A Two Stage Approach for the Automatic Detection of Insomnia

Area of Science:

• Neuroscience
• Medical Informatics
• Sleep Medicine

Background:

• Chronic insomnia significantly impacts quality of life and incurs substantial societal costs.
• Automated tools for timely insomnia detection by clinicians are limited.
• Electroencephalography (EEG) is a key diagnostic tool for sleep disorders.

Purpose of the Study:

• To develop and evaluate a two-stage automated system for detecting insomnia using overnight EEG recordings.
• To leverage deep neural networks (DNNs) for sleep stage scoring and epoch-level insomnia detection.
• To create subject-level features from DNN outputs for final binary classification of insomnia.

Main Methods:

• A two-stage approach utilizing DNNs for sleep stage scoring and epoch-level insomnia detection.
• Feature extraction included temporal and spectral features from 2 EEG channels.
• Subject-level features derived from sleep stage ratios, transition ratios, and insomniac epoch probabilities were used to train binary classifiers (LDA, CART, SVM).

Main Results:

• The system achieved an F1 score of 0.88, sensitivity of 84%, and specificity of 91% in classifying participants as control or insomniac.
• The evaluation was performed on data from 115 participants (61 control, 54 with insomnia).
• The proposed method demonstrates high accuracy in distinguishing individuals with and without insomnia.

Conclusions:

• The developed two-stage automated system effectively detects insomnia from overnight EEG recordings.
• This approach offers a promising tool to assist clinicians in the timely diagnosis of insomnia.
• The findings highlight the potential of DNNs and derived features for objective sleep disorder assessment.