Cleveland Clinic It is teaming up with San Francisco’s Piramidal startup to create a massive AI model for monitoring brain health of patients in intensive-care units.
In place of training on text, this system uses data from electroencephalograms (EEG), which are collected by electrodes that are placed on the skull and read out on a computer as a series wavy lines. EEG records the brain’s electrical activity—and changes in this activity can indicate a problem. Doctors scan EEG in an ICU to look for signs of altered consciousness or brain dysfunction.
Doctors rely currently on EEG continuous monitoring to detect abnormal activity of the brain in ICU patients, but cannot monitor each patient individually in real-time. EEGs are usually generated once every 24 or 12 hours. They’re then analysed to see if a particular patient is having a problem. Manually reviewing a full day of brainwaves can take up to two hours.
“This type of thing is time-consuming. It is subjective, and it is experience- and expertise-dependent,” Imad Najm is a neurosurgeon and the director of Cleveland Clinic’s Neurological institute.
It is a system designed by Piramidal, Cleveland Clinic to identify abnormalities and interpret EEG signals in seconds. Doctors can then intervene faster.
“Our model plays that role of constantly monitoring patients in the ICU and letting the doctors know what’s happening with the patient and how their brain health is evolving in real-time,” Kris Pahuja, chief product officer at Piramidal.
Pahuja and CEO Dimitris Fotis Sakellariou founded Piramidal in 2023, with the goal of building a foundation model for the brain—an AI system that can read and interpret neural signals broadly across different people. Sakellariou was a researcher in EEG for fifteen years before he became a scientist. Pahuja has worked in product strategy for Google and Spotify. The startup is backed Y Combinator and raised $6m in seed financing last year.
The Cleveland Clinic, among other partners and publically available EEG datasets were used to build the company’s ICU brain models. Sakellariou said the model incorporates almost a million EEG monitoring hours from “dozens of thousands” Both neurologically healthy patients and those with unhealthy conditions are included. The brain activity patterns of each individual are very different, and so a model that captures common features is difficult to build.
“The beauty of a foundation model is the same way ChatGPT can generalize text, it can adapt to your tone, it can adapt to your way of writing—our model is able to adapt to the brains of different people,” Sakellariou says.
The Cleveland Clinic team and Piramidal are currently using retrospective data from patients to refine the model. They plan to test their model within the next 6-8 months in an ICU with real-time patient data, a small number of doctors and beds. They then plan to roll the software out to all ICUs. Najm claims that eventually, the software would allow hospital staff to keep track of hundreds of patients simultaneously.
