Artificial Intelligence-Assisted motion capture for medical applications: a comparative study between markerless and passive marker motion capture

We aimed to determine whether artificial intelligence (AI)-assisted markerless motion capture software is useful in the clinical medicine and rehabilitation fields. Currently, it is unclear whether the AI-assisted markerless method can be applied to individuals with lower limb dysfunction, such as those using an ankle foot orthosis or a crutch. However, as many patients with lower limb paralysis and foot orthosis users lose metatarsophalangeal (MP) joint flexion during the stance phase, it is necessary to estimate the accuracy of foot recognition under fixed MP joint motion. The hip, knee, and ankle joint angles during treadmill walking were determined using OpenPose (a markerless method) and the conventional passive marker motion capture method; the results from both methods were compared. We also examined whether an ankle foot orthosis and a crutch could influence the recognition ability of OpenPose. The hip and knee joint data obtained by the passive marker method (MAC3D), OpenPose, and manual video analysis using Kinovea software showed significant correlation. Compared with the ankle joint data obtained by OpenPose and Kinovea, which were strongly correlated, those obtained by MAC3D presented a weaker correlation. OpenPose can be an adequate substitute for conventional passive marker motion capture for both normal gait and abnormal gait with an orthosis or a crutch. Furthermore, OpenPose is applicable to patients with impaired MP joint motion. The use of OpenPose can reduce the complexity and cost associated with conventional passive marker motion capture without compromising recognition accuracy.