Written by Ralph Sun
One in three clinically unresponsive ICU patients learned to control a tone with their thoughts, using electrodes already attached to their heads. The breakthrough is less about decoding than about where it happens, and who it happens to.
A new study from Columbia reports something that should be difficult to read without a reaction. Forty-one clinically unresponsive patients with acute brain injury, tested a few days after they were hospitalised, were able to raise and lower the frequency of a beeping tone using their minds alone, through noninvasive EEG electrodes that recorded and translated their brainwaves.
“Research in the past decade by us and others has shown that many patients with acute brain injury who appear to be in a coma actually have some level of preserved consciousness, even though they are unable to speak or move their hands or eyes to communicate,” says Jan Claassen, a professor of neurology at Columbia and medical director of the neurocritical care unit at NewYork-Presbyterian/Columbia. “Our brain-computer interface is only a first step, but it shows that, in principle, we could hopefully in the future give some of these patients a voice.”
The context makes the result less surprising and more serious. In 2019, Claassen’s team used standard EEG and AI to analyse the brain waves of unresponsive brain-injured patients and detected signs of hidden consciousness in 15% of them — a figure that shocked critical care, which had previously documented the phenomenon in only a handful of cases. Those patients did not respond to commands to blink, make a fist or wriggle their toes, but the EEG showed they were trying to comply and could not move their muscles. Later work traced the dissociation to a breakdown in communication between the regions handling higher-order cognition and those controlling movement. The condition is called cognitive motor dissociation, and patients with it are more likely to regain consciousness and recover significant day-to-day functioning.
The task design is where the cleverness sits. A tone plays through the patient’s headphones, and a recorded voice asks them to adjust its frequency by trying to open and close their hands. Electrical impulses from the brain are relayed to a bedside computer that analyses the signals in real time, and the voice encourages them when the frequency changes and congratulates them when they hit the target. The signal being decoded is intent, not movement — the one thing a patient with intact cognition and a broken motor pathway can still produce.
After a brief training period, about one in three patients could control the tones, including every patient whose hidden consciousness had been detected by a separate exam. They were able to do it within days of severe brain injury, which is not what you would expect from a system that had been tuned over weeks.
Why this counts as a major advance, in Claassen’s own framing, is worth quoting directly: fMRI has been shown to obtain yes-or-no answers from a single unresponsive patient, but it is not practical for routine use. EEG, by contrast, is used in clinical practice every day in the ICU. The only sophisticated part is the software analysing the data in real time and modulating the feedback.
That is the point I would put at the centre of this. The decisive property is not signal quality. It is eligibility. An acutely brain-injured patient cannot consent to surgery, cannot be implanted, and may recover — so implantable interfaces, however impressive, are structurally out of reach in that ward. The interface that works there is the one already attached to the head for monitoring. A modest decoding task performed with routine electrodes reaches people that a superior decoder never will.
There is an ethical edge that the field should probably address before it arrives. The study’s own roadmap points toward personalised features, a loved one’s voice, culturally relevant details, and functions to detect pain, hunger and awareness. The moment you can receive an answer to “are you in pain”, you acquire an obligation to act on it. Detection without response would be worse than the silence that preceded it.
The limits deserve equal billing. One in three patients responded, meaning two in three did not. The study demonstrated control of a tone, not reliable communication of needs. Cognitive motor dissociation still cannot be identified outside research studies, and forty-one patients is a first step rather than a standard of care.
The headline is that unresponsive patients can control a computer with their thoughts. The more consequential part is quieter: the tool was already in the room, and the thing that had not been tried was asking them.
