Validation of a video-based pose estimation algorithm for the assessment of balance error scoring system in single limb stance test
Year: 2025
Type: article
Abstract: BACKGROUND: Single Limb Stance Test (SLST) is a reliable and validated test to estimate balance performance. However, assessment of the SLST performance, e.g. by using the Balance Error Scoring System (BESS), can be time-consuming and subjective. To deliver effective balance interventions, a reliable and accessible balance assessment method is imperative. RESEARCH QUESTION: Can video-based pose estimation be effectively utilized to validate the BESS assessment for SLST, as compared to both human observation and marker-based assessment methods? METHODS: 60-second eyes-closed SLST trials were recorded using an iPad camera and a marker-based motion capture system. Mediapipe was applied to estimate the whole-body kinematics from the video recordings. The kinematic data were processed by threshold-based error detection algorithms to calculate the corresponding BESS total and sub-scores. To validate the video-based BESS assessment, the results were compared to human and marker-based motion capture system BESS assessments using repeated measures ANOVA and correlation coefficients (CC). RESULTS: There was no significant difference in the BESS total score between the assessment methods and the correlation between assessment methods was good with CC's ranging from 0.69 to 0.77. However, a significant difference in BESS forefoot and heel lifting sub-scores between the video-based and the human or marker-based assessments was found because Mediapipe failed to capture the detail of the foot motion. SIGNIFICANCE: Video-based pose estimation is a reliable and accessible method to assess SLST performance. It can be used to examine and speed up SLST assessment using the BESS total score. However, future research and development in capturing foot motion is needed.
Source: Gait & Posture
Authors: Yu Yuan Lee, Lena Fennen, Rosemary Dubbeldam
Institution: University of Münster
Language: English
Field: Health Professions
Domain: Health Sciences
Open Access status: hybrid
Estimated APC paid: $3,510
OpenAlex ID: W4410019891