PhaseMix: A Periodic Motion Fusion Method for Adult Spinal Deformity Classification

Abstract

Human gait plays a crucial role in medical diagnostics, particularly in the context of adult spinal deformities (ASD). Recent research has applied deep-learning techniques to process video data for ASD diagnosis. Although this method successfully captured dynamic postural information, it does not account for the periodicity and symmetry of motion, particularly the periodicity of gait and the inherent symmetry of locomotor postures in the gait cycle. This omission may have led to the loss of critical information necessary for an accurate diagnosis. To resolve this issue, we present a method that can effectively capture cycle and action symmetry information in motion and exploit the characteristics of gait motion. The proposed method is subsequently trained using the fused data within the deep learning model. Our experiments, performed on a video dataset consisting of 81 patients, showed that our method outperformed baseline approaches. The proposed method achieved an accuracy of 71.43, precision of 72.80, and F1 score of 71.15. The complete set of codes, models, and results can be accessed from the GitHub repository: https://github.com/ChenKaiXuSan/Skeleton_ASD_PyTorch. In this study, we present an innovative video-based approach to support clinical diagnoses of gait disorders, with a focus on periodic motion and postural symmetry. The application of deep learning in diagnostic methods has been enhanced by utilizing the kinematic properties of walking movements. The experimental results demonstrate that the proposed method can provide more accurate diagnostic results when using limited video data.

Keywords

• Adult spinal deformity
• deep learning
• human action recognition
• healthcare
• gait posture
• periodic motion