基于机器学习的个性化睡眠分期方法研究
首发时间:2024-04-26
摘要:睡眠分期是医生诊断患者睡眠障碍和评估患者睡眠质量的基础。传统睡眠分期方法不仅耗时长、过程繁琐,也很难考虑不同患者之间的个体差异。基于机器学习的睡眠分期方法正逐渐引起学术界和医学届的重视。直接应用现有的基于群体数据的机器学习模型于新个体患者数据进行睡眠分期时,分期结果的准确性难以保证。为解决这一问题,本文将基于患者的脑电信号和眼电信号数据开展个性化自动睡眠分期方法的研究。首先,为获取两类信号中不同尺度的特征信息以及重要的上下文信息,本文基于双流卷积网络和自注意力机制构建了睡眠分期自动分类机器学习模型;模型中添加了挤压激励网络,以融合脑电信号和眼电信号中的互补特征。此外,本模型还利用元学习思想先使用群体数据对模型进行预训练,然后使用新个体数据对预训练模型进行微调以得到患者的个性化睡眠分期模型。实验结果表明,本文设计的机器学习模型在 Sleep-EDF 和 ISRUC-S3 数据集上的 准确率分别达到 77.9\% 和 68.7\%,相比于基线模型提高了 1.4\% 和 2.7\%。
关键词: 睡眠分期; 机器学习; 个性化; 脑电信号; 眼电信号
For information in English, please click here
Research on personalized sleep staging method based on machine learning
Abstract:Sleep staging is the basis for doctors to diagnose patients with sleep disorders and evaluate patients' sleep quality. Traditional sleep staging methods are not only time-consuming and complicated, but also difficult to consider individual differences between different patients. The sleep staging method based on machine learning is gradually attracting the attention of the academic and medical circles. When the existing machine learning model based on population data is directly applied to the sleep staging of new individual patient data, the accuracy of staging result is difficult to guarantee. In order to solve this problem, this paper will carry out a personalized automatic sleep staging method based on the electroencephalogram and electrooculogram data of patients. In order to obtain the feature information of different scales and important context information in the two types of signals, this paper constructs a machine learning model for automatic classification of sleep stages based on the two-flow convolutional network and self-attention mechanism. The squeeze excitation network is added to the model to fuse complementary features in electroencephalogram and electrooculogram signals. Experimental result shows that the accuracy of the machine learning model designed in this paper on the Sleep-EDF and ISRUC-S3 datasets reaches 77.9\% and 68.7\%, which is 1.4\% and 2.7\% higher than that of the baseline model.
Keywords: Sleep staging Machine learning Personalization Electroencephalogram Electrooculogram
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于机器学习的个性化睡眠分期方法研究
评论
全部评论0/1000