A knowledge graph-based post-stroke gait assessment system: A pilot study

Abstract

Instrumented gait analysis (IGA) has been widely used in research, but not typically in clinical practice, as it requires expertise in data analysis and interpretation. Post-stroke clinical gait assessment could be improved by integrating artificial intelligence into IGA, but previous gait assessment systems have relatively low clinical utility. This study aims to develop a clinically oriented automatic post-stroke gait assessment system based on knowledge graph (KG) to better support clinicians. A domain KG is first constructed in the field of gait analysis. This system can process IGA data to identify gait abnormalities and their potential causes based on kinematic analysis and KG. A preliminary evaluation with twenty post-stroke patients and four domain experts tested the system's performance in clinical settings, showing an average recall, precision, and F-score of 1, 0.78, and 0.89. Four clinical professionals showed high behavioural intention to use the system in clinical settings (4.33 ± 0.41 on a 5-point Likert scale based on the Technology Acceptance Model). The results depicted that this system has potential to be applied in clinical settings to provide useful supplementary insights for clinicians, which may promote the interpretation and clinical utility of IGA. The schema of this KG could be generalised or extended to gait analysis related to other diseases.