Patients interacting with conversational AI through a smartphone application can communicate directly with their epilepsy monitoring devices, according to a clinical study published on October 13, 2026. The system reduced false alarms by about 80 percent while operating at the Hospital of the University of Pennsylvania.
Instead of passively collecting continuous streams of physiological data that only clinical specialists review days later, the platform establishes a two-way communication channel between patients and their implanted or scalp electrodes. Researchers at the Perelman School of Medicine and the Center for Neuroengineering and Therapeutics at the University of Pennsylvania tested the technology on adults undergoing inpatient monitoring.
University of Pennsylvania Researchers Build Three-Part Smartphone Platform
The proof-of-concept project was constructed by a team of Penn Bioengineering doctoral students, led by Zack Goldblum, Haoer Shi, and Zhongchuan Xu. Software first captures continuous electroencephalography data from electrodes placed in the participant’s brain.
Participants in the unit could ask questions regarding their brain activity, request visual representations of data trends, report symptoms, and receive surveys or alerts throughout the day. Similar conversational tools were accessible to clinicians, allowing medical staff to review system assessments and add their own notes regarding true seizures and false alarms through an expert-in-the-loop algorithm.
The research team tested the platform on 13 patients admitted to the epilepsy monitoring unit at the Hospital of the University of Pennsylvania over several days.
The conversational platform answered more than 92 percent of brain activity queries with accurate, relevant text and charts. Meanwhile, users responded to more than 67 percent of real-time seizure detections by labeling them as true, false, or uncertain.
Completely unprompted, several patients asked if they could take the system home with them, which was one of the best signals we could receive.
Zack Goldblum, graduate researcher in the Litt lab, via NIH
False Alarms Drop by 80 Percent As Algorithms Learn From Patient Feedback
Continuous monitoring devices routinely flag suspicious brain activity that experts must manually review and discard. By incorporating immediate patient feedback, the software fine-tuned its underlying models on the fly. That patient-driven curation reduced false alarms by approximately 80 percent across the study period. Incorporating expert clinical feedback further trimmed alarms by another two percent.
Participants reported that the software helped clarify patterns they had previously struggled to recognize.
Researchers Apply Architecture to Parkinson’s Disease and Insulin Pumps
The development team is preparing studies alongside Lauren Hammer, an assistant professor of neurology and bioengineering, to automatically tune deep brain stimulation devices for patients with Parkinson’s disease. Additional applications are underway for cardiac devices, insulin delivery, insomnia tracking, and emerging treatments for depression, obsessive-compulsive disorder, addiction, and cognitive disorders.
“Imagine an insulin pump that can answer questions about how a meal will affect blood sugar, or a Parkinson’s device that adjusts treatment based on how a patient feels during a walk or writing a letter.”
Brian Litt, corresponding author and professor of neurology and bioengineering at the University of Pennsylvania, via Newswise
The research builds on Brian Litt’s initiative to develop advanced brain-computer interfaces, supported by a $5.6 million NIH Pioneer Award. Additional clinical facilities involved with Penn Medicine’s enterprise include Penn Medicine at Home, GSPP Rehabilitation, Lancaster Behavioral Health Hospital, and Princeton House Behavioral Health. Investigators are now planning large-scale outpatient trials to determine whether continuous conversational engagement can actively reduce seizure frequency and healthcare utilization.