Objective Assessment of Navigation Trajectory in Walking Corsi Test Using Pose Estimation Algorithms

This study utilizes pose estimation algorithms to present a novel objective assessment of navigation trajectories in the walking Corsi test, known as a visuospatial working memory assessment task. Spatial navigation is critical in cognitive assessments, especially for detecting impairments in neurodegenerative conditions such as Alzheimer's disease. By leveraging the OpenPose model, this approach tracks specific body keypoints, primarily focusing on the left ankle, to monitor and evaluate participant movement during the walking Corsi test. A Kalman filter was applied to enhance the accuracy of keypoint tracking, refining the model's output by reducing noise and compensating for motion inaccuracies. Data were collected from 35 healthy participants, and the resulting navigation trajectories were subject to perspective correction, keypoint extraction, and noise reduction. The processed trajectories were analyzed using root mean square error, mean absolute error, and R2 metrics to assess model performance at various stages of data processing. These metrics revealed significant improvements in accuracy after implementing pre-processing techniques and the Kalman filter tracker algorithm, with the model achieving a highly accurate R2 value of 0.995. The findings demonstrate the effectiveness of biomechatronic systems in delivering precise spatial navigation and memory assessments. This work holds promise for clinical diagnostics and research applications, providing an innovative, objective tool for assessing cognitive functions, particularly in neurodegenerative diseases.