Master’s thesis presentation by Cecilia Hernqvist and Matilda Rosander, MPBME

Welcome to the master’s thesis presentation:”Machine learning for symptoms quantification of Parkinson’s disease patients” by Cecilia Hernqvist and Matilda Rosander, MPBME

When? 13:00, Thursday, 15 June, 2017

Where? Landahlsrummet (room 7430), Hörsalsvägen 11, 7th floor

Examiner: Tomas McKelvey

Abstract
Parkinson’s disease is a disease affecting the nervous system, with about 20 000 people diagnosed in Sweden. The most significant symptoms influence the motor skills, which may greatly lower the quality of the patients’ life. Adjusting the medication dose is difficult and therefore a tool for simplifying the process is needed. Besides this, diagnostics regarding the severity of symptoms often differ between individual physicians. The purpose of this thesis is to help develop a diagnostics tool by using machine learning to classify the severeness of a symptom called bradykinesia. The data used in the project was gathered from gyroscope and accelerometer sensors, which were attached to both the wrists and ankles while the patient conducted two prescribed movement activities; continuous heel tapping and hand rotation. Twenty different machine learning models were trained and tested for the classification based on the UPDRS scale (Unified Parkinson’s Disease Rating Scale). The two most promising models were the support vector machine together with a cubic and a polynomial kernel, respectively. To train the models, at set of 164 features were extracted from the data, whereas different hypotheses were formed based on subsets of the original feature set. It was found that the use of all available features from both the hand and foot data generally gave the best results. One promising outcome was regarding classification between healthy subjects and patient signals deemed as symptom-free, where several models were able to distinguish between the two classes. This result indicates that machine learning algorithms are able to detect signal characteristics not noticeable to the visual inspection of the physician.