Automated sleep scoring system using multi-channel data and machine learning.

Area of Science:

• Medical Informatics
• Machine Learning
• Sleep Medicine

Background:

• Manual sleep staging is time-consuming, subjective, and requires specialists.
• Accurate sleep staging is crucial for diagnosing and treating sleep disorders.
• Automated methods offer a potential solution to improve efficiency and objectivity.

Purpose of the Study:

• To develop and evaluate an accurate, automated sleep-scoring system.
• To compare the performance of Random Forest, Extra Trees, and Decision Tree classifiers for sleep staging.
• To leverage a comprehensive dataset from 19 polysomnography (PSG) sensors.

Main Methods:

• Utilized sleep data from 50 patients recorded by 19 Philips Alice clinic PSG sensors.
• Applied data under-sampling techniques to manage large datasets (approx. 87 million data points across 19 channels).
• Trained and tested Random Forest, Extra Trees, and Decision Tree classifiers on the processed dataset.

Main Results:

• Achieved average accuracies of 95.258% (Extra Trees), 95.17% (Random Forest), and 91.318% (Decision Tree).
• Obtained average precision, recall, and F1-scores of 0.95362, 0.95258, and 0.94568, respectively.
• Demonstrated superior performance compared to previous studies through a unique database and 19-channel data utilization.

Conclusions:

• The proposed automated sleep-scoring system demonstrates high accuracy and efficiency.
• This method can potentially reduce the burden on sleep specialists and accelerate the sleep scoring process.
• The use of 19 PSG channels and a proprietary database contributes to the system's effectiveness.