Cross-frequency integration with multiple brain regions for imagined speech

Abstract

Imagined speech, or covert speech, refers to mentally simulating speech without vocalization, engaging brain regions like the auditory cortex and motor areas. This phenomenon is studied to develop brain-computer interfaces for communication, particularly for individuals with conditions like paralysis. Since imagination arises from distributed brain-wide neural activity rather than a single region. Although auditory and motor areas of brain regions are involved in imagination process. For identification of imagined speech, analysis of all brain regions is still an under-explored area. In the proposed work, a cross-frequency integration technique with k-NN classifier is introduced, where each brain rhythm is combined with others except for identical rhythms, and the temporal features of each rhythm combination are extracted. The combinations of brain regions are evaluated using a subset of channels obtained from the channel wrapper method. Further, PCA feature decorrelation and cross-validation techniques are employed. The proposed method achieved an average classification accuracy of 74.4% ± 0.836 (mean ± STD across 10-fold stratified cross-validation) for 5-class classification task and a maximum classification accuracy of 75.8%. The proposed method achieves higher classification accuracy than previously reported state-of-the-art methods. Among all the combinations, the Theta-Gamma combination of brain rhythms achieved the best performance in frontal midline and bilateral frontal pole of the brain.