Brain2Qwerty v2 Achieves 61% Word Accuracy in Non-Invasive Brain Communication

Brain2Qwerty v2 has achieved a word accuracy rate of 61%, significantly higher than the 8% accuracy of other non-invasive methods. Developed as part of a collaboration with the Basque Center on Cognition, Brain, and Language (BCBL), this model utilizes approximately 22,000 sentences for training and involved nine volunteer participants, each contributing around 10 hours of recorded brain activity. The best participant reached a word accuracy rate of 78%. This follows the introduction of Brain2Qwerty v1 last year, which laid the groundwork for this advancement. The project is supported by the Digital Brain Project, which has allocated $5 million for open datasets to further enhance research in this area. The model leverages tools like NeuralSet for processing brain data at scale and NeuralBench for model evaluation. The implications of this research could be significant for individuals with brain lesions that hinder communication, as it presents a non-invasive alternative to traditional methods like stereotactic electroencephalography and electrocorticography.