Machine learning methods for multiple sclerosis detection based on raw data from an instrumented walkway

Hu, Wenting (2022) Machine learning methods for multiple sclerosis detection based on raw data from an instrumented walkway. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (2MB)

Abstract

Background: Multiple sclerosis (MS) patients may experience varying gait disorders. In MS disease evaluation and diagnosis, instrumented walkways using embedded pressure sensors are widely used to provide information regarding gait disturbances. The information is delivered as predefined parameters, which may obscure salient features and patterns in the raw sensor data. This thesis applied machine learning techniques to raw walkway data to distinguish MS patients from healthy controls while further distinguishing the impairment levels of MS patients. Methods: New features were constructed to supplement the standard parameters. A severity level was determined using patients' ratings of the severity of their gait problems on the MS Impact Scale-29. Two experiments were conducted. The first experiment focused on discerning healthy controls from MS patients. The second experiment attempted to classify patients with different impairment levels. Results: The MS vs. Healthy experiment achieved a good baseline accuracy of 81% using the standard feature set and received a 7% improvement using the augmented set. The mild MS vs. moderate MS experiment achieved an accuracy of 76% using the standard set, which was further improved by 2% using the augmented set. Conclusion: These experiments demonstrate that the newly generated features improve the machine learning model results with excellent accuracy and precision.

Actions (login required)

View Item