AI Models to Interpret Brainwaves Like Human Experts

Artificial intelligence (AI) models for interpreting routine clinical electroencephalograms (EEGs) perform similarly to human experts, according to research results published online June 20. JAMA Neurology.

Dr. Jesper Tveit, Holberg EEG, Bergen, Norway, et al. have developed and validated an AI model, Standardized Computer-Based EEG-Artificial Intelligence Reporting (SCORE-AI), that can distinguish between abnormal and normal EEG recordings. Categorize abnormal EEG recordings. SCORE-AI was developed and validated using brain waves recorded between 2014 and he 2020. A total of 30,493 records were included in the development dataset and annotated by 17 experts. SCORE-AI was validated using three independent datasets. A multicenter dataset consisting of 100 representative EEGs. A single-center data set of 9,785 EEGs. Also a dataset of 60 of his EEGs containing external reference standards for benchmarking with previously published models.

Researchers found that SCORE-AI achieved high accuracy. For different categories of his EEG abnormalities, the area under the receiver operating characteristic curve varied from 0.89 to 0.96, showing similar performance to human experts. Previously published benchmarks against three AI models were limited to comparing epileptic anomaly detection. SCORE-AI’s accuracy was 88.3%, significantly higher than previously published models and comparable to human experts.

“Our convolutional neural network model SCORE-AI achieved expert-level performance in routine clinical EEG readings,” the authors wrote. “Its application could help provide useful clinical information in remote and underserved areas where expertise in EEG interpretation is minimal or unavailable.”

Several authors disclosed ties to the pharmaceutical industry.