Advancing Sleep Studies: An Exploration of Phase-Based Approaches for Sleep Staging

• Published in: 2024 35th Irish Signals and Systems Conference (ISSC) pp. 1 - 6
• Main Authors: McCausland, Christopher, Biglarbeigi, Pardis, Bond, Raymond, Finlay, Dewar
• Format: Conference Proceeding
• Language: English
• Published: IEEE 13.06.2024
• Subjects: Accuracy; Band-pass filters; Correlation; Electroencephalogram; Interprebility; Machine Learning; Phase Locking Value; Sleep staging; Support vector machines; Time-frequency analysis; Training; Visualization
• ISSN: 2688-1454
• DOI: 10.1109/ISSC61953.2024.10602946

Sleep staging is an important but burdensome process required for the assessment of overnight polysomnography. A major criticism of sleep staging is the unacceptable levels of inter-scorer reliability, caused in part, by the subjectiveness of visual interpretation. More recent literature has also proposed that differing regions of the brain can be in different sleep stages simultaneously which may add further confusion. This study assessed the potential of phase-based methods to discern sleep stages. Phase-based features were generated from inter-electrode and sub-frequency-bands of EEG signals. Classification was subsequently undertaken with a support vector machine used due to the high dimenionality of the data. The SVM was trained on 213,055 sleep stage observations with 23,675 test observations. An accuracy of 80.4% was achieved which was inline with current interscorer reliability maximums, without over-fitting. Observed performance metrics were 0.80, 0.80, 0.95, 0.80 and 0.75 for Sensitivity, PPV, Specificity, F1 Score and Cohen's kappa respectively.