Classification of Insomnia using Cyclic Alternating patterns in Sleep microstructure

This work analyses the importance of sleep microstructural patterns for identifying sleep instability and arousals, which can be extremely useful to diagnose insomnia. We derived the quantitative metrics from the sleep macrostructures as well as microstructural patterns such as cyclic alternating patterns (CAP). Studies on classification of insomnia based on CAP subphases is limited so far. In this work, we explored the utility of different subphases of CAP in distinguishing insomniacs from healthy controls. We utilized sleep data of 9 insomnia patients and 16 healthy subjects from a publicly available dataset called CAP Sleep database. SVM classifier with linear kernel is deployed for the classification. We achieved a high accuracy of 90 ± 10% across ten different random runs with 80%-20% validation approach. Also, we obtained an accuracy of 88±12% using leave-one-subject-out approach. All the performances are evaluated in a subject-independent fashion. Proposed approach outperforms the existing studies on classification of insomnia on the CAP database.Clinical relevance— For diagnosis of insomnia, confirmation of sleep arousals plays a vital role. Although sleep macrostructures facilitate classification of sleep disorders, sleep arousals being directly associated with changes in sleep microstructure, cause sleep instability, particularly in insomnia, and thereby prove to be useful for classifying insomnia. Measurement of changes in phases A and B according to sleep cycles would be credible information for confirmatory diagnosis of insomnia through convenient sleep monitoring without impacting sleep pattern.