Parkinson’s Disease Tremor Assessment Using Inertial Sensors

• Autor(a) principal: Ferreira, Beatriz dos Reis Lopo
• Data de Publicação: 2024
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Texto Completo: http://hdl.handle.net/10400.6/14664
• topic: Doença de Parkinson; Kurosis; Machine Learning; Pontuação da Updrs; Sensores Inerciais; Tremor

Parkinson’s disease (PD) is a chronic, progressive, and neurodegenerative disorder, predicted to be diagnosed for 12 million people by 2040. One of the cardinal symptoms of this disease is tremor. Tremor is characterized as an involuntary and oscillatory movement of a body part and can be divided into rest tremor, postural tremor, and kinetic tremor. The tremor associated with PD is characterized by a 3-6 Hz, regular, asymmetrical tremor and is commonly a rest and/or postural tremor. Nowadays, PD and tremor are usually evaluated by a trained specialist who assesses the symptoms according to the Unified Parkinson’s Disease Rating Scale (UPDRS). However, due to being subjective and representing only a small sample of how symptoms affect the subject during the day, this method exhibits a high within-subject variability and a low test-retest reliability. Consequently, other methods to evaluate tremor that don’t have the same limitations are being proposed and implemented. These methods rely on the use of inertial sensors, like an accelerometer and a gyroscope, and the computation of data collected using these sensors. In this dissertation, a systematic literature review is presented and a mobile app is proposed for the collection of accelerometer and gyroscope sensor data during the performance of five tests, three of them are based on movements performed for the UPDRS evaluation and two of them intend to recreate activities of daily living. This app also includes three daily questionnaires that contextualize the signals collected. Furthermore, a computation framework for the evaluation of tremor is proposed, including the preprocessing, feature extraction, and data analysis steps. The data analysis step is divided into two tasks, the distinction between people with Parkinson’s disease (PwPD) and healthy controls (HC) and the estimation of UPDRS rest tremor scores. A Bagging tree classifier was implemented for both tasks, achieving a good result only for the distinction between the two groups, with a success rate of 85.3%. In addition, a method based on the kurtosis and a method based on the number of 10-second windows in the signal where the fundamental frequency is in the rest tremor frequency band. These methods obtained success rates of 83.3% and 87.88%, respectively.