Why is communicating so hard with ALS?
ALS is a disease that gradually destroys the nerves that control muscles. The 45-year-old patient in the study had tetraparesis (weakness in all four limbs) and severe dysarthria: his speech muscles obeyed poorly and he was very hard to understand.
Brain-computer interfaces read brain activity and turn it into text. Until now, they needed long training and still made too many mistakes to support a real conversation.
The study is a BrainGate2 clinical trial (NCT00912041). The authors note that earlier brain-computer interfaces were held back by extensive training requirements and limited accuracy, which is what this work set out to improve.
What did the researchers do?
Five years after his illness began, the participant had four tiny microelectrode arrays placed in a part of the brain (the left ventral precentral gyrus) that commands the movements of speech. The implants record activity from 256 electrodes.
When he tries to speak, an AI decodes that signal into words, shown on a screen and then spoken by text-to-speech software set to sound like his voice before ALS.
The team tested decoding in two settings: prompted sentences and unstructured conversation, where the participant says what he actually wants to say. Decoded words appeared on screen and were then spoken aloud by the voice software.
What were the results?
On the first day of use, 25 days after surgery, the system reached 99.6% accuracy with a 50-word vocabulary. Calibration needed 30 minutes of recordings while the participant tried to speak.
On the second day, after 1.4 extra hours of training, accuracy was 90.2% with a 125,000-word vocabulary. With more data it reached 97.5% and held there for 8.4 months. The participant held self-paced conversations at about 32 words per minute for more than 248 cumulative hours.
In other words, the system needed only about half an hour of recordings to start working. Accuracy was highest with the small 50-word vocabulary, dipped to 90.2% when the vocabulary grew to 125,000 words on the second day, and then rose again with more training data.
Key figures
- 99.6%accuracy on the first day, with a 50-word vocabulary
- 90.2%accuracy on the second day with a 125,000-word vocabulary
- 97.5%accuracy sustained over 8.4 months, at about 32 words per minute
What does this change for patients?
For people locked inside their bodies by paralysis, getting back a quick conversation would be a major gain in freedom. This study shows training time can be short and accuracy very high.
It is not yet a treatment people can get. It is a demonstration that such a device can work in one person over several months.
The authors conclude that performance reached a level suitable to restore conversational communication after brief training, in a person with ALS and severe dysarthria.
What are the limits?
Only one patient was followed. We do not know whether results would be the same in other people, with other diseases or other damaged parts of the brain.
The device needs brain surgery and a complex recording setup. Larger and longer trials will be needed to judge safety, cost and ease of everyday use.
The study also reports results for a single device and a single surgical placement, so we cannot yet say how long such implants last beyond the 8.4 months followed here.
When words fail: Mima, a listening ear
This implant was tested in one man with ALS, as research. It shows faster communication is possible, but it is not available to patients.
Mima does not replace this technology. She can help you or a relative note what matters and prepare questions for the team, at your own pace. A question: "Which communication aids can I use right now?"
Frequently asked questions
Can a brain implant give speech back?
In this study, an implant linked to an AI let a man with ALS communicate with 97.5% accuracy over 8.4 months. It involved one patient, so it is a first demonstration.
How long does it take to train the system?
Calibration needed 30 minutes of recordings, then 1.4 extra hours of training on day two to move to a 125,000-word vocabulary.
Can people already get such an implant?
No. The study is clinical research on one participant, registered as NCT00912041. The device is not presented as an available treatment.
Is the voice the patient's own?
Decoded words are spoken by text-to-speech software designed to sound like the voice he had before ALS.
The original study
Card NS, Wairagkar M, Iacobacci C, Stavisky SD, Brandman DM, et al. “An Accurate and Rapidly Calibrating Speech Neuroprosthesis.” The New England Journal of Medicine, August 2024. DOI : 10.1056/NEJMoa2314132
Read the full study in the New England Journal of MedicineThis article summarises a scientific study and is not medical advice.