Automatic Artifacts Detection and Classification in Sleep EEG Signals Using Descriptive Statistics and Histogram Analysis: Comparison of Two Detectors

Abstract

The algorithm for artifacts detection and classification, applying different sets of constraint rules, was proposed. Two automatic artifacts detectors based on the proposed algorithm and thresholding techniques that use descriptive statistics and a histogram analysis, were developed. At first, the performance of both detectors was evaluated by matching with the human expert scoring. Detectors were tested with various threshold values; and ones that provided the best performance results were selected. At the end, two detectors were compared with each other and the best detector was identified. Detected artifacts were classified into three different types: body movement, muscle and slow eye movement artifacts. Analyses of detection and classification results were performed separately for the whole night sleep and for NREM sleep stages 1, 2, 3 and 4. The artifacts detector, based on the application of a threshold, calculated from a histogram, has provided the best detection results with approximately 85% of sensitivity and 84% of correct classification; the best classification results were obtained for body movement artifacts, with approximately 94% of specificity and 99% of correct classification in the analysis, performed for NREM sleep stages 1, 2, 3, and 4.