A new artificial intelligence (AI) model may improve patient diagnosis and care among specialized epilepsy centers in underserved areas.
The use of an artificial intelligence (AI) model to interpret routine clinical electroencephalograms (EEGs) was able to achieve diagnostic accuracy in detecting and identifying epilepsy readings, similar to that of human experts, according to a new study.
“These results warrant clinical implementation with a potential to improve patient care in remote and underserved areas where EEG expertise is scarce or unavailable,” wrote the researchers of the study. In addition, SCORE-AI [Standardized Computer-based Organized Reporting of EEG-Artificial Intelligence] may help improve efficiency and reduce excessive workloads for experts in tertiary care centers who regularly interpret high volumes of EEG recordings.”
This clinical validation study is published in JAMA Neurology.
Although previous AI models have been able to detect normal and abnormal epileptiform activity, the researchers believe that a fully automated AI model is needed in routine clinical practice. To the researchers knowledge, this AI model is the first of its kind to complete a fully automated and comprehensive clinically relevant assessment of routine EEGs.
The study investigators developed SCORE-AI with the aim of being able to distinguish between normal and abnormal EEG readings, as well as classify abnormal readings into clinical decision-making categories: epileptiform-focal, epileptiform-generalized, nonepileptiform-focal, and nonepiletiform-diffuse.
SCORE-AI was developed and validated using 30,493 anonymized EEGs recorded from 2014 to 2020, which were analyzed from January 17, 2022, to November 14, 2022, and annotated by 17 experts. All patients 3 months and older and not critically ill were eligible for the study. Of the patients in this dataset, 14,980 were male and the median age was 25.3 (95% CI, 1.3-76.2) years.
Additionally, SCORE-AI was validated using 3 independent datasets: a multicenter dataset of 100 patients, in which 61 were male patients and the median age was 25.8 (95% CI, 4.1-85.5) years, evaluated by 11 experts; a single-center dataset of 9785 patients, of which 5168 were male patients and the median age was 35.4 (95% CI, 0.6-87.4) years, evaluated by 14 experts; and a benchmarking dataset of 60 patients from 3 published AI models using external reference standards, with 27 male patients and median age of 23 (95% CI, 3-75) years.
The researchers found that SCORE-AI was able to achieve high accuracy, between 0.89 and 0.96 under the receiver operative characteristic curve, for identifying the different categories of abnormal EEG readings, similar to that of humans. Furthermore, benchmarking against 3 previously published AI models, SCORE-AI achieved higher accuracy (88.3%; 95% CI, 79.2%-94.9%; P < .001).
The researchers acknowledge some limitations to the study, such as that it did not include newborns or critically ill patients and the model was trained to identify biomarkers interpreted by experts.
Despite these limitations, the researchers believe the study shows how AI can be used to improve patient diagnosis and care, especially among underserved areas.
“Its application may help to provide useful clinical information in remote and underserved areas where expertise in EEG interpretation is minimal or unavailable,” wrote the researchers. “Importantly, it may also help reduce the potential for EEG misinterpretation and subsequent mistreatment, improve interrater agreement to optimize routine interpretation by neurologists, and increase efficiency by decompressing excessive workloads for human experts interpreting high volumes of EEGs.”
Reference
Tveit J, Aurlien H, Plis S, et al. Automated interpretation of clinical electroencephalograms using artificial intelligence. JAMA Neurol. Published online June 20, 2023. doi:10.1001/jamaneurol.2023.1645
Public Hospitals More Likely to Extend Unprofitable Services After 340B Participation, Study Finds
May 10th 2024Public hospitals were significantly more likely to sustain access to unprofitable services following 340B Drug Pricing Program participation, while nonprofit hospitals were mostly unaffected, according to a recent study.
Read More
Frameworks for Advancing Health Equity: Urban Health Outreach
May 9th 2024In the series debut episode of "Frameworks for Advancing Health Equity," Mary Sligh, CRNP, and Chelsea Chappars, of Allegheny Health Network, explain how the Urban Health Outreach program aims to improve health equity for individuals experiencing homelessness.
Listen
Health Equity, Accessibility Essential to Supporting Women Affected by Hemophilia
May 10th 2024Due to often being overlooked as only carriers of hemophilia, the review authors sought to highlight the needs of female caretakers for patients with the disorder and the possibility of women themselves being affected by it.
Read More
CMS Medicare Final Rule: Advancing Benefits, Competition, and Consumer Protection
May 7th 2024On this episode of Managed Care Cast, we're talking with Karen Iapoce, senior director of government products and programs at ZeOmega, about the recent CMS final rule on Medicare Part D and Medicare Advantage.
Listen
Breast Cancer Treatment Disparities Impact Survival Based on Race, Age, Socioeconomics
May 9th 2024A new study has linked racial and ethnic disparities with factors like age, income, and insurance to breast cancer treatment decline. Patients who received all treatments had better survival, highlighting the need for interventions to improve access and reduce disparities.
Read More