Behavior recognition and assessment of spinal dysfunction based on an attention mechanism

Behavior recognition and assessment of spinal dysfunction are crucial for accurately diagnosing complex motor patterns, postural deviations, and movement abnormalities. Traditional approaches often rely on handcrafted features and static analysis, which struggle to capture the dynamic and context-dependent nature of human motion. We develop an attention-augmented deep neural framework designed to advance behavioral data analytics and facilitate early recognition of spinal dysfunctions.

The proposed framework consists of two core components: the Spinal Dysfunction Attention Recognition Network (SDARN) and the Adaptive Attention-Based Strategy (AABS). SDARN integrates advanced feature extraction, attention-driven temporal modeling, and classification modules to effectively learn spatial and temporal dependencies in movement sequences. This design enables the network to focus on relevant motion segments that correspond to abnormal spinal patterns. Meanwhile, AABS dynamically refines the attention distribution by incorporating domain-specific constraints and temporal regularization, enhancing both robustness and interpretability. Together, these components form a unified system capable of adaptive learning from complex behavioral data.

Experimental evaluations conducted on benchmark datasets confirm that the proposed method achieves notable gains in recognition accuracy and interpretability compared to conventional models, providing a promising tool for clinical assessment and rehabilitation of spinal dysfunctions.