Development and Performance Analysis of a Semi-Supervised Gait Recognition Model for Pediatric Abnormalities Using a...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Pediatric Orthopedics

Background:

• Pediatric gait abnormalities are linked to musculoskeletal problems and increased injury risk.
• Early detection and accessible screening are crucial for timely intervention.
• Current assessment methods may have limitations in accessibility and scalability.

Purpose of the Study:

• To develop and validate a video-based semi-supervised Abnormal Gait Recognition Module (AGRM) for pediatric gait assessment.
• To improve diagnostic performance and clinical interpretability of gait abnormality detection.
• To address the need for accessible and accurate pediatric gait screening tools.

Main Methods:

• Developed a 3D ResNet backbone integrated with a Mean Teacher Module (MTM) and Spatial Hierarchical Pooling Module (SHPM).
• Trained and validated the AGRM on a hybrid dataset of pediatric gait videos and the CASIA-B dataset.
• Evaluated performance using accuracy, macro-precision, macro-recall, and macro-F1 scores for binary and three-class classification tasks.

Main Results:

• The AGRM achieved 70.5% accuracy and 0.718 macro-F1 score in three-class classification (normal, genu varum, genu valgum).
• In binary classification (normal vs. abnormal), the model yielded 80.3% precision and 79.2% recall.
• Ablation studies confirmed the significant contributions of MTM and SHPM; Grad-CAM visualization highlighted attention to lower extremity and knee regions.

Conclusions:

• The developed AGRM demonstrates robust performance and generalization in identifying pediatric gait abnormalities.
• The video-based AI approach effectively captures pathological gait characteristics and offers a promising tool for early screening.
• This technology can enhance accessible pediatric musculoskeletal assessment in clinical and community settings.