Hybrid MODWT-CSP Approach for Motor Imagery Classification in MEG-BCI Systems
33rd Conference on Signal Processing and Communications Applications-SIU-Annual, İstanbul, Turkey, 25 - 28 June 2025, (Full Text)
- Publication Type: Conference Paper / Full Text
- Doi Number: 10.1109/siu66497.2025.11111752
- City: İstanbul
- Country: Turkey
- Istanbul University Affiliated: No
Abstract
MEG-BCI systems are effective for motor imagery (MI) classification owing to their high temporal and spatial resolutions. This study proposes the MODWT + CSP method using a 306-channel MEG dataset recorded on different days for four imaging tasks: two-hand (H) and two-foot (F) movement imagery, subtraction (S), and word formation (W). The method was evaluated using four classifiers (SVM, k-NN, ANN, and NB), with ANN achieving the highest accuracy. The experimental results indicated the best classification performance in the H-W (77.41%) and F-W (75.94%) task pairs. However, the performance varied among individuals, with participants 3, 5, 8, and 17 exceeding 90% accuracy in certain tasks. These findings confirm the success of MODWT + CSP + ANN in MEG-BCI systems and emphasize the need for adaptive approaches to address interindividual variability.