47th Annual Meeting The American Society of Pediatric Neurosurgeons

Introduction: 3D motion analysis is invaluable when assessing patients with movement disorders.This technology however, is expensive and labour intensive, thus not readily available in many centres.Recent developments in artificial intelligence for pose estimation allow for development of automated evaluation algorithms for 3D movement using 2D video data. Objectives:We have created an AI augmented pipeline to evaluate patient extremity function using novel pose estimation algorithms.We seek to validate the accuracy and reliability of this tool.Methods: An open-source pose estimation AI was adapted to interpret smartphone video data using validated scales of gait and upper extremity function (Edinburgh Visual Gait Scale [EVGS] and Modified Mallet Scale [MMS]).Video acquisition was carried out on a consumer grade smartphone.AI analysis was compared to a trained human interpretation for comparison.Results: 50 patients have been recruited with 50 gait and 25 upper extremity videos acquired.For the lower extremity, out of the 17 EVGS parameters, 9 parameters have shown a strong correlation between automated and human interpretation while 8 parameters show a good to moderate correlation.Trunk and angle-related parameters such as flexion and extension of the knees and hips perform better relative to other parameters.In the upper extremities, parameters of abduction, hand to neck, hand to spine, and hand on mouth have strong correlation between automated and human interpretation.Global external rotation of the shoulder is not evaluated well.The sub optimally performing parameters may be improved with increased sample size as well as refinements to the AI algorithm. Conclusion:Preliminary results demonstrate that the AI augmented evaluation provides consistent analysis in most movement parameters.We aim to continue to improve this tool kit.Developments are currently underway to expand the scope and implementation of this technology in clinical settings.