Master thesis presentation: Symptoms Quantication for Parkinson’s Disease by Marieke Wendebourg
On Friday 10 June, Marieke Wendebourg is presenting her master thesis with the title “Symptoms Quantication for Parkinson’s Disease”.
When? Friday, 10 June, 2016 at 10:00 am.
Where? Lunnerummet 3311, Chalmers
Examinator: Tomas McKelvey
Today, Parkinson’s disease is the second most common age related degenerative disorder presenting a complex set of both cognitive and motor symptoms. Medication for the treatment of motor symptoms exists but the development of effective treatment plans without technical aids is tedious. These aids could include sensor systems for the objective evaluation and quanti cation of symptoms in short- and long-term settings for both the clinical and the home environment. Recently, several studies have shown the feasibility of symptom quanti cation with the help of gyroscopes or accelerometers. Utilizing such measurements, this work aims at a comparison of several supervised learning methods in order to find the most suitable model structure and therefore the best modeling approach for the quantification of bradykinesia in Parkinson’s patients using the example of repeated forearm-rotation, which is a routine motion from Parkinson’s test protocols.
The measurement characteristics applied for model development were based on knowledge about the considered movement and motion patterns in Parkinson’s disease as well as on insights provided by the literature on other studies concerning the quanti cation of Parkinson’s symptoms. The considered parametric and non-parametric models were developed for a number of sensor subsets and compared in terms of cross-validated mean squared prediction errors obtained for data not utilized during model development. As expected, it was found that when considering only gyroscopes, those measurements of angular velocities around the axis of the forearm were most relevant to model development.
Additionally, results generally improved when using principal component analysis for dimension reduction prior to model development. The best results were obtained for local regression when applied to only two characteristics of measurements of angular velocities around the forearm.