Medical implants are, for the most part, silent data collectors: they record, clinicians review later, and patients rarely see any of it until a scheduled appointment. A team at the University of Pennsylvania has tested a very different model — a medical device that patients can actually talk to, and that talks back.
In a proof-of-concept study published in Nature Medicine, researchers at Penn's Perelman School of Medicine and the Center for Neuroengineering and Therapeutics (CNT) describe an AI-driven platform that let people being monitored for epilepsy exchange messages with their brain monitoring system in real time — asking questions about their own brain activity, receiving personalized alerts, reporting symptoms, and even helping train the system to recognize their seizures more accurately. For more context on this story, see our ongoing latest AI developments.
How the Platform Works
The study enrolled 13 adults admitted to Penn's Epilepsy Monitoring Unit for continuous brain monitoring with electroencephalography (EEG) — some to pinpoint the source of their seizures, others as part of pre-surgical evaluation. The platform they tested consisted of three components: software that continuously captured EEG data from electrodes in each participant's brain; a secure, HIPAA-compliant cloud environment analyzing the incoming data in near real time; and a smartphone app featuring a Penn-developed AI conversational agent, built with technology similar to that behind ChatGPT.
Through the app, patients could ask questions about their brain activity, request visualizations of trends in their data, report symptoms, and receive alerts or surveys throughout the day. The system could also reach out unprompted when it spotted changes it judged relevant, and adapt "on the fly" based on patient responses.
"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, like your own personal Jiminy Cricket on your shoulder," said senior author Brian Litt, MD, a professor of Neurology, Neurosurgery, and Bioengineering. "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."
What 1,307 Messages Revealed
During the study period, participants exchanged 1,307 messages with the system. Most conversations centered on seizure events, sleep, mood, and energy levels — factors with direct bearing on seizure activity. Patients were also surveyed three times daily about sleep, anxiety, mood, well-being, and other symptoms.
The feedback loop worked in both directions. Participants could tell the system which seizure detections were correct and which were false alarms, and the platform substantially improved its ability to identify seizures accurately as a result. Clinicians could review the system's assessments and add their own annotations through what the team calls a smart "expert in the loop" algorithm, which concentrated expert input on only the highest-value cases.
Several participants said the platform helped them make sense of patterns they had noticed but never understood. One reported that it helped "solidify what I felt to be true between some of my episodes and emotions." Others valued access to their sleep data and the possibility of warning before major seizures.
From Epilepsy to Insulin Pumps and Deep Brain Stimulation
The team sees the platform as a template for the next generation of medical devices far beyond epilepsy. Potential applications include cardiac devices, insulin pumps, and emerging neuropsychiatric treatments for Parkinson's disease, depression, obsessive-compulsive disorder, addiction, and cognitive disorders.
Work is already underway: the researchers are preparing studies with Lauren Hammer, MD, PhD, an assistant professor of Neurology and Bioengineering, to apply the platform to automatically tune deep brain stimulation devices used to treat Parkinson's disease. Further projects aim to incorporate wearables for insomnia, epilepsy, and monitoring of brain and vital signs.
"Patients often understand aspects of their disease that sensors cannot measure directly," Litt said — envisioning, as the release puts it, an insulin pump that could answer questions about how a meal will affect blood sugar. As such systems become more personalized, the researchers expect them to require less clinical supervision, making long-term outpatient use practical.
A Shift in What Medical Devices Are For
The study's broader significance is conceptual as much as technical. Where today's devices mostly collect data for periodic clinician review, the Penn platform treats the device, the patient, and the care team as collaborators in an ongoing conversation. The findings were published in Nature Medicine by the Perelman School of Medicine and CNT team, with the platform itself constructed by Penn Bioengineering PhD students led by Zack Goldblum, Haoer Shi, and Zhongchuan Xu.
The experiment is small — 13 patients in a controlled monitoring unit — and the technology is early. But it offers a concrete glimpse of what happens when the data a medical device collects stops flowing one way, and becomes something patients can question, correct, and learn from in real time.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
