Evaluation of temporal, spatial and spectral filtering in CSP-based methods for decoding pedaling-based motor tasks
Cristian Felipe Blanco-Díaz1,2, Cristian David Guerrero-Mendez1,2, Denis Delisle-Rodriguez3
1Postgraduate Program in Electrical Engineering, Federal University of Espirito Santo (UFES), 29075-910 Vitória, Brazil.
Biomedical Physics & Engineering Express
|February 28, 2024
Summary
This study introduces advanced machine learning methods to accurately decode pedaling movements from EEG signals, improving Brain-Computer Interfaces for stroke rehabilitation. These findings enhance the potential for robotic-assisted therapy and daily living activities.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Rehabilitation Technology
Background:
- Stroke significantly impairs Activities of Daily Living (ADLs) due to loss of motor control.
- Brain-Computer Interfaces (BCIs) combined with robotic systems offer a promising avenue for neurorehabilitation.
- Electroencephalography (EEG)-based BCIs face challenges with artifacts, complicating movement discrimination, especially for pedaling tasks.
Purpose of the Study:
- To propose and evaluate Common Spatial Patterns (CSP)-based methods for classifying pedaling motor tasks using EEG.
- To address the challenge of movement discrimination in EEG-based BCIs for pedaling, a less explored area.
- To enhance the accuracy of decoding motor intentions for rehabilitation applications.
Main Methods:
- Implemented Filter Bank Common Spatial Patterns (FBCSP) and Filter Bank Common Spatial-Spectral Patterns (FBCSSP).
- Utilized different spatial filtering configurations and filter bank combinations.
- Analyzed an in-house EEG dataset from 8 participants performing pedaling tasks.
Main Results:
- The optimal configuration involved a two-filter bank (8-19 Hz and 19-30 Hz) with a 1.5-2.5s time window and two spatial filters.
- Achieved an accuracy of approximately 0.81.
- Reported False Positive Rates below 0.19 and a Kappa index of 0.61.
Conclusions:
- EEG oscillatory patterns during pedaling can be accurately classified using machine learning.
- The proposed FBCSP and FBCSSP methods show significant potential for decoding pedaling tasks.
- This research supports the future application of these methods in rehabilitation contexts, such as Motorized Mini Exercise Bike (MMEB)-based BCIs.
