Related Experiment Video
Updated: Aug 10, 2026
11:25
Simultaneous Scalp Electroencephalography (EEG), Electromyography (EMG), and Whole-body Segmental Inertial Recording for Multi-modal Neural Decoding
Published on: July 26, 2013
Integrating statistical parametric mapping, functional principal component analysis, explainable machine learning,
Kofi Nyantakyi Appiah1, Edward Wilson Ansah2, Frank Sarfo Dofour3
1Department of Physical Education, Wesley College of Education, Kumasi, Ghana. nyantakyi.appiah@gmail.com.
BMC Biomedical Engineering
|August 8, 2026
Summary
Increasing walking speed altered muscle activation patterns but not ground reaction forces in healthy adults. This study introduces a multimodal gait analysis framework for future clinical applications.
Area of Science:
- Biomechanics and Motor Control
- Human Movement Analysis
- Wearable Sensing Technologies
Background:
- Gait speed is a critical indicator in neurological and orthopedic conditions.
- Waveform-level adaptations in ground reaction forces (GRF) and electromyography (EMG) during gait are not well understood.
- Existing analyses often lack integration of continuous data, dimensionality reduction, explainable AI, and equivalence testing.
Purpose of the Study:
- To compare three-axis GRF and multi-muscle EMG during slow (0.5 m/s) versus fast (1.0 m/s) treadmill walking.
- To utilize statistical parametric mapping (SPM), functional principal component analysis (fPCA), explainable machine learning, and equivalence testing for multimodal gait analysis.
- To investigate waveform-level adaptations in GRF and EMG signals.
Main Methods:
- Analysis of 58 healthy adults (55 with complete EMG data).
- Paired SPM with cluster-based permutation for waveform difference assessment.
- fPCA-derived features input into a Random Forest model with SHAP interpretability; Two One-Sided Tests (TOST) for vertical GRF equivalence.
Main Results:
- No significant cluster-level differences were found in any GRF component between speeds.
- Significant EMG clusters were identified in multiple lower limb muscles, including tibialis anterior, gastrocnemius, vastus lateralis, rectus femoris, and semitendinosus.
- The Random Forest model achieved 87.2% accuracy, with tibialis anterior fPCA features being the most significant predictor; TOST did not confirm vertical GRF equivalence.
Conclusions:
- Moderate increases in walking speed induce widespread changes in multi-muscle activation but not significant GRF waveform alterations.
- The integrated SPM-fPCA-SHAP-TOST pipeline offers an interpretable framework for multimodal gait analysis.
- This framework can be a foundation for future research in rehabilitation engineering, digital biomarkers, and wearable sensing.