Authors :
Presenting Author: Joanne Lau, MD, MS – Stanford University
John-Paul Sumaquero, BS, R EEG/EP T, CLTM, CNIM – Stanford University
Kevin Graber, MD – Stanford University
Rationale:
The integration of artificial intelligence (AI) into clinical practice represents a significant advancement in modern medicine. Ceribell Rapid Response Electroencephalography (EEG) has incorporated Claritγ AI, an automated algorithm designed to detect seizure activity and quantify seizure burden, alongside the Automated Insights Bar that highlights regions warranting closer review. Stanford Hospital implemented a clinical protocol in July 2025 leveraging Claritγ–derived seizure burden to help inform clinical decision-making and optimize resource utilization, particularly overnight. The study evaluates this system on real-world clinical care and healthcare resource management.
Methods:
We are conducting a prospective observational study of patients ≥ 17 years undergoing evaluation for possible seizures with the Ceribell EEG system. Seizure activity and burden are quantified by Claritγ as the percentage of ten-second epochs with seizure activity within 5-minute segments (e.g. ≥ 90% seizure burden indicates ≥ 4.5 minutes of seizure activity in the last 5 minutes and triggers status epilepticus alerts). The sensitivity and specificity of Claritγ–derived seizure burden, along with the performance of the Automated Insights Bar within regions of Claritγ–detected seizure burden, are compared with final EEG interpretations reviewed by two epileptologists. The Ceribell Rapid Response EEG is used to derive 2HELPS2B scores to stratify seizure risk and guide monitoring duration, which is analyzed alongside transition rates to conventional EEG, subsequent recording length, concordance of EEG interpretation, and seizure incidence on conventional EEG to evaluate resource utilization.
Results:
Since protocol implementation, 52 Ceribell EEGs have been placed, with 16 conducted overnight, primarily in the Emergency Department and Intensive Care Unit. Overnight mean recording duration was 321 minutes, during which Claritγ detected a maximal seizure burden ranging from 0% to 76.67%. Claritγ did not miss any seizure activity compared to the final EEG interpretations. The sensitivity and specificity of overnight Claritγ–detected seizure burden (> 3.33%) were 100% and 46.15%, respectively. Within the segments of detected seizure burden, the presence of Automated Insight Bars had a sensitivity of 100% and specificity of 28.57%. The 2HELPS2B scores overnight ranged from 0-2. Among 6 with a 2HELPS2B score of 0, 33% were converted to conventional EEG for suspected functional neurologic disorder with video monitoring. For 9 with a 2HELPS2B score of 1, 44% were converted due to abnormal Ceribell findings or clinical history. All EEGs with a score of 2 were converted to conventional EEG. Data collection is ongoing.
Conclusions:
Preliminary findings indicate high sensitivity of Claritγ and its Automated Insights Bar for seizure detection, and demonstrates the utility of Ceribell EEG, particularly overnight. This highlights the potential of AI to enhance patient care and clinical workflows.
Funding:
No specific grant funding.