A Gait Symmetry-Fused Informer Model for Predicting Asymmetric Lower Limb Joint Angles in Knee Osteoarthritis Patients

Knee osteoarthritis (KOA), a prevalent musculoskeletal disorder characterized by asymmetric gait, significantly impairs walking ability. While lower limb exoskeleton systems offer promise for gait correction, precise joint angle prediction remains challenging: the asymmetric gait of KOA patients creates poor inter-limb similarity in joint angle data, making it difficult for traditional models to discern correlations and effectively learn from such data. This deficiency is particularly pronounced when predicting highly asymmetric knee joints, where low data similarity hinders feature extraction and modeling accuracy. To address this, the study collected lower limb joint angle data from five KOA patients during asymmetric gait using inertial measurement units, integrating gait symmetry features into an enhanced Informer model for multi-joint prediction. Experimental results showed the model achieved a 200 ms prediction with a mean absolute error (MAE) of 0.95°for knees-reducing MAE by 60% compared to the CNN-BiLSTM baseline-and 0.88°-1.34°for relatively symmetric hips. These results validate the model's capability to handle varying joint symmetry by leveraging gait symmetry information, supporting its use in optimizing exoskeleton systems for real-time gait modulation and personalized KOA rehabilitation.