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

Area of Science:

• Biomechanics
• Clinical Medicine
• Rehabilitation Engineering

Background:

• Markerless motion capture using AI is emerging in clinical fields.
• Its utility in patients with lower limb dysfunction, including orthosis or crutch users, remains unclear.
• Assessing foot and ankle motion, particularly metatarsophalangeal (MP) joint flexion, is crucial for gait analysis in these populations.

Purpose of the Study:

• To evaluate the clinical utility of AI-assisted markerless motion capture (OpenPose) in gait analysis.
• To compare OpenPose's accuracy against conventional passive marker motion capture (MAC3D) and manual video analysis (Kinovea).
• To determine the influence of ankle foot orthoses and crutches on OpenPose's recognition accuracy.

Main Methods:

• Treadmill walking data for hip, knee, and ankle joint angles were collected.
• Data were analyzed using OpenPose (markerless AI), MAC3D (passive markers), and Kinovea (manual video analysis).
• Comparisons were made between the different methods, assessing the impact of assistive devices.

Main Results:

• OpenPose demonstrated significant correlation with MAC3D and Kinovea for hip and knee joint angles.
• OpenPose and Kinovea showed strong correlation for ankle joint angles, outperforming MAC3D.
• OpenPose accurately captured gait in individuals using orthoses or crutches and those with impaired MP joint motion.

Conclusions:

• AI-assisted markerless motion capture (OpenPose) is a viable alternative to conventional methods for gait analysis.
• It offers comparable accuracy for hip and knee joints and superior accuracy for ankle joints.
• OpenPose simplifies and reduces the cost of motion capture without sacrificing accuracy, benefiting clinical medicine and rehabilitation.