Pathological gait reconstruction and recognition framework for cerebral palsy based on wearable body sensor networks

Document Type

Article

Publication Date

3-1-2026

Abstract

Abnormal gait stands as a key clinical symptom in cerebral palsy. A thorough analysis of the pathological gait exhibited by cerebral palsy (CP), coupled with an interpretation of the associated kinematic mechanisms, holds immense importance for their rehabilitation therapy. This paper proposes a framework for three-dimensional reconstruction and recognition of pathological gait for CP based on wearable body sensor networks. In our approach, a body sensor network (BSN) which can sense the kinematic signals is constructed for the data collection of CP gait. After then, an error-state-Kalman filter (ESKF) algorithm is proposed for multi-sensor data fusion to estimate the attitude of each body segment, and a lower limb kinematic model was created to achieve three-dimensional representation of pathological gait, the kinematic characteristics of different types of pathological gaits is also analyzed through statistical analysis. In addition, a parallel dual-branch optimized multi-modal fusion backbone network is established to extract multi-dimensional features from sensor time series and skeletons, providing classification results for different types of pathological gaits.Finally, we conducted multiple experiments, including K-fold cross validation, leave-one-subject-out cross (LOSO) validation and leave-multiple-subject-out (LMSO) validation, to analyze the performance of the proposed model in pathological gait recognition tasks. Furthermore, we assessed the impact of the number of sensors on pathological gait recognition performance to optimize the sensor configuration. The experimental results demonstrate that our proposed data fusion method can precisely capture the kinematic changes in the lower limbs of individuals with CP, and the proposed pathological gait recognition model exhibits superior performance compared to the current state-of-the-art methods.

Keywords

Body sensor network, Cerebral palsy, Motion capture, Multi-sensor data fusion, Pathological gait analysis

Publication Title

Information Fusion

ISSN

1566-2535

DOI

10.1016/j.inffus.2025.103915

Funders

Nanjing Medical University (2024NJMUWLXK05, CMCM202305, NMUR20240010), Natural Science) Research in Higher Education Institutions in Jiangsu Province (25KJA416001), National Natural Science Foundation of China (62203150, 62276090), National Key Research and Development Program of China (2022YFC2405600), Cooperation Projects in Jiangsu Province (BZ202406)

Volume

127

Publisher

Elsevier B.V.

Share

COinS