Detection and Severity Assessment of Parkinson's Disease Through Analyzing Wearable Sensor Data Using Gramian Angular...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Parkinson's disease (PD) is a prevalent neurodegenerative disorder affecting millions globally, primarily the elderly.
• Current PD diagnosis relies on subjective clinical assessments of motor symptoms like bradykinesia and rigidity.
• There is a need for objective, quantitative methods for early PD detection and severity monitoring.

Purpose of the Study:

• To develop and validate a novel method for diagnosing PD and assessing its severity using gait signals.
• To leverage Gramian Angular Fields (GAFs) and deep Convolutional Neural Networks (CNNs) for enhanced PD detection.
• To explore the potential of short gait signal recordings for accessible PD assessment.

Main Methods:

• Utilized Gramian Angular Fields (GAFs) to transform time-series gait data into image representations.
• Applied deep Convolutional Neural Networks (CNNs) for the classification and severity estimation of PD.
• Collected gait data using pressure sensors embedded in shoe insoles.

Main Results:

• Achieved high diagnostic accuracy for PD with 98.6% accuracy, 99.2% true positive rate, and 98.5% true negative rate.
• Demonstrated strong correlation (R² > 0.8) between gait signals and Hoehn and Yahr/Timed Up and Go (TUG) test scores for severity estimation.
• Showed lower prediction accuracy for UPDRS and UPDRS motor scores (R² < 0.2).
• Effective diagnosis and severity assessment were achieved using short gait signal windows (as little as 10 seconds).

Conclusions:

• The GAFs-CNN model offers a highly accurate and objective method for Parkinson's disease diagnosis and severity assessment.
• Gait analysis using this method shows promise for developing shorter, more accessible tools for PD monitoring.
• Further research could refine the model for improved prediction of specific PD severity scales like UPDRS.