Abstract
Category:
Basic Sciences/Biologics; Other
Introduction/Purpose:
Gait analysis is essential for diagnosing foot and ankle disorders. However, conventional gait analysis methods are costly and complex, especially when evaluating large populations or requiring swift diagnostics. Our study focuses on an innovative application of deep learning (DL) technique to simplify gait analysis. We propose a markerless motion capture system utilizing a single-camera setup, bypassing infrared markers and multi-camera. This novel method aims to be cost-effective and adaptable, enhancing diagnostic efficiency. It is designed to accurately capture gait patterns, offering a comprehensive view of patient-specific conditions when combined with other diagnostic tools such as weight-bearing radiographs and computed tomography. This study evaluates the reliability of the DL-assisted markerless motion capture technique for clinical gait analysis, potentially revolutionizing the evaluation of foot and ankle pathologies.
Methods:
From March 2023 to February 2024, 18 patients who underwent ankle arthrodesis or subtalar arthrodesis for degenerative arthritis were analyzed using the Remobody-S device (Remo Inc., South Korea) integrated with the HuMoR deep learning architecture. This equipment interprets 3D skeletal coordinates, joint range of motion (ROM), and gait metrics from 2D video input. We focused on the hip, knee, and ankle ROMs during the gait cycle. Patients walked 4 meters preoperatively and six months postoperatively, with their movements captured by an attached 2D camera. Differences in gait metrics and joint ROMs between the affected and unaffected sides were compared using the intraclass correlation coefficient (ICC). Clinical outcomes were evaluated using the visual analogue scale (VAS) and Lower extremity function score (LEFS).
Results:
Results indicated that the significant preoperative differences in joint ROMs became comparable postoperatively (Figure 1). The ICCs for hip, knee, and ankle ROMs approached 1 postoperatively, denoting reduced discrepancies between both sides (Table 1). The mean VAS improved from 6.67 ± 1.23 preoperatively to 3.67 ± 0.55 postoperatively (p = 0.003). The mean LEFS improved from 36.3 ± 12.9 preoperatively to 53.0 ± 17.4 postoperatively (p = 0.251).
Conclusion:
The DL-assisted markerless motion capture technique for gait analysis reliably depicted postoperative gait improvement, consistent with clinical outcomes. Its potential clinical application warrants further research.
