AdWorld Businesses for Sale
Visit World Businesses for Sale
Thinking of selling your business? Buy or sell a business locally, nationally or worldwide. 10% off with code AWN10.
Epilepsy patients chatted with their EEG brain monitor
The full story behind my Short: Penn researchers linked hospital EEG to a secure AI chat app: 13 epilepsy patients exchanged 1,307 messages, and their labels on seizure alerts cut false alarms by about 80% in a simulation (Nature Medicine, 13 Oct 2026).
The short version
Penn researchers built an AI app that let 13 hospital epilepsy patients chat with their EEG brain monitor by phone. Across 1,307 messages, patients asked about seizures, sleep and brain spikes, got alerts, and marked seizure alerts as real, false or unsure. The AI answered 92% (35 of 38) of data questions correctly, and in a simulation, retraining the seizure detector on patients' own labels cut false alarms by about 80%, from 0.74 to 0.10 per hour, with no expert input. Five of the 13 asked to keep using it after discharge. Caveats: small, short inpatient study; the retraining was simulated (Goldblum et al., Nature Medicine, 13 Oct 2026).
13epilepsy patients on EEG [1]
1,307messages exchanged [1]
92%data questions answered correctly (35/38) [1]
~80%fewer false alarms, simulated (0.74 to 0.10/hr) [1]
What happened
Thirteen adults in the Hospital of the University of Pennsylvania epilepsy monitoring unit were monitored with scalp EEG (11) or intracranial EEG (2). Their EEG streamed to a HIPAA-compliant cloud, where pretrained models estimated seizure probability, sleep stages and spike rates. A large language model (GPT-4o-mini in a secure tenancy) chatted with patients through a phone app, sent alerts and surveys, and answered data questions by querying their own database. Patients responded to 67.2% of the 58 real-time seizure alerts [1].
Watch the Short
About this page
The Short uses illustrative photos from Wikimedia Commons (credited below) and my own sketches; they are not photos of the Penn study, its patients or its app.
The narration is an AI clone of my own voice (Chatterbox), used with my consent.
Why it matters
- Devices that listen. Most medical devices cannot share what they detect with patients in real time; this one could, and it learned from replies [1].
- Fewer false alarms. Patient labels alone retrained the detector in a pseudoprospective simulation, cutting false alarms by about 80% while keeping sensitivity [1].
- Early days. 13 patients, a median of about 5 days, an artificial hospital setting, and only the baseline detector was used live [1].
Full narration (transcript)
What if your medical device could text you back? So let's talk about what Penn actually built here. Thirteen hospital patients with epilepsy were wired to EEG. Their brain data streamed to a secure AI they could chat with by phone. They asked about their seizures, their sleep, and spikes in brain activity. In days, 1,307 messages flew back and forth, about 100 per patient. And it answered 92 percent of data questions correctly. But here's where it gets interesting. Patients could mark each seizure alert as real, false or unsure. In a simulation, those labels retrained the seizure detector, with no expert help. False alarms fell by about 80 percent: from about one every 80 minutes to one every 10 hours. Five of the 13 asked to keep using it after leaving hospital. Caveat: a small, short study, and the retraining was simulated. Why it matters: devices that listen could learn faster, and cry wolf less. For the full story, follow the link in the description.
Sources
- [1] Goldblum Z., Shi H. et al. (13 Oct 2026). “Human–artificial intelligence collaboration for comanaging epilepsy in real time.” Nature Medicine. doi.org/10.1038/s41591-026-04680-4
- [2] Medical Xpress (Oct 2026). “AI platform enables real conversation with patients” (news report on the study). medicalxpress.com
Credits
Photos and images via Wikimedia Commons (illustrative): Hand holding smartphone with blank white screen.jpg (Santeri Viinamäki, CC BY-SA 4.0); Hospital of the University of Pennsylvania, 2024.jpg (JJonahJackalope, CC BY-SA 4.0); EEG Recording Cap.jpg (Chris Hope, CC BY 2.0); Patient room with hospital bed.jpg (Diane A. Reid (Photographer), Public domain); Smartphone Use.jpg (Océanos y dados, CC0); EEG Closing eyes.png (Otoomuch, CC BY-SA 3.0); BalticServers data center.jpg (BalticServers.com, CC BY-SA 3.0); Hand holding Smartphone.jpg (Mictlancihuatl, CC BY-SA 4.0); UPenn Hospital Campus (53589299182).jpg (ajay_suresh, CC BY 2.0). Sketches and animation: my own. Narration: synthetic voice.