Evaluation of the Feasibility and Efficacy of a Chronic Brain-computer Interface for Speech Rehabilitation in Patients With Locked-in-syndrome (LIS)

Eighteen million people worldwide are affected by speech disorders. Locked-in syndrome (LIS) represents the most extreme form of communication disability resulting from motor impairment. In France, the Association du Locked-In Syndrome (ALIS) reports approximately 500 individuals with LIS, most of whom live at home. The quality of life of people with LIS depends strongly on their ability to communicate, and speech synthesis is the mode of communication restoration most desired by individuals with LIS. Several surveys have shown that improving communication abilities in these individuals leads to a significant improvement in their quality of life as well as that of their caregivers.

Natural speech allows the production of an average of 150 words per minute. Non-invasive communication methods, whether based on residual motor function (eye-blink code) or on brain-machine or brain-computer interfaces (Brain-Computer Interface, BCI) using scalp electroencephalographic (EEG) signals, involve a high cognitive load and have low efficiency (spelling only a few letters per minute). Invasive BCIs for speech rehabilitation aim to overcome the limitations of non-invasive devices (cognitive overload and slow speech rate). The intention to act (the act of speaking) is predicted by an algorithm based on the direct decoding of neuronal activity from the sensorimotor cortex (the area where articulatory muscles are represented).

To date, studies testing speech rehabilitation BCIs in humans remain rare. A subdural electrocorticographic (ECoG) invasive BCI enabled speech decoding (words and sentences from a limited repertoire) for chronic use (2 years). Real-time control of an on-screen cursor allowing spelling of up to 90 letters per minute (equivalent to text messaging) has also been achieved using an intracortical invasive BCI (Utah Array). Very recently, up to 60 words per minute were produced using an intracortical invasive BCI (Utah Array) implanted in the ventral premotor cortex, although with a connector potentially contaminated by acoustic audio feedback. Furthermore, these devices still rely on transcutaneous connectors, which may be sources of infection and prevent routine daily-life use.

In summary, there is currently no fully implantable, wireless invasive "speech BCI" with real-time speech synthesis suitable for long-term home use. The present study will use an intracranial extradural invasive BCI combined with a speech synthesizer, with the aim of developing a communication tool suitable for everyday use. More specifically, the SpeechBCI protocol will propose two complementary BCI approaches in the same subject: a speech BCI (primary objective, BCI_PAROLE device) and a cursor BCI (secondary objective). Both BCIs will use the WIMAGINE intracranial epidural system, enabling ECoG signal acquisition with a very limited risk of infection and brain injury and providing signals that are more stable over time compared with intracortical or subdural ECoG devices that retain transcutaneous connectors. These systems will allow long-term and ecological use, as the WIMAGINE implant is wireless and offers excellent long-term signal stability. The WIMAGINE implant has already been successfully tested for controlling an exoskeleton in a tetraplegic subject (operational for over 6 years) and very recently for controlling walking in real-life conditions via a spinal cord stimulator in a paraplegic individual.

The BCI_PAROLE device will integrate a speech synthesizer providing real-time auditory feedback to the speaker. The BCI-CURSEUR device will allow the subject to control an on-screen cursor to access various communication functionalities (web access, emails, chats, etc.). This will provide a complementary communication solution to real-time speech production.

The hypothesis of this stydy is that the intention to speak (attempted speech) will be decoded by the BCI_PAROLE device in individuals with LIS because, as with limb paralysis, paralysis of articulatory and phonatory muscles does not prevent the corresponding cortical map from producing specific signals. Similarly, the BCI-CURSEUR device will decode the intention to move a cursor and perform actions on a computer screen.

Neuronal electrical signals will be recorded bilaterally from the ventral motor cortex (representation of lips, cheeks, tongue, palate, and larynx-the vocal tract) and from the dorsal part of the motor cortex (larynx and hand, the latter for controlling a two-degree-of-freedom cursor), using a total of 128 electrodes (64 on each cerebral hemisphere). The implant will be optimally positioned over speech motor areas using preoperative functional imaging in order to optimize speech decoding by the BCI_PAROLE device.