Integrating AI Into EEG Evaluation and Epilepsy Care (Podcast)

When it comes to epilepsy care, opportunities for efficiencies from artificial intelligence (AI) abound. Consider interpretation of a continuous EEG study, which can take about two hours even for a well-trained epileptologist.

“EEG evaluation is almost ready-made to be a proof of concept for the development of an AI model to improve patient care in epilepsy,” says Imad Najm, MD, Director of Cleveland Clinic’s Epilepsy Center. “At Cleveland Clinic, EEGs have been digitized since the late 1980s, so as one of the largest epilepsy programs in the U.S., we have hundreds of thousands of hours of digital EEG recordings. More importantly, these EEGs are from patients for whom we also have full digital records, so for the most part we know their diagnosis, their treatments and their treatment outcomes. So for any type of AI model we develop to read EEGs, in addition to detecting EEG patterns, we can build the model to ascribe these patterns to a validated diagnosis and ultimately help inform treatment. That’s why EEG is an ideal application to start with.”

He notes that in addition to workload efficiencies, this approach increases the objectivity of EEG evaluation. “An EEG can be read in two different ways by two different experts,” Dr. Najm says. “We want to minimize or eliminate those discrepancies, and we believe AI can help us do that while still giving clinicians the final say.”

This approach to EEG evaluation, which Cleveland Clinic aims to introduce to clinical use in the next few months, is one point of discussion in the latest episode of Cleveland Clinic’s Neuro Pathways podcast. In the episode, Dr. Najm explores how AI and advanced data infrastructure are transforming epilepsy diagnosis and treatment by enhancing EEG analysis and clinical decision-making. Subtopics addressed include the following:

• Current challenges in EEG interpretation
• The rationale for AI integration in EEG analysis
• Current state and future implementation of AI models
• AI capabilities and limitations
• Multimodal data integration and AI-based analyses
• Impact on clinical workflow and patient care
• Future directions

Click the podcast player above to listen to the 30-minute episode now or read on for an edited excerpt of its transcript. Check out more Neuro Pathways episodes at clevelandclinic.org/neuropodcast or wherever you get your podcasts.

This activity has been approved for AMA PRA Category 1 Credit™ and ANCC contact hours. After listening to the podcast, you can claim your credit here.

Excerpt from the podcast

Podcast host Glen Stevens, DO, PhD: Is Cleveland Clinic currently using some form of AI to assist in EEG review in clinical settings?

Imad Najm, MD: Not yet. We are now in the final stages of preparing for that, which requires FDA approval of the AI model we intend to use. We hope that will happen in the fourth quarter of 2026 and we will be able to start to use the model at a small scale within our operation. Then, once we move to our new Neurological Institute building in early 2027, we will have at least one part of our central monitoring unit equipped with an AI model for live analysis of EEGs recorded anywhere in the inpatient hospital setting outside the epilepsy monitoring units. That would include neurological intensive care and medical intensive care units at any of our locations within the Cleveland Clinic health system.

That will be the first step. Our planned second step will be to develop and implement a dedicated model for live analysis and post-processing of EEGs for detection of seizures and abnormal EEG activity among patients in our epilepsy monitoring units.

Dr. Stevens: To what extent will the AI be able to generate readings that are good to go versus you, as the physician, needing to provide interpretation?

Dr. Najm: An AI model is only as good as the data that is used to develop and validate it and make it work as a learning model over the long term. So, at this point, we are training our model to read EEGs, identify abnormalities, and assign these abnormalities to some labels that our experts agreed upon. Our AI model is not currently trained to interpret an EEG in the context of a patient's condition. This is a much higher level where physicians need to have the final say.