Dinh Huy G, Zhou Joanne Y, Benmira Adam, Kenney Deborah E, Ladd Amy L
Department of Orthopaedic Surgery, Stanford University, Stanford, CA 94305, USA.
Department of Orthopaedics, Emory University, Atlanta, GA 30322, USA.
Sensors (Basel). 2025 Jul 26;25(15):4633. doi: 10.3390/s25154633.
Motion analysis can be useful for multiplanar analysis of hand kinematics. The carpometacarpal (CMC) joint has been traditionally difficult to capture with surface-based motion analysis but is the most commonly arthritic joint of the hand and is of particular clinical interest. Traditional 3D motion capture of the CMC joint using multiple cameras and reflective markers and manual goniometer measurement has been challenging to integrate into clinical workflow. We therefore propose a markerless single-camera artificial intelligence (AI)-assisted motion capture method to provide real-time estimation of clinically relevant parameters. Our study enrolled five healthy subjects, two male and three female. Fourteen clinical parameters were extracted from thumb interphalangeal (IP), metacarpal phalangeal (MP), and CMC joint motions using manual goniometry and live motion capture with the Google AI MediaPipe Hands landmarker model. Motion capture measurements were assessed for accuracy, precision, and correlation with manual goniometry. Motion capture demonstrated sufficient accuracy in 11 and precision in all 14 parameters, with mean error of -2.13 ± 2.81° (95% confidence interval [CI]: -5.31, 1.05). Strong agreement was observed between both modalities across all subjects, with a combined Pearson correlation coefficient of 0.97 ( < 0.001) and an intraclass correlation coefficient of 0.97 ( < 0.001). The results suggest AI-assisted live motion capture can be an accurate and practical thumb assessment tool, particularly in virtual patient encounters, for enhanced range of motion (ROM) analysis.
运动分析对于手部运动学的多平面分析可能是有用的。腕掌（CMC）关节传统上难以通过基于表面的运动分析来捕捉，但它是手部最常见的患关节炎的关节，具有特殊的临床意义。使用多个摄像头和反光标记对CMC关节进行传统的三维运动捕捉以及手动量角器测量，一直难以整合到临床工作流程中。因此，我们提出一种无标记单摄像头人工智能（AI）辅助的运动捕捉方法，以提供临床相关参数的实时估计。我们的研究招募了五名健康受试者，两名男性和三名女性。使用手动量角法以及通过谷歌AI MediaPipe Hands地标模型进行实时运动捕捉，从拇指指间（IP）、掌指（MP）和CMC关节运动中提取了14个临床参数。对运动捕捉测量的准确性、精密度以及与手动量角法的相关性进行了评估。运动捕捉在11个参数中显示出足够的准确性，在所有14个参数中显示出精密度，平均误差为-2.13±2.81°（95%置信区间[CI]：-5.31，1.05）。在所有受试者中，两种方式之间观察到高度一致性，联合皮尔逊相关系数为0.97（<0.001），组内相关系数为0.97（<0.001）。结果表明，AI辅助的实时运动捕捉可以成为一种准确且实用的拇指评估工具，特别是在虚拟患者会诊中，用于增强运动范围（ROM）分析。
