Related Experiment Video
Updated: Aug 5, 2026
06:52
Deep-Learning Based Multi-Joint Synchronous Tracking for Objective Quantification of Hindlimb Locomotor Kinematics in Rats
Published on: April 3, 2026
Spatio-temporal graph convolutional networks with transfer learning for continuous ground reaction force estimation
Qinghua Meng1, Zhiyuan Yang1, Yijia Xue1
1Tianjin University of Sport, Tianjin, China.
Frontiers in Bioengineering and Biotechnology
|July 30, 2026
Summary
This study introduces a novel marker-based method to estimate ground reaction forces (GRF) in hemiplegic gait without force plates. The approach accurately captures gait dynamics and asymmetry for rehabilitation assessment.
Area of Science:
- Biomechanics
- Rehabilitation Engineering
- Medical Technology
Background:
- Three-dimensional ground reaction force (GRF) is crucial for assessing weight-bearing, propulsion, and asymmetry in hemiplegic gait.
- Current GRF measurement methods using force plates are confined to laboratory settings, hindering continuous rehabilitation monitoring.
- Estimating continuous 3D GRF in hemiplegic gait without force-plate input, while maintaining kinematic coordination and asymmetry, presents a methodological challenge.
Purpose of the Study:
- To develop and validate a force-plate-independent, marker-based method for estimating continuous stance-phase 3D GRF in post-stroke hemiplegic gait.
- To integrate spatio-temporal graph convolutional network (ST-GCN) with transfer learning for accurate GRF prediction.
- To preserve whole-body kinematic coordination and detect affected-unaffected side asymmetry in hemiplegic gait.
Main Methods:
- A spatio-temporal graph convolutional network (ST-GCN) model was developed, utilizing 39 Plug-in Gait markers as graph nodes with kinematic features.
- Two-stage transfer learning was employed, pretraining the model on healthy gait data and fine-tuning it on hemiplegic gait data.
- Leave-one-subject-out cross-validation was used to evaluate the model's performance on 30 chronic stroke patients.
Main Results:
- The proposed ST-GCN model achieved high accuracy in predicting vertical (r=0.984), anterior-posterior (r=0.956), and medio-lateral (r=0.912) GRF.
- The model demonstrated superior performance compared to baseline methods including MLP, 2D-CNN, BiLSTM, and lightweight Transformer.
- Bland-Altman analysis confirmed minimal bias in key kinetic parameters like peak forces and asymmetry indices, with all participants within 95% limits of agreement.
Conclusions:
- The developed force-plate-independent framework accurately reconstructs GRF waveform morphology and preserves kinetic asymmetry in hemiplegic gait.
- This marker-based method offers a promising laboratory approach for kinetic gait assessment in hemiplegia.
- The system is suitable for laboratory-based kinetic gait assessment but not yet for wearable or home-based clinical monitoring.
