A Novel Approach for Transfer Learning Motor Imagery Classification Based on IVA

• Published in: 2023 31st European Signal Processing Conference (EUSIPCO) pp. 1210 - 1214
• Main Authors: Moraes, Caroline P. A., Fantinato, Denis G., Neves, Aline
• Format: Conference Proceeding
• Language: English
• Published: EURASIP 04.09.2023
• Subjects: Brain-Computer Interface; Correlation; Electroen-cephalogram; Electroencephalography; Independent Vector Analysis; Minimization; Motor Imagery; Signal processing; Signal processing algorithms; Solids; Transfer learning
• DOI: 10.23919/EUSIPCO58844.2023.10289953

Motor imagery (MI) classification based on electroencephalogram (EEG) signals performs an important role in neurological rehabilitation for therapeutic proposes. Independent Component Analysis (ICA) is a set of techniques with a solid framework and is widely used in the signal processing area. Inspired by ICA, Independent Vector Analysis (IVA) is an extension of the problem for multiple datasets and explores the correlation between different datasets through the use of Mutual Information (Mutinf). The statistical dependency between datasets through Mutinf could help in MI classification since it allows a generic and homogeneous treatment of the whole data and a possible knowledge transfer between patients. This paper proposes an innovative approach for the Transfer Learning MI task by exploring the minimization of mutual information through IVA applied to motor imagery. The results show a high correlation and small standard deviation cross-subjects.