| 韩传奇* **,崔莉*.基于压缩卷积神经网络的心律不齐分类方法[J].高技术通讯(中文),2023,33(9):895~904 |
| 基于压缩卷积神经网络的心律不齐分类方法 |
| An arrhythmia classification method based on compressed convolutional neural network |
| |
| DOI：10. 3772/ j. issn. 1002-0470. 2023. 09. 001 |
| 中文关键词: 可穿戴设备； 心律不齐分类； 压缩卷积神经网络； 类别不平衡； 损失函数 |
| 英文关键词: wearable device, arrhythmia classification, compressed convolutional neural network, class imbalance, loss function |
| 基金项目: |
| 作者 | 单位 | | 韩传奇* ** | （*中国科学院计算技术研究所北京 100190）
（**中国科学院大学北京 100190） | | 崔莉* | |
|
| 摘要点击次数: 665 |
| 全文下载次数: 548 |
| 中文摘要: |
| 心律不齐是一种常见的心脏疾病，严重时可能会危及生命，因此对该疾病开展早期筛查和分类在临床医学中具有重要意义。搭载心电信号（ECG）传感器的可穿戴设备凭借低成本和便捷等特点，是实现日常心脏健康监测的理想平台之一。然而受制于计算能力等因素的限制，可穿戴设备需要将数据上传到云端进行分析，增加了等待时延和用户隐私泄露风险。另一方面，现有心律不齐分类算法在训练时受疾病样本分布不平衡等因素的影响，在识别部分异常病症时的表现不尽人意，限制了其应用范围。为解决上述问题，本文提出了一种基于压缩卷积神经网络的心律不齐分类算法，增强了其在移动平台上的部署能力。同时在训练过程中通过将类别先验分布引入损失函数中，提升了算法对异常病症的识别能力。实验结果表明，本文提出的压缩模型相比经典模型在减少98.2%参数量的同时，超越了许多相关工作取得了0.759的宏F1值。 |
| 英文摘要: |
| Arrhythmia is a common heart disease, which may be life threatening in severe cases. Therefore, early screening and classification of the disease are of great significance in clinical medicine. Wearable devices equipped with electrocardiogram (ECG) sensors are one of the ideal platforms for heart condition monitoring with the characteristics of low cost and convenience. However, due to the limitation of computing power and other factors, wearable devices need to upload data to the cloud for further analysis, which increases the waiting delay and the risk of user privacy disclosure. On the other hand, the imbalanced class distribution could weaken the existing arrhythmia classification algorithm in identifying certain abnormal diseases and limit its application scenarios. To solve the above problems, this paper proposes an arrhythmia classification algorithm based on a compressed convolution neural network, which enhances its deployment capability on mobile platforms. At the same time, by introducing the category prior distribution into the loss function, the algorithm’s ability to identify abnormal symptoms is improved. Experimental results show that the proposed algorithm reduces the number of parameters by 98.2% compared with the classical one and still surpasses many related works with a macro F1 value of 0.759. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
