From Theory to Practice: Moving Toward Artificial Intelligence-powered Computer Vision Applications for Peripheral Motor Nerve Assessment of the Hand.

Abstract

Computer vision has emerged as a useful technology that may prove capable of facilitating remote clinical examinations in hand surgery. This study's primary aim is to evaluate the efficacy of computer vision for assessing peripheral motor function and range of motion of the hand for future clinic and telemedicine purposes. Five healthy volunteer subjects (10 hands total) were filmed performing three static hand examinations ("peace sign," "hitchhiker thumb," and "OK sign") as well as apposition. Videos were processed using the proprietary H.AI.ND program based on the MediaPipe API (Google, v0.9.2.1), generating temporal and spatial data for joint angle analysis. The median joint angles determined for each test were compared with their manually derived counterparts to assess accuracy and reliability. The measurements were compared at a population level using Wilcoxon signed rank tests and at the individual video level using interclass correlation analyses. The artificial intelligence-generated angle outputs demonstrated a high level of reliability when compared with manually determined measurements for the 3 clinical positions included in this study. Assessment of compound appositional movement also demonstrated high reliability with time-dependent multijoint evaluation. Goniometric analysis through computer vision applications may provide an easy and reliable alternative for hand evaluation in the normal population for both static and dynamic function. Further study is warranted to evaluate this program's potential role for diagnostic assessment in the diseased population before and after surgical investigation.