Efficient system for classifying cyclic alternating pattern phases in sleep.

Area of Science:

• Neuroscience and Biomedical Engineering
• Sleep Medicine and Signal Processing

Background:

• Electroencephalogram (EEG) signals are crucial for sleep analysis.
• Cyclic Alternating Patterns (CAP) in EEG during sleep are vital for diagnosing sleep disorders.
• Accurate differentiation of CAP phases (A and B) is essential for detailed sleep analysis.

Purpose of the Study:

• To develop an accurate and easily implementable system for distinguishing between CAP phase A and phase B in EEG signals.
• To enhance the early diagnosis of sleep disorders through improved CAP analysis.
• To compare the performance of different machine learning algorithms for CAP phase classification.

Main Methods:

• EEG signal preprocessing: denoising and segmentation.
• Signal decomposition into frequency sub-bands using zero-phase filtering.
• Statistical feature extraction and selection using the Kruskal-Wallis test.
• Classification using k-nearest neighbour (kNN), support vector machine (SVM), bagged tree (BT), and neural network (NN) algorithms.

Main Results:

• The bagged tree (BT) classifier achieved the highest accuracy (83.29%) and F-1 score (83.58%) on a combined balanced dataset.
• The proposed method demonstrated superior accuracy and efficiency compared to existing techniques.
• Successful classification was achieved for datasets including both healthy subjects and those with insomnia.

Conclusions:

• The developed system provides an accurate and efficient method for differentiating CAP phases in EEG signals.
• The proposed approach holds potential for widespread clinical application in sleep disorder diagnosis.
• Machine learning, particularly the bagged tree classifier, shows promise for automated sleep pattern analysis.