The Hand Pose Estimation Model in Measuring Range of Motion

Abstract

Objective. The COVID pandemic has challenged medical practitioners to perform clinical examinations remotely, including assessing the range of motion of the finger joints. This sparked the development of the 3D (three-dimensional) Hand Pose Estimation Model, a software that can generate hand pose estimates and compute hand joint angles from a 2D (two-dimensional) image. The study aims to assess the accuracy of the 3D Hand Pose Estimation Model when compared with a goniometer and radiography. Methodology. The 3D Hand Pose Estimation Model was developed by training a machine learning model with a parametric hand model and 2D hand images. Ten healthy participants with no history of trauma, disease, or deformity of the hand were enrolled in the study. Active flexion and extension joint angles of the metacarpophalangeal, proximal interphalangeal, and distal interphalangeal joints of the fingers, excluding the thumb, were measured using the 3D Hand Pose Estimation Model, a goniometer, and radiographs. Results. The mean joint angles derived from the 3D Hand Pose Estimation Model and goniometer were not significantly different in 18 out of 24 joint angles (75%). While measurements from both instruments differed greatly from those taken on radiographs, more goniometric measurements are within five degrees of the radiographic measurements. Conclusion. The 3D Hand Pose Estimation Model can estimate joint angles given a 2D image. Improvements in the model can be made with the aid of the data obtained from this study.

Keywords

• hand
• range of motion
• estimation model
• machine learning