Related Experiment Video
Updated: Jan 9, 2026
07:24
Home-Based Monitor for Gait and Activity Analysis
Published on: August 8, 2019
7.2K
Self-supervised Contrastive Learning to Monitor Free-Body Movement Daily Activities of Parkinson's Disease Patients
Summary
This study introduces a new self-supervised learning method using a wrist sensor to accurately monitor Parkinson's disease (PD) patients' daily activities, improving deep model generalization for better patient care.
Area of Science:
- Biomedical Engineering
- Machine Learning
- Neurology
Background:
- Parkinson's disease (PD) monitoring requires accurate human activity recognition.
- Wearable sensors offer a less invasive approach to track daily living activities.
- Deep learning models struggle with generalization in PD patients due to abnormal movements.
Purpose of the Study:
- To develop a novel self-supervised contrastive learning method for enhanced deep model generalization in PD patients.
- To accurately recognize activities of daily living (ADL) using a single wrist-mounted accelerometer.
- To improve the reliability of wearable sensor-based monitoring for PD individuals.
Main Methods:
- Utilized a self-supervised contrastive learning methodology.
- Employed a single wrist-mounted accelerometer sensor for data collection.
- Trained deep models on data from 15 Parkinson's disease patients.
Main Results:
- Successfully recognized three unconstrained ADL profiles: Ambulation, Sitting, and Standing.
- Achieved an average weighted F1-score of 84.85%.
- Outperformed recent studies using deep learning and wearable sensors for PD activity monitoring.
Conclusions:
- The proposed method enhances deep model generalization for PD patients.
- This approach enables reliable daily monitoring with a less invasive sensor setup.
- Findings support personalized physical therapy and informed clinical decisions for PD management.
