Artificial intelligence-assisted detection of epileptic spasms using electroencephalographic-video analysis

Area of Science:

• Medical technology
• Artificial intelligence in medicine
• Neurology

Background:

• Epileptic spasms (ES) are a severe epilepsy syndrome often challenging to diagnose.
• Accurate and timely ES detection is crucial for effective treatment and improved patient outcomes.
• Current diagnostic methods, primarily electroencephalography (EEG), can be limited in sensitivity, especially for subtle spasms.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) diagnostic tool for automatic epileptic spasms (ES) detection.
• To leverage hybrid electroencephalographic (EEG)-video signals for enhanced diagnostic accuracy.
• To compare the AI tool's performance against experienced electroencephalographers.

Main Methods:

• A retrospective cohort study utilizing a multimodal fusion approach combining video and EEG signals.
• Internal cross-validation and multicenter external validation were performed.
• The AI model was trained and tested on data from 252 patients with ES and 78 controls, including extensive video-EEG recordings.

Main Results:

• The hybrid AI model achieved superior performance over EEG-only models, with an area under the receiver operating characteristic curve of .9820.
• In clinical validation, the AI demonstrated diagnostic sensitivity of .735 and specificity of .995, comparable to experienced electroencephalographers.
• AI assistance significantly improved electroencephalographers' sensitivity in detecting subtle ES by 16%-21% without compromising specificity.

Conclusions:

• The developed AI diagnostic tool achieves clinical performance comparable to senior electroencephalographers for ES detection.
• The AI tool exhibits greater robustness in identifying subtle ES and can enhance diagnostic sensitivity when used collaboratively with clinicians.
• This technology holds significant promise for improving pediatric epilepsy care, particularly in resource-limited settings.