Effective Diagnosis of Various Sleep Disorders by LEE Classifier: LightGBM-EOG-EEG

Area of Science:

• Neurology
• Medical Informatics
• Machine Learning

Background:

• Accurate diagnosis of sleep disorders is crucial for effective treatment.
• Existing diagnostic methods can be time-consuming and resource-intensive.
• Machine learning offers potential for automated and efficient sleep disorder classification.

Purpose of the Study:

• To develop and evaluate a novel machine learning approach for classifying seven common sleep disorders.
• To compare the diagnostic performance of models using electrooculogram (EOG) and electroencephalogram (EEG) data.
• To achieve state-of-the-art accuracy in multi-class sleep disorder diagnosis.

Main Methods:

• Utilized the Light Gradient Boosting Machine (LightGBM) decision tree model for classification.
• Evaluated performance on the publicly available CAP dataset comprising 108 subjects.
• Developed a LightGBM model using EOG data (L-EOG) and a combined EOG-EEG model (LEE).
• Incorporated binary classifiers and thresholding rules to refine classification accuracy.

Main Results:

• The L-EOG classifier achieved 83.3% accuracy for seven-class sleep disorder classification.
• The LEE classifier, integrating EOG and EEG, improved accuracy to 93.3%.
• A final thresholding rule enhanced performance to 94.4% accuracy.
• This represents the highest reported accuracy for seven-class sleep disorder classification in the literature.

Conclusions:

• The proposed LightGBM-based method, particularly the LEE classifier, demonstrates high accuracy in diagnosing multiple sleep disorders.
• Combining EOG and EEG data significantly enhances diagnostic performance compared to EOG alone.
• This approach shows promise for improving the efficiency and accuracy of sleep disorder diagnosis.