In the intricate setting of a neurological intensive care unit (ICU), the capability to identify and react to subclinical seizures and subtle shifts in neurological status is crucial. Unfortunately, the prevailing standard—continuous EEG monitoring—faces significant challenges, notably a dependence on manual reviews and a shortage of specialized personnel for on-the-spot analysis. In response to these hurdles, Cleveland Clinic has partnered with Piramidal, a neurotechnology startup specializing in artificial intelligence (AI), to develop an advanced AI model aimed at providing real-time, scalable, and impartial EEG interpretations for hospitalized patients.
Advertisement
Cleveland Clinic is a non-profit academic medical center. Advertising on our site helps support our mission. We do not endorse non-Cleveland Clinic products or services.Policy
This initiative, spearheaded by Imad Najm, MD, Director of the Cleveland Clinic Epilepsy Center, aims to transition EEGs from a retrospective analysis tool into a proactive clinical ally in real-time. “This serves as a proof of concept that utilizing AI models can enhance and objectify live EEG monitoring, reading, and interpretation,” Dr. Najm explains.
Unmet demand for neurocritical care monitoring
The need for bedside EEG monitoring has increased as healthcare providers appreciate its role in diagnosing serious medical conditions and shaping treatment plans. Nonetheless, the current process is labor-intensive, where a qualified technologist might spend up to two hours reviewing a 24-hour EEG, followed by around 15 minutes for a physician to finalize the report.
Furthermore, current monitoring often lacks real-time analysis. Although the data is continuously recorded, EEGs are generally reviewed by clinicians every one to two hours. This lag can be detrimental in an ICU, where the burden and duration of seizures significantly affect neurological outcomes. “It’s costly, not scalable, and interpretation can be quite subjective,” notes Dr. Najm.
AI model for brainwave analysis
The partnership with Piramidal employs the company’s AI model, which is trained to process EEG data and interpret neural signals across a wide range of patient demographics. Unlike AI solutions tied to specific hardware, this platform is vendor-neutral, capable of integrating and converting data from various EEG machine manufacturers into a standardized format for analysis.
This technology can evaluate 24 hours of EEG data in mere seconds. Beyond its speed, the model offers a detailed level of quantification that is both challenging and time-consuming to achieve manually—such as determining seizure burden (total minutes of seizure activity within a 24-hour window) and precisely timing each event.
Model development and validation
The algorithm was developed using a comprehensive array of publicly accessible EEG data and refined with meticulously curated data from the Cleveland Clinic, followed by rigorous testing and validation in a recent pilot study. Prior to the study, a team consisting of three senior EEG analysts and three expert physicians meticulously annotated thousands of hours of EEG data from the Cleveland Clinic, reaching a consensus on seizure onset, timing, localization (e.g., left, right, frontal, generalized), and types of interictal epileptic activity.
In this pilot study, which retrospectively analyzed previous recordings, the system showcased approximately 90% sensitivity and specificity for seizure detection and lateralization. Specifically, the algorithm accurately identified nine out of ten seizures, with nine out of ten events flagged as seizures confirmed as true positives, resulting in only 10% false positives attributed to activity resembling seizure patterns.
The system extends beyond seizure detection and categorization; it can also quantify seizure burden and identify critical clinical patterns relevant in an ICU, such as:
- Periodic lateralized epileptiform discharges (PLEDs), indicative of acute or subacute brain injury
- Triphasic waves, often a sign of metabolic or systemic encephalopathy (e.g., hepatic or renal failure)
- Bilateral independent periodic discharges, associated with severe acute global brain injury
Dr. Najm emphasizes that validation of the model’s capability for recognizing non-seizure patterns, such as PLEDs and triphasic waves, is ongoing, with the current 90% accuracy statistics specifically pertaining to seizure detection and localization/lateralization.
A workflow for scalable monitoring
The primary aim of this technological advancement is to establish an EEG central monitoring unit model. Dr. Najm envisions a centralized hub equipped with a massive 98-inch primary monitor surrounded by smaller displays connected to numerous monitored beds throughout the healthcare system. In this “center stage” workflow, the AI software continuously oversees all patients, automatically pushing data for any patient experiencing a seizure or significant pattern change to the large central screen. This focus enables monitoring technologists to direct their attention where it’s most needed, allowing for timely confirmation or reclassification of findings and potential alerts to the medical team.
The model also generates graphical summary reports detailing seizure frequency, timing, and burden over customizable timeframes, such as 10 or 24 hours. Individual events can be represented as interactive markers that link directly to the associated raw EEG tracing for physician evaluation. Thus, the EEG review—which currently consumes two to three hours—could potentially be reduced to just 10 to 15 minutes. “Clinicians will transition from primary analysis to confirmation and final interpretation,” Dr. Najm observes.
Implementation and future horizons
As the platform approaches FDA approval—anticipated in the coming months with support from Cleveland Clinic pilot data—the epilepsy team is gearing up for its application in clinical practice.
The centralized EEG monitoring hub will be permanently established within Cleveland Clinic’s cutting-edge Neurological Institute building on its Main Campus, set to open in early 2027. Additionally, every inpatient bed in the new building, except those in the epilepsy monitoring unit, will be outfitted for AI-supported EEG monitoring, enabling this capability across the healthcare system’s various facilities.
While the initial focus is on the ICU—where the key clinical inquiry often revolves around the binary question of “seizure or no seizure?”—future applications are expected to extend to the epilepsy monitoring unit, requiring more intricate localization and semiological analysis to aid medical decisions.
Beyond this, Dr. Najm envisions a comprehensive “AI toolbox” for managing epilepsy, integrating EEG data with seizure semiology, neuroimaging, genetics, and longitudinal outcome data for a holistic, thoroughly validated platform aimed at diagnosis, treatment, and outcome prediction.
“This is merely the initial step toward transforming patient management,” concludes Dr. Najm. “Eventually, patients worldwide could receive the same high-quality, objective, and responsive neurological care using these tools at scale.”
