By Matt Silver
SAN ANTONIO, Tex -- June 12, 2019 -- Artificial intelligence (AI) can help predict the severity of obstructive sleep apnea (OSA) using simple overnight electroencephalogram (EEG) tests, according to a study presented here at SLEEP 2019, the 33rd Annual Meeting of the Associated Professional Sleep Societies (APSS).
“What we’ve shown is that existing neurodiagnostic EEG data flows that are most commonly used to evaluate seizure disorders hold promise for preliminary assessment and screening of moderate to severe sleep apnea,” said Chris Fernandez, PhD, University of Wisconsin School of Medicine and Public Health, Madison, Wisconsin. “[It] has the potential to enable enhanced care coordination bidirectionally between sleep medicine and neurological specialist physicians.”
EEG studies typically require several full nights of sleep to provide enough data. However, the current study has shown the potential for AI to make EEG studies, with respect to sleep apnoea, much quicker.
“With a single sample of blood, it is possible to run multiple diagnostic tests on that sample, contributing to a high diagnostic utility of each collected sample,” explained Dr. Fernandez. “Within this analogy, we sought to evaluate the applicability for multiple diagnostic tests to be run on a single patient waveform sample, such as an ambulatory EEG study, holter monitor study, period of inpatient monitoring, and others.”
The researchers performed a cross-sectional analysis of 4,650 patients who completed an overnight polysomnography (PSG) study but only included standard 10-20 EEG sensor array data in their analysis. They then used AI methods, including bidirectional-LSTM and deep-CNN to model the relationship between global and local EEG phenotypes and OSA severity.
The AI-based models provided an average accuracy, sensitivity, and specificity of 91.1%, 86.9%, and 99.5%, respectively, for predicting moderate and severe OSA.
“What surprised us the most about the results was just how flexible the AI model was, in terms of the capability to tune the statistical performance for different clinical scenarios and settings,” said Dr. Fernandez. “In particular, the model was configured for nearly 100% specificity while correctly identifying more than 8 in 10 cases of moderate to severe OSA, providing potential for this method to serve as an objective screening tool to complement existing, validated screening tools like the STOP-BANG, Epworth Sleepiness Scale, and the 4-Variable screening tool, among others.”
With further validation, AI-based risk estimates could be incorporated into diagnostic EEG reports, providing clinicians with an additional means for identifying patients with moderate and severe OSA that may benefit from follow‑up diagnosis and treatment.
[Presentation title: Using Novel EEG Phenotypes and Artificial Intelligence to Estimate OSA Severity. Abstract 0932]
To read more Conference Dispatch articles, click here.