Deep neural network-based Analysis between LDH and Healthy Individuals

Lumbar Disc Herniation (LDH) is a prevalent musculoskeletal disorder that significantly affects quality of life and imposes a growing burden on public health systems. This study proposes a gait feature extraction framework based on deep neural networks to analyze LDH and healthy individuals' motion patterns using 2D video data. Using the MoveNet model, we extracted keypoint sequences from multiple perspectives and performed comparative analysis of gait trajectories. The results reveal that patients with LDH exhibit different gait deviations compared to healthy controls, particularly on the trajectories of the hips, knees, and ankles during stepping. Quantitative evaluation using SVM classifiers and KDE visualization confirms the effectiveness of using six targeted key points for classification, achieving higher precision than using the full 17 points. These findings provide new information on the characteristics of motor impairment of LDH and demonstrate the potential of vision-based neural network models for auxiliary diagnosis and long-term monitoring.