Cross-speed and Cross-posture Gait Phase Partitioning Based on Multi-modal Data.

Gait abnormality represents a widespread motor symptom that can stem from various diseases, including stroke, Parkinson's disease, and spinal cord injury. During the rehabilitation process of gait abnormality, recognizing the gait phase of patients is crucial. However, patients usually cannot realize walking at a stable speed or in a normal body posture. Therefore, the partitioning of gait into distinct phases under different circumstances is imperative. In this study, generalized models for cross-speed and cross-posture gait phase partitioning was developed using multimodal gait data collected at different speeds and postures. Multiple classifiers were employed to construct the cross-condition models, and the evaluation was systematically performed using the F1 score metric for different modalities, classifiers and conditions under both intra-subject and inter-subject scenarios. The results showed that the RF classifier outperformed others (91.58% F1 score), and the EMG modality yielded the best performance across different modalities (89.15% F1 score). Moreover, the inter-subject model yielded more favorable outcomes compared to the intra-subject model. This systematic evaluation and analysis contribute to the development of generalized gait phase partitioning models that can be applied across different speeds and postures in patients with gait abnormality, thereby facilitating their rehabilitation process.