Project Title
Team
- e19142, Hashini Illangarathne, email
- e19249, Sayumi Muthukumarana, email
- e19452, Ashan Wimalasiri, email
Supervisors
Table of content
- Abstract
- Related works
- Methodology
- Experiment Setup and Implementation
- Results and Analysis
- Conclusion
- Publications
- Links
Abstract
Brain Computer Interfaces(BCIs) have made a huge impact in neurotechnology by enabling direct communication between the human brain and electronic external systems. Motor Imagery (MI) based BCIs, which utilize electroencephalography(EEG) signals to interpret motor imagery tasks have demonstrated significant potential in assistive technologies and neurorehabilitation. Traditional MI classification approaches typically distinguish between broad movements. However, novel research suggests that targeting single-joint MI EEG classification (Eg:Wrist, Elbow, Knee) can enhance precision, providing more intuitive and natural control for BCI applications. This study focuses on a single-joint, specifically the wrist, based MI task classification. The introduction of a novel model capable of distinguishing distinct wrist movements and investigating model performance across distinct joint movements, is the main aim of this study. Thus, contributing to improved precision in BCI applications.
Related works
Traditional MI classification research so far focuses on distinguishing between broad movement categories, such as left-hand and right-hand imagery.
- [Lotte et al.] employed CSP-based feature extraction for limb-level MI tasks but did not address finer, joint-specific movements.
- [Fang et al.] utilized deep CNNs for hand-movement classification but did not distinguish between different joints like the wrist or elbow.
Methodology
The primary objective is to classify MI tasks for single joint movements, focusing on the wrist, with high accuracy to enhance the usability of Brain-Computer Interface(BCI) applications in neurorehabilitation and assistive technology.
EEG Data
EEG signals of 25 healthy subjects performing wrist movements, pronation and supination, are obtained through the publicly available dataset on GigaDB(https://gigadb.org/dataset/100788).
Data Preprocessing
Data Preprocessing is required to be carried out in order to develop models based on traditional machine learning. Techniques proposed to be utilized would include,
- 60 Hz notch filter will be applied to raw EEG data to reduce the effect of external electrical noises such as DC noise of power supply and the scan rate of the monitor.
- Band-pass filter will be applied to remove both low-frequency and high-frequency components that are irrelevant to the Motor Imagery (MI) task.
- Independent Component Analysis(ICA) technique will be applied to remove artifacts such as eye-blinking effects, muscle movement, and other non-neural interferences.
Feature Extraction
- Common Spatial Patterns(CSP) to maximize variance between different MI tasks.
- Wavelet Transform for multi-resolution analysis.
Classification Approaches
Traditional Machine Learning Models include,
- Support Vector Machines (SVM)
- Linear Discriminant Analysis (LDA)
- Random Forest (RF)
Deep Learning Models include,
- Recurrent Neural Networks (RNN)
- Transformers
Model Evaluation
The model’s performance will be assessed using standard classification metrics, including accuracy, precision, recall, F1-score, and Matthews Correlation Coefficient (MCC). Additionally, Area Under the ROC Curve (AUC-ROC) will provide insights into the model’s discriminative capabilities. Statistical significance tests, such as paired t-tests, will be conducted to validate performance improvements.
Experiment Setup and Implementation
The higher overview of steps involved in classifying MI tasks is shown below.
Results and Analysis
Results have not yet been finalized.
Conclusion
This proposed research approach focuses on classifying wrist Motor Imagery(MI) tasks using EEG signals, specifically pronation and supination movements. The approach consists of preprocessing, feature extraction, model training, and evaluation to enhance classification accuracy. This study aims to improve MI-based BCI systems by addressing challenges in classifying single-joint movements, contributing to advancements in assistive technologies and neurorehabilitation.
Publications
This project has not yet been published.