AI medical devices that let patients hold real-time conversations with their own brain monitors could change how chronic conditions are managed, according to a new pilot study. Researchers at the University of Pennsylvania built a smartphone platform linking live EEG data to a conversational AI assistant, letting 13 epilepsy patients ask questions about their brain activity and flag possible seizures. Published Oct. 13 in the journal Nature Medicine, the study found that patient feedback helped cut false seizure alarms by 80 percent in a retrospective simulation.
AI Medical Devices Connect Patients to Live Brain Data
The study, titled “Human-AI Collaboration for Co-Managing Epilepsy in Real Time,” was led by senior author Brian Litt, a professor of neurology, neurosurgery, and bioengineering at Penn’s Perelman School of Medicine and its Center for Neuroengineering and Therapeutics. The work was funded by the National Institutes of Health. The researchers describe the system as a new class of AI medical devices designed to collaborate with patients rather than simply record data for clinicians. The participants were adults in the epilepsy monitoring unit at the Hospital of the University of Pennsylvania, who spent several days texting their own brain monitors through a smartphone app.
The platform had three components: software capturing EEG data from electrodes on the participants’ brains, a secure HIPAA-compliant cloud analyzing the data in near real time, and a smartphone app with a Penn-developed AI conversational agent similar to ChatGPT. Through the app, patients could ask whether they had experienced a seizure, how they had slept, and what their brain activity looked like, receiving answers as text and charts. The goal is for AI medical devices to surface insights in the moment, rather than days later when a specialist reviews the data.
Researchers analyzed 1,307 messages exchanged between patients and the system. Most covered seizure events, sleep, mood, and energy levels. Patients were also surveyed three times daily about sleep, anxiety, mood, and well-being. Several participants said the platform helped explain patterns they had previously observed but struggled to understand, including one who said it confirmed a link they had suspected between their episodes and their emotions, a result the researchers see as evidence that AI medical devices can help patients understand their own conditions better.
Crucially, the conversation went both ways. When the app flagged a possible seizure, patients marked it as real, a false alarm, or uncertain, and researchers used those labels to retrain the seizure-detection algorithm. In a simulation run after the fact, false alarms fell 80 percent across patients’ hospital stays, showing how AI medical devices could improve their own accuracy by learning from the people who use them. Clinicians could also converse with the system to review its assessments, through an “expert in the loop” process that reduced expert input to only the highest-value annotations.
AI Medical Devices Point Toward Patient-Led Care
The research points to a shift from today’s AI medical devices, which largely collect data for clinicians to review periodically, toward systems that actively collaborate with patients. “These devices are made first and foremost for doctors, however, they do see a lot of information that patients would benefit from knowing,” Litt said. “There are insights buried there about how their behavior impacts health, but they currently have no way of learning them in the moment,” Litt said. The study notes that none of the three devices the FDA has approved to treat epilepsy can share information with patients in real time.
“The idea is that instead of waiting for a specialist to analyze the data collected by your device, the device could detect changes and alert you immediately,” Litt said. “More importantly, both the patient and the device can learn together over time. The device learns what matters to the patient, and the patient learns how their behavior affects their health.” The researchers identified insulin delivery, cardiac monitoring, and Parkinson’s deep brain stimulation as areas for further exploration, though the published evidence concerns epilepsy monitoring only.
The study is small and early stage: 13 patients in a supervised hospital setting, with results validated only in a retrospective simulation rather than a live outpatient trial. The system is a research prototype, not a publicly available app. Turning experimental AI medical devices into approved clinical tools will take larger trials and regulatory review, but the study shows how such devices could turn patients from passive data sources into active partners in their own care.