinson's disease progression, qualified by clinical diagnostic rating scale UPDRS. Designed system must be able to generalize in order to correctly assess UPDRS scale value from future speech signal measurements. Utilized methods in this thesis are artificial neural networks, particularly modifications of Error Backpropagation algorithm, along with Adaptive Neuro Fuzzy Interference System (ANFIS). Calculations are parallelized as much as possible, so that many model simulations are performed in order to find the optimal solution. In case of success, it would be possible to develop an embedded system, which could continuously diagnose Parkinson's disease progression from home. That would reduce patient's travelling expenses to medical centres.