Automated UPDRS Gait Scoring Using Wearable Sensor Fusion and Deep Learning

Area of Science:

• Biomedical Engineering
• Neurology
• Machine Learning

Background:

• Quantitative assessment of Parkinson's disease (PD) is crucial for effective management.
• Current clinical evaluations, like the Unified Parkinson's Disease Rating Scale (UPDRS), rely on subjective manual ratings, leading to time inefficiencies and inter-rater variability.

Purpose of the Study:

• To develop and validate a fully automated framework for UPDRS gait-scoring using wearable sensor data and deep learning.
• To create an objective and efficient method for assessing PD severity through gait analysis.

Main Methods:

• Integration of surface electromyography (EMG) signals and inertial measurement unit (IMU) data into a single deep learning model.
• An end-to-end network with specialized branches (diagnosis, evaluation, balance) and a fusion-detection module to emulate clinical assessments.
• Validation using lower limb EMG and shank-mounted IMUs on 21 PD patients and healthy controls during a walking task.

Main Results:

• The automated system achieved a mean classification accuracy of 92.8% across UPDRS levels 0-2.
• The method demonstrated minimal subject effort and sensor setup requirements.
• Significant reduction in clinician workload compared to traditional UPDRS evaluations was observed.

Conclusions:

• Wearable sensor-driven deep learning offers a rapid and reliable approach for PD gait assessment.
• The proposed framework has the potential for widespread application in both clinical settings and home-based monitoring.
• This technology can enhance the objectivity and efficiency of PD diagnosis, treatment, and rehabilitation.