Predicting parkinson's disease using gait analysis
Fwu, Joey Hui Ning
Date of Issue2014
School of Computer Engineering
Centre for Computational Intelligence
Parkinson’s Disease (PD) is characterised by the progressive degeneration of dopamine producing cells in the pars compacta of the stantia nigra. Dopamine is a neurotransmitter that allows communication between the substantia nigra and corpus striatum, the part of the brain which coordinates proper muscle movement. Decline in dopaminergic neurons results in a fall of dopamine concentration which in turn causes the communication between substantia nigra and corpus striatum to be hindered. This causes PD patients to lose some control of their body muscle, and gives rise to early stage symptoms like involuntary shaking, rigidity, postural instability and vast decrease in speed of motion. The continuous loss of these dopamine producing neurons will cause these symptoms to become worse. Unfortunately, there is still no known cure for PD. However, medication can be used to control patient’s symptoms. Therefore, accurate and early diagnosis is of outmost importance. To date, there is also no specific medical test that can be used to diagnose PD definitely. Approximately 10 million people are estimated to be living with PD worldwide, with no discrimination of race and culture, and only 6.3 million has been diagnosed. Hence, there is a need for us to develop methods that can help diagnose PD accurately to allow patients to seek treatment early. Gait is the pattern of movement of the limbs of animals during locomotion, and it is often used by neurologist in the diagnosis of PD. PD patients often exhibit an asymmetric walk, with slow and shuffling steps accompanied by a forward bending posture. For this purpose, gait analysis approach is used for the prediction of PD using machine learning approaches. Seven machine learning algorithms, Naïve Bayes, Bayes Network, Decicion Tree, Bagging, Random Forest, Support Vector Machine (SVM) and Projection Based Learning for Meta-cognitive Radial Basis Function Network (PBL-McRBFN) were used in this study. Gait analysis data obtained from Physionet is used in this project for the prediction of PD. Wavelet transform is used to select the top ten attributes with highest relevance and least redundancy. The performances of these machine learning algorithms were evaluated using these 10 gait attributes. In all these data set, the results of these seven machine learning approach are compared using the three performance measures: overall classification accuracy, average per-class classification accuracy, and F-measure. The result shows that machine learning approaches on gait analysis are competent in diagnosing PD accurately.
Final Year Project (FYP)
Nanyang Technological University