Jain Ritika, Ganesan Ramakrishnan Angarai
IEEE J Biomed Health Inform. 2025 Apr;29(4):2581-2588. doi: 10.1109/JBHI.2024.3524079. Epub 2025 Apr 4.
A novel method is proposed for diagnosing the sleep disorders of insomnia, narcolepsy, periodic leg movement syndrome, nocturnal frontal lobe epilepsy, rapid eye movement behavior disorder, and sleep-disordered breathing. We use the light gradient boosting decision tree model (LightGBM) for the classification of healthy controls and different sleep disorders. The proposed approach is evaluated on the publicly available CAP dataset of 108 subjects. The LightGBM classifier using only an electrooculogram (EOG) channel (L-EOG) achieves a seven-class classification accuracy of 83.3%. This performance improves to 93.3% with the LightGBM-EOG-EEG (LEE) classifier, which harnesses the combined strengths of EOG and electroencephalogram (EEG) channels. LEE classifier employs four binary classifiers that disambiguate the classes confused by the L-EOG classifier. An additional 1% improvement is achieved by applying a simple thresholding decision rule to the results of the LEE classifier, resulting in an overall accuracy of 94.4%, the best in the literature for the seven-class classification of sleep disorders.
提出了一种用于诊断失眠、发作性睡病、周期性腿部运动综合征、夜间额叶癫痫、快速眼动睡眠行为障碍和睡眠呼吸紊乱等睡眠障碍的新方法。我们使用轻量级梯度提升决策树模型（LightGBM）对健康对照者和不同睡眠障碍进行分类。在公开可用的108名受试者的CAP数据集上对所提出的方法进行了评估。仅使用眼电图（EOG）通道的LightGBM分类器（L-EOG）实现了七分类准确率为83.3%。通过利用EOG和脑电图（EEG）通道的综合优势的LightGBM-EOG-EEG（LEE）分类器，该性能提高到了93.3%。LEE分类器采用四个二分类器来区分被L-EOG分类器混淆的类别。通过对LEE分类器的结果应用简单的阈值决策规则，又实现了1%的提升，从而使总体准确率达到94.4%，这是睡眠障碍七分类文献中的最佳结果。
