Detection of Parkinson's Disease Using Wrist Accelerometer Data and Passive Monitoring

Area of Science:

• Biomedical Engineering
• Neurology
• Wearable Technology

Background:

• Parkinson's disease (PD) is a neurodegenerative disorder characterized by movement abnormalities.
• Early detection and continuous monitoring are crucial for managing PD.
• Wearable technologies, particularly wrist-worn accelerometers, offer a promising solution for unobtrusive, long-term patient monitoring.

Purpose of the Study:

• To investigate the efficacy of wrist-worn accelerometry for early detection of Parkinson's disease.
• To compare different feature engineering methods (statistical vs. document-of-words) for PD detection.
• To evaluate the impact of data duration (3 vs. 7 days) on classification performance.

Main Methods:

• Utilized a dataset of one-week wrist-worn accelerometry data from individuals with PD and healthy controls.
• Applied two feature engineering techniques: epoch-based statistical features and the document-of-words method.
• Employed various machine learning classifiers, including SVM, to analyze classification performance across different windowing strategies and data durations.

Main Results:

• Parkinson's disease was detected with an average accuracy of 85% ± 15% using an SVM classifier with combined features from all windowing strategies.
• The document-of-words method significantly outperformed the statistical feature engineering approach.
• Classification performance using three days of data was comparable to that achieved with seven days, indicating reduced monitoring time requirements.

Conclusions:

• Wrist-worn accelerometry is a viable tool for the early detection and severity tracking of Parkinson's disease.
• The document-of-words feature engineering method enhances diagnostic accuracy.
• A shorter data collection period of three days can yield comparable diagnostic performance to seven days, improving patient convenience and data acquisition efficiency.