Brain implant allows ALS patient to speak with 99% accuracy

What happened?

A team from the University of California, Davis, has published a study in Nature Medicine in which a patient with amyotrophic lateral sclerosis (ALS), Casey Harrell, used a brain implant to communicate for over 3,800 hours over two years, producing nearly 2 million words. The system, called BrainGate, achieves 99% accuracy and an average speed of 56 words per minute, allowing the patient to work full-time without constant researcher assistance. This milestone represents the longest and most autonomous use of a brain-computer interface (BCI) for speech, surpassing previous records in duration and accuracy.

Why is it important?

This breakthrough is significant because it addresses two key barriers to BCI adoption: the need for daily calibration and reliance on intensive technical supervision. In previous studies, patients required frequent system adjustments to maintain accuracy, limiting practical use. In this case, Harrell's implant has worked stably for months without recalibration, thanks to machine learning algorithms that adapt to long-term neural changes. The speed of 56 words per minute, though lower than natural speech (~150 wpm), doubles the speed of earlier systems like the 2021 one (50-word vocabulary) and approaches functional communication rates. The 99% accuracy is comparable to commercial speech recognition systems, reducing user frustration. Additionally, the patient has been able to resume his work as a technology consultant, demonstrating that BCI can be integrated into daily life and professional activity.

Consequences and context

The success of the implant opens the door for patients with severe paralysis to regain communication and work capacity, which has profound economic and social implications. According to data from the ALS Association, approximately 30,000 Americans live with ALS, and many lose the ability to speak in advanced stages. Technologies like BrainGate could restore not only their voice but also independence and participation in society. However, significant challenges remain: the device requires invasive surgery to implant electrodes in the motor cortex, carrying risks of infection and high costs (estimated in tens of thousands of dollars). Moreover, the system is not commercially available; it has only been tested in a handful of patients in clinical trials. Compared to the first speech implant in 2021, which handled only 50 words, this system uses a vocabulary of over 125,000 words, representing a qualitative leap in expressive capacity. It also outperforms non-invasive systems like Synchron's (FDA-approved in 2021), which has lower accuracy and speed. The study highlights the importance of adaptive algorithms: the machine learning model continuously updates to track changes in neural signals, enabling prolonged use without performance degradation. This approach could be applied to other BCIs to restore movement or vision.

What readers should know

The BrainGate technology, originally developed at Brown University, uses a 100-channel electrode array implanted in the motor cortex to record neural activity related to the intention to speak. Signals are decoded using deep learning algorithms that predict phonemes and words, generating synthesized text displayed on a screen or played through artificial voice. In Harrell's case, the system has been integrated with his personal computer, allowing him to write emails, participate in meetings, and browse the internet. While 99% accuracy is impressive, the system still makes errors, especially with homophones (e.g., "wave" vs. "waive") and in noisy environments. Researchers plan to improve speed by adding predictive language models and emotional intonation, as well as reducing hardware size for portability. They are also exploring wireless versions to eliminate cables connecting the implant to an external computer. For interested readers, this study marks a turning point: BCI for speech has moved from proof-of-concept to a functional and sustainable tool. However, commercialization is still far off; more trials with diverse patients and regulatory approval are needed. In the future, similar technologies could help not only people with ALS but also those with spinal cord injuries, strokes, or cerebral palsy. Casey Harrell's case demonstrates that yesterday's science fiction is taking firm steps toward clinical reality.