Wearable-Sensor-Based Weakly Supervised Parkinson's Disease Assessment with Data Augmentation.

Area of Science:

• Biomedical Engineering
• Neurology
• Data Science

Background:

• Parkinson's disease (PD) is a leading cause of dementia, necessitating accurate monitoring tools.
• Wearable technology offers potential for computer-aided diagnosis and long-term monitoring of PD.
• Assessing PD severity efficiently and accurately in free-living environments remains a challenge due to poor annotation and class imbalance.

Purpose of the Study:

• To develop a novel framework for assessing Parkinson's disease severity using wearable sensor data in real-world settings.
• To address challenges of poor annotation and class imbalance in wearable-based PD assessment.
• To enable more accurate self-diagnosis and remote intervention guidance for PD patients.

Main Methods:

• Utilized clustering methods to learn latent categories from activities.
• Employed latent Dirichlet allocation (LDA) topic models to capture multi-activity latent features.
• Augmented bag-level data while retaining key instance prototypes to mitigate class imbalance.

Main Results:

• Collected a dataset from 83 individuals in free-living conditions using wearable sensors.
• Achieved 73.48% accuracy in fine-grained classification of PD severity (normal, mild, moderate, severe) based on hand movements.
• Demonstrated the framework's efficacy in real-world conditions.

Conclusions:

• The proposed framework effectively assesses Parkinson's disease severity using wearable sensor data in free-living environments.
• This approach can improve the accuracy of PD self-diagnosis and facilitate remote drug intervention guidance.
• The study highlights the potential of advanced data analysis techniques for managing neurodegenerative diseases.