Detection of Periodic Leg Movements by Machine Learning Methods Using Polysomnographic Parameters Other Than Leg...

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Data Science

Background:

• Polysomnography (PSG) utilizes numerous channels, leading to patient discomfort and technical challenges.
• The presence of leg electromyography (EMG) channels increases complexity and data storage requirements in sleep studies.

Purpose of the Study:

• To develop and validate a method for detecting periodic leg movement (PLM) during sleep using PSG data, excluding leg EMG channels.
• To assess the efficacy of digital signal processing (DSP) and machine learning techniques for PLM detection without leg EMG.

Main Methods:

• Retrospective analysis of PSG records from 153 patients diagnosed with PLM disorder.
• Development of novel software integrating DSP, statistical methods, and machine learning algorithms.
• Classification of PLM using K-nearest neighbour, multilayer perceptron, random forests, and logistic regression.

Main Results:

• The K-nearest neighbour algorithm achieved the highest average classification rate of 91.87% with a low RMSE of 0.2850.
• Multilayer perceptron showed a lower average classification rate (83.29%) and higher RMSE (0.3705).
• High accuracy in PLM classification was achieved without relying on leg EMG recordings.

Conclusions:

• Periodic leg movement (PLM) can be reliably detected using standard PSG channels, excluding leg EMG.
• This approach simplifies the polysomnography recording process, reduces patient burden, and minimizes data handling issues.
• Machine learning offers a viable solution for accurate PLM detection, improving the efficiency of sleep disorder diagnosis.