Quantitative Gait Analysis Using a Pose-Estimation Algorithm with a Single 2D-Video of Parkinson's Disease Patients

Area of Science:

• Biomedical Engineering
• Neurology
• Movement Disorders

Background:

• Clinician-based gait assessments for Parkinson's disease (PD) lack objectivity.
• Sensor-based gait analysis is often inaccessible for widespread clinical use.

Purpose of the Study:

• To develop an accessible and objective tool for evaluating gait in PD patients using 2D video analysis.
• To validate the accuracy of video-based gait parameter measurements against a gold standard sensor system.

Main Methods:

• Prospective recording of 2D videos of PD patients and healthy controls performing the Timed Up and Go (TUG) test.
• Utilizing a deep-learning pose-estimation algorithm to track toe and heel movements and calculate gait parameters.
• Simultaneous gait evaluation using a pressure-sensor system (GAITRite) for validation.

Main Results:

• Video-derived gait parameters showed excellent agreement with GAITRite measurements (ICC > 0.9).
• Key gait parameters (step length, velocity, turning time) significantly correlated with PD severity and freezing of gait scores.
• Objective improvements in gait parameters were observed with antiparkinsonian medication.

Conclusions:

• 2D video-based gait tracking offers an objective and accessible method for assessing Parkinson's disease.
• This technology has the potential to improve gait evaluation and treatment monitoring in PD.