Elsevier

Neurocomputing

Volumes 58–60, June 2004, Pages 1137-1143
Neurocomputing

Sleep stage classification using fuzzy sets and machine learning techniques

https://doi.org/10.1016/j.neucom.2004.01.178Get rights and content

Abstract

The hypnogram is determined after a study of electrophysiological records. In this paper we present the Intelligent system for sleep stages classification (ISSSC). This system is divided into four different modules: the first processes the electrophysiological signals and determines its most relevant parameters; the second module establishes fuzzy rules that will be used during the classification process; the third module is an inference module, it implements a fuzzy model. Finally the system builds the patient's hypnogram and provides us different outputs. We present the classification results obtained from applying the systems to classify patients with different sleep disorders.

Introduction

Sleep stages automatic classification has become a key problem in neuroinformatics and generally speaks in neuroscience. It is used to describe the patient's transition through the sleep stages (hypnogram) taking into account the study of the electrophysiological records such as Electroencephalograms (EEG), Electro-oculograms (EOG) and Electromyograms (EMG).

The most popular technique to process the sleep stages based on rules were proposed by Rechtschaffen and Kales (R&K) in 1968 [10] Today it is widely recognized that these rules have severe limitations which were not foreseen 30 years ago. Nevertheless, these rules have survived all criticisms raised in the past.

In this work we will assume that the sleep stage scoring is a kind of classification problem known as classification with learning or supervised classifications. In this problem the human sleep stages can be classified into one of 6 discrete stages according to R&K rules.

To solve our classification problem we use Machine Learning techniques and fuzzy logic theory.

Machine learning is an interdisciplinary field with connections to artificial intelligence, information theory and statistics. Machine learning algorithms learn automatically from experience and use different forms to represent knowledge. Among the many ways to represent knowledge, if–then rules [4], [2] have been used with other techniques to make inferences and classify some new cases.

On the other hand, the fuzzy rules, an example of if–then rules, are based on the use of fuzzy logic to represent knowledge. General fuzzy rules can be represented as: If x is A then y is B; where, B and A are fuzzy sets belonging to the Y and X linguistic variables, respectively.

Fuzzy sets [6] are usually identified by membership functions. A fuzzy set μ of X is a mapping involving the set X up to the unit interval: μ:X→[0,1]. Here μ(x) is known as the x-value membership degree.

In Section 2 we provide details of the modules that constitute the Intelligent System for Sleep Stages Classification (ISSSC), and in Section 3, the results of the use of this system are validated through the analysis of the system's implementation and of elements that affect the classification process. Finally, in the conclusions of this paper we evaluate the usefulness of the system.

Section snippets

Intelligent system for sleep stages classification (ISSSC)

ISSSC version 1.0, consists of four different modules (Fig. 1). The first module processes the patient's electrophysiological records that are stored in binary files.

We carried out this research in the Clinical Neurophysiology Offices of the Neurosciences Center of Cuba. We recorded the EEG, EOG and EMG using Medicid -3E. In each case we work with silver-disk electrodes stuck with collodion to the skin or to the scalp.

The EEG derivations were: Fz, C3, C4, O1 and O2 of the 1020 system. We

Validation

Evaluation is a necessary and important process in the development and testing of any software. The main purpose of this process is to determine the system's degree of correct response according to the expectations.

Brender [5] proposes some metrics and other measures that express different quality parameters of medical knowledge. Among the metric measures used in the evaluation of the system we can mention total behavior, behavior conditioned to the class (expected and inferred values), kappa

Conclusions

The ISSSC version 1.0 system is a helping system for medical diagnosis. This system allows us to create the hypnogram of patients taking into account the electrophysiological signals: EEG, EOG and EMG. The system is a useful tool for detecting sleep disorders, it also helps to spread the expert's expertise. The ISSSC system consist of four modules: signals preprocessing, machine learning, inference and inference corrector module. Each module works independent of others. This characteristic of

References (12)

There are more references available in the full text version of this article.

Cited by (0)

View full text