Meistaravörn við verkfræðideild Þorbjörg Ída Ívarsdóttir
MSc í heilbrigðisverkfræði
Fimmtudaginn 12. janúar kl. 14:00 mun Þorbjörg Ída Ívarsdóttir verja 60 ECTS verkefni sitt til meistaragráðu í heilbrigðisverkfræði „Predicting early-stage Parkinson’s disease using virtual reality and biomedical parameters“. Fyrirlesturinn fer fram í M210 og eru allir velkomnir.
Nemandi: Þorbjörg Ída Ívarsdóttir
Leiðbeinandi: Paolo Gargiulo
Prófdómari: Lotta María Ellingsen
Útdráttur:
Parkinson’s disease (PD) is the second most common neurodegenerative disease in the world today. It is mainly characterized by motor symptoms. Postural control deficiencies commonly affect PD patients. Currently, there is no known cure but treatment options can slow down the course of the disease. Therefore early diagnosis of PD is of great importance in order for treatment to be initiated.
The aim of the study was to evaluate the quantitative response of early-stage Parkinson’s patients during the BioVRSea paradigms as well as assessing the interplay of different biomedical parameters using machine learning technologies to classify healthy individuals and early-stage Parkinson’s patients.
Measurements were performed on 9 early-stage PD patients and 37 age-matched healthy subjects. Both before and after the experiment the participants filled out a Motion Sickness Susceptibility Questionnaire (MSSQ), giving a score on their motion sickness symptoms.
In this study, a BioVRSea system using a combination of different biomedical parameters, Center of Pressure (CoP), electroencephalography (EEG), electromyography (EMG) and heart rate (HR) was used. From all of the biomedical parameters a total of 1885 features were collected. To decrease dimensionality they were reduced to the top 20 features. From those top 20 features, there were 10 features from the EEG, 5 from the EMG and 5 from the CoP. Additionally, two other feature selections were used. One consisted of indexes extracted from the MSSQ, while the other consisted of the indexes and the top 20 features combined. These three feature selections were then used to demonstrate the possibility to classify between the healthy subject and the early-stage Parkinson’s subjects. Machine learning was able to predict whether a subject belonged to the healthy or the PD group with 95.7% accuracy with the top 20 features, 80.4% using the questionnaire and 91.3% by combining the top 20 features and the questionnaire.
These results could pave the way to efficient evaluation of early-stage PD.
