Shared-task Self-supervised Learning for Estimating Free Movement Unified Parkinson's Disease Rating Scale III

Area of Science:

• Biomedical Engineering
• Neurology
• Machine Learning

Background:

• The Unified Parkinson's Disease Rating Scale (UP-DRS) is crucial for diagnosing and assessing Parkinson's disease (PD) severity.
• Current clinical assessments require in-person examinations, posing challenges for continuous monitoring.
• Wearable sensors and machine learning offer a promising avenue for remote PD severity estimation.

Purpose of the Study:

• To develop and validate a machine learning model for estimating UPDRS Part III scores using wearable sensor data.
• To investigate the efficacy of a multi-channel convolutional neural network and self-supervised learning for PD severity assessment.
• To provide a reliable method for at-home estimation of PD severity, supporting clinical decision-making.

Main Methods:

• Utilized motion data, including gyroscope signals and their spectrograms, from two wearable sensors.
• Developed a multi-channel convolutional neural network (CNN) for UPDRS Part III estimation.
• Implemented a novel shared-task self-supervised learning approach to enhance estimation accuracy during free-body movements.

Main Results:

• The proposed model achieved an improved correlation coefficient (0.81) compared to baseline (0.67) with clinical UPDRS-III examinations.
• The mean absolute error in UPDRS-III estimation was reduced from 7.75 to 6.96.
• The system demonstrated reliable PD severity score estimation during daily activities using data from 24 PD subjects.

Conclusions:

• The developed approach shows significant potential for accurate, remote monitoring of Parkinson's disease severity.
• This technology can empower physicians with continuous patient data for better disease management and medication adjustment.
• The findings support the integration of wearable technology and AI for more effective Parkinson's disease care.