Intra-subject class-incremental deep learning approach for EEG-based imagined speech recognition

Brain-computer interfaces (BCIs) aim to decode brain signals and transform them into commands for device operation. The present study aimed to decode the brain activity during imagined speech. The BCI must identify imagined words within a given vocabulary and thus perform the requested action. A possible scenario when using this approach is the gradual addition of new words to the vocabulary using incremental learning methods. An issue with incremental learning methods is degradation of the decoding capacity of the original model when new classes are added. In this study, a class-incremental neural network method is proposed to increase the vocabulary of imagined speech. The results indicate a stable model that did not degenerate when a new word was integrated. The proposed method allows for the inclusion of newly imagined words without a significant loss of total accuracy for the two datasets.