EEG motor imagery classification using machine learning techniques

• R.T. Páez-Amaro ^[1] ; E. Moreno-Barbosa ^[1] ; J.M. Hernández-López ^[1] ; C.H. Zepeda-Fernández ^[1] ; L.F. Rebolledo-Herrera ^[1] ; B. de Celis-Alonso ^[1]
  1. [1] Benemérita Universidad Autónoma de Puebla

     Benemérita Universidad Autónoma de Puebla

     México

     
• Localización: Revista Mexicana de Física, ISSN-e 0035-001X, Vol. 68, Nº. 4, 2022, págs. 669-680
• Idioma: inglés
• Enlaces
  • Texto completo
• Resumen

  • Background. A brain-machine interface (BMI) is a device or experimental setup that receives a brain signal, classifies it and then uses it as a computer command. There is not a consensus on which kind of learning methodology (deep learning, convolutional networks, AI, etc.) and/or type of algorithms in each methodology, are best to run BMI’s. Objective. The aim of this work was to build a low-cost, portable, easy-to-use and a reliable Motor Imagery Electro-encephalography (EEG-MI) based BMI; comparing different algorithms to find the one that best satisfies such conditions. Methods. In this study, motor imagery (MI) EEG signals, from both PhysioNet public data and our own laboratory data obtained using an Emotiv headset, were classified with four machine learning algorithms. These algorithms were: Common spatial patterns (CSP) combined with linear discriminant analysis (LDA), Deep neural network (DNN), convolutional neural network (CNN) and finally Riemannian minimum distance to mean (RMDM). Results. The mean accuracy for each method was 78%, 66%, 60% and 80% respectively. The best results were obtained for the baseline vs Motor Imagery (MI) comparison. With global-training public data, an accuracy between 86.4% and 99.9% was achieved. With global-training lab data, the accuracy was above 99% for the CSP and RMDM cases. For lab data, the classification/prediction computing time per event were 8.3 ms, 18.1 ms, 62 ms and 9.9 ms, respectively. In thediscussion a comparison between the results presented here and state-of-the-art of methodologies and algorithms for BMI’s can be found. Conclusions. The CSP and RMDM algorithms resulted in fast (computing time) and effective (success rate) tools for their implementation as deep learning algorithms in BMIs.