Litcius/Paper detail

Multivariate Dynamic Mode Decomposition for Automatic Imagined Speech Recognition Using Multichannel EEG Signals

Alavala Siva Sankar Reddy, Ram Bilas Pachori

2024IEEE Sensors Letters20 citationsDOI

Abstract

In this letter, the multivariate dynamic mode decomposition (MDMD) is proposed for multivariate pattern analysis across multichannel electroencephalogram (MC-EEG) sensor data for improving decomposition and enhancing the performance of automatic imagined speech recognition (AISR) system. Using the proposed MDMD, the MC-EEG signal is decomposed into dynamic modes, which shows the mutual characteristics across all cross channels. Further, different features, namely, frequency, power, and average absolute amplitude have been derived from each computed dynamic mode. The proposed method has been tested on the publicly available dataset of imagined speech EEG sensor data, comprising four different types of imagined prompts. The MDMD-based classification frameworks using random forest and K-nearest neighbor have been developed and achieved significant accuracy of 88.9 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\pm$</tex-math></inline-formula> 2.44% for long words and 73.93 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\pm$</tex-math></inline-formula> 2.86% for short words imagined speech MC-EEG classes, respectively. The proposed MDMD method is capable of delivering improved AISR accuracy for MC-EEG data, and the developed classification framework can be an efficient brain–computer interface tool for persons with speech disabilities.

Topics & Concepts

ElectroencephalographyComputer scienceSpeech recognitionMultivariate statisticsBrain–computer interfacePattern recognition (psychology)SIGNAL (programming language)Artificial intelligenceRandom forestMode (computer interface)k-nearest neighbors algorithmMachine learningPsychologyProgramming languagePsychiatryOperating systemEEG and Brain-Computer InterfacesBlind Source Separation Techniques
Multivariate Dynamic Mode Decomposition for Automatic Imagined Speech Recognition Using Multichannel EEG Signals | Litcius