PSGMiner: A modular software for polysomnographic analysis

• Published in: Computers in biology and medicine Vol. 73; pp. 1 - 9
• Main Author: Umut, Ilhan
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.06.2016; Elsevier Limited
• Subjects: Adult; Aged; Aged, 80 and over; Algorithms; Automatic Data Processing - methods; Classification; Computer software; Digital signal processing; Electroencephalography; Female; Fractals; Heart rate; Humans; Internal Medicine; Laboratories; Machine learning; Male; Middle Aged; Other; Physiology; Polysomnography; Polysomnography - methods; Researchers; Signal processing; Sleep disorders; Software; Software packages; Studies; Polysomnography; Computer software; Digital signal processing; Machine learning
• ISSN: 0010-4825; 1879-0534
• DOI: 10.1016/j.compbiomed.2016.03.023

Abstract Purpose Sleep disorders affect a great percentage of the population. The diagnosis of these disorders is usually made by polysomnography. This paper details the development of new software to carry out feature extraction in order to perform robust analysis and classification of sleep events using polysomnographic data. The software, called PSGMiner, is a tool, which visualizes, processes and classifies bioelectrical data. The purpose of this program is to provide researchers with a platform with which to test new hypotheses by creating tests to check for correlations that are not available in commercially available software. The software is freely available under the GPL3 License. Method PSGMiner is composed of a number of diverse modules such as feature extraction, annotation, and machine learning modules, all of which are accessible from the main module. Using the software, it is possible to extract features of polysomnography using digital signal processing and statistical methods and to perform different analyses. The features can be classified through the use of five classification algorithms. PSGMiner offers an architecture designed for integrating new methods. Comparison with existing methods Automatic scoring, which is available in almost all commercial PSG software, is not inherently available in this program, though it can be implemented by two different methodologies (machine learning and algorithms). While similar software focuses on a certain signal or event composed of a small number of modules with no expansion possibility, the software introduced here can handle all polysomnographic signals and events. Conclusions The software simplifies the processing of polysomnographic signals for researchers and physicians that are not experts in computer programming. It can find correlations between different events which could help predict an oncoming event such as sleep apnea. The software could also be used for educational purposes.