Monitoring Parkinson's Disease from phonation improvement by Log Likelihood Ratios

• Published in: 2015 4th International Work Conference on Bioinspired Intelligence (IWOBI) pp. 105 - 110
• Main Authors: Gomez-Vilda, Pedro, Alvarez-Marquina, Agustin, Rodellar-Biarge, Victoria, Nieto-Lluis, Victor, Martinez-Olalla, Rafael, Vicente-Torcal, M. Carmen, Lazaro-Carrascosa, Carlos
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.06.2015
• Subjects: Aging and Well Being; Biomechanics; Biomedical monitoring; Diseases; e-Health; Monitoring; Monitoring Neurodegeneration; Neuromotor Diseases; Sociology; Speech; Speech Processing; Statistics
• ISBN: 1467378453; 9781467378451
• DOI: 10.1109/IWOBI.2015.7160152

Parkinson's Disease (PD), contrary to other neurodegenerative diseases, supports certain treatments which can improve patients' conditions or at least mitigate disease effects. Treatments, either pharmacological, surgical or rehabilitative need longitudinal monitoring of patients to assess the progression or regression of thier condition, to optimize resources and benefits. As it is well known, PD leaves important marks in phonation, thus correlates obtained from spoken recordings taken at periodic intervals may be used in longitudinal monitoring of PD. The most preferred correlates are mel-cepstral coefficients, distortion features (jitter, shimmer, HNR, PPE, etc.), tremor indicators, or biomechanical coefficients. Feature templates estimated from each periodic evaluation have to be compared to establish potential progression or regression. The present work is devoted to propose a comparison framework based on Log Likelihood Ratios. This methodology shows to be very sensitive and allows a three-band based comparison: pre-treatment status vs post-treatment status in reference to a control subject or to a control population. Results from a database of eight male patients recorded weekly during a month are shown with comments regarding their severity condition. The conclusions derived show that several distortion, biomechanical and tremor features are quite relevant in monitoring PD phonation.