Brain Computer Interface for Communication and Speech Restoration

UTHealth investigators have developed a speech brain-computer interface (BCI) to decode speech using neural activity, especially for patients with aphasia due to brain damages in language areas. This novel BCI decodes speech by widespread electrode coverage in multiple cortical sites. It achieved 73% phoneme-level correctness during articulation and 87% in the best subject, via transfer learning and a sequence-to-sequence (Seq2Seq) model.

Background

More than one million people in the US alone have been rendered aphasic, which is a condition that affects a person’s ability to communicate. Aphasia may be due to damage to language areas by stroke, traumatic brain injury, neoplasia, or degenerative diseases. Recent advancements in BCI have demonstrated the potential to decode speech using neural activity. However, traditional speech BCI relies heavily on the detection of direct signals from the motor cortex of the brain, proving insufficient for patients with damaged language cortices.

Significance and Impact

UTHealth researchers Drs. Nitin Tandon and John Seymour developed a novel BCI with capability to detect speech intention for patients who do not have a normal, preserved language cortex either by direct injury or disconnection. The novel BCI deviates from traditional BCIs that are narrowly focused on detection of signals from the motor cortex of an intact brain. The novel BCI addresses the unmet need for patients with brain damages in language areas to communicate more effectively.

Technology Highlights

Customizability for individual brains using transfer learning techniques with relevant training data sets and continuously trainable convolutional layer.
Potential to be neural prosthetics for aphasic patients with lesions and insufficient fluency of word production to initialize models otherwise, bypassing limitations in number of collected neural signals posed by conventional speech BCIs.
Potential to benefit a wide range of patients, including those with damaged language cortices or those with locked-in syndrome.
The Seq2Seq model benefits from temporal context, language structure, and sequence modeling, which performed significantly better than other models in predicting phonemes across all patients.

Related Publication:

Tessy M Thomas et al. J. Neural Eng. (2023) 20 046030
Singh Aditya, et al. Nature Communications 16.1 (2025): 8749

Intellectual Property Status

US Utility Patent App. 19/298,712

Available for licensing

About the Inventors