Stony Brook's AI spots signs of coma recovery days before doctors

Imagine waking up in an ICU, hearing a doctor ask you to squeeze their hand, but your body refuses to move a single millimeter.

You hear them tell your family you are in a coma. In medicine, this terrifying gap is called covert consciousness: your brain is listening and trying to answer, but the signal cannot reach your limbs.

Brain scans can sometimes catch these internal signals, but they are too bulky and expensive for daily bedside checks.

Researchers led by Sima Mofakham at Stony Brook University revealed an AI tool called SeeMe that tracks covert consciousness using standard video.

Tracking awareness down to single pores

The face is driven by broad neural networks across both brain hemispheres. When motor control begins to flicker back to life, the face is usually where tiny signals surface first.

SeeMe uses computer vision to track microscopic facial movements down to the level of individual skin pores. Twitches completely invisible to a doctor's naked eye show up instantly on the system.

The team tested the setup on 16 healthy volunteers and 37 patients with severe brain injuries who appeared deeply comatose. Staff asked each patient to stick out their tongue, open their eyes, and smile.

SeeMe captured eye reactions in 30 patients and nearly all mouth movements, clocking a detection rate nearly double that of trained physicians.

Catching recovery days ahead of clinicians

Across roughly 90 percent of the patients, SeeMe spotted early signs of consciousness about four days earlier than bedside doctors.

On average, the algorithm flagged eye-opening responses four days before standard clinical exams caught them, and detected mouth reactions about eight days ahead.

In one case, an older car-crash patient showed mouth twitches picked up by SeeMe on day 18 of admission. He was not able to follow physical motor commands during routine exams until day 37.

Another coma patient showed subtle eye and mouth twitches on day 19, opened his eyes three days later, and steadily recovered.

The researchers noted that the strength and frequency of these tiny twitches closely matched how well patients recovered at discharge and six months later.

The team cautioned that SeeMe is built to assist, not replace, medical staff. A negative reading should never be taken as proof that a patient has no chance of waking up.

Why it matters

Waiting by an ICU bed is agonizing. Families and doctors routinely make life-or-death decisions without knowing if an unresponsive patient can actually hear them.

SeeMe changes the math by sidestepping multimillion-dollar MRI machines. Running pore-level tracking on standard cameras means continuous, non-invasive consciousness monitoring could actually scale to ordinary hospital wards.

We think the clever part here is taking a traditional clinical instinct—watching a patient's face—and pushing it beyond human visual limits with computer vision, rather than forcing intrusive brain probes onto frail patients.

The clear risk is clinical overreliance. If an algorithm flags recovery days ahead, it could tempt teams to jump to conclusions, just as a false negative might prematurely crush hope.

Still, if the team succeeds in turning these tiny micro-twitches into a binary yes-or-no answering system, it could give trapped patients a voice long before they can physically squeeze a hand.