Abnormal Gait Recognition of Patients with Stroke Based on Deep Learning Fusion
10.16156/j.1004-7220.2025.04.022
- VernacularTitle:基于深度学习融合的卒中患者异常步态识别
- Author:
Chenhao LI
1
;
Peng YANG
;
Chenglong FENG
;
Haifeng ZHANG
;
Chenghua JIANG
;
Wenxin NIU
Author Information
1. 上海工程技术大学机械与汽车工程学院,上海 201620;同济大学附属养志康复医院(上海阳光康复中心)转化研究中心,上海 201619
- Publication Type:Journal Article
- Keywords:
gait recognition;
stroke;
bidirectional long short-term memory network;
convolutional neural network;
residual network
- From:
Journal of Medical Biomechanics
2025;40(4):955-962
- CountryChina
- Language:Chinese
-
Abstract:
Objective To address the personalized differences in motion gait between stroke patients and healthy older adults,as well as the issue of abnormal gait recognition,a deep learning fusion-based approach is proposed to effectively improve the accuracy of abnormal gait recognition.Methods A model fusing convolutional neural networks(CNN)and bidirectional long short-term memory networks(BiLSTM)was adopted,with the introduction of a residual network(ResNet).Unilateral ankle joint movement data at different walking speeds within a comfortable range were collected from healthy older adults and stroke patients.Signals from inertial sensors and electromyography sensors were used as inputs,while gait features were analyzed and gait differences between the two groups were compared.The effectiveness of the model was validated by comparing the classification performance of traditional deep learning models and CNN-ResNet-BiLSTM models with different layer combinations in terms of abnormal gait recognition accuracy.Results The CNN-ResNet-BiLSTM model,which introduced residual connectivity,performed excellently in abnormal gait recognition.Compared with traditional deep learning models such as the gated recurrent unit(GRU)and long short-term memory network(LSTM),its prediction accuracy was improved by 13.6％and 8.36％,respectively.Additionally,compared with other model combinations,this model achieved an overall accuracy of 97.78％.Conclusions The algorithm proposed in this study can be applied to stroke-related abnormal gait detection,providing technique support for the early diagnosis and precise monitoring of such diseases.
