A Method of Data Augmentation for Shunt Murmur Angiostenosis Detection
A Method of Data Augmentation for Shunt Murmur Angiostenosis Detection
复制标题
分流杂音血管狭窄检测的数据增强方法
DOI：
10.1007/978-3-030-50454-0_58
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
Furuya Ken’ichi
中科院分区：
文献类型：
--
作者：
Nishijima Keisuke;Higashi Daisuke;Furuya Ken’ichi
由于血液透析患者年龄的增长，出现了血管狭窄、闭塞等问题。因此，患者必须每天检查分流功能。然而，分析分流杂音是困难的，因为存在个体差异，判断分流功能需要经验。本研究旨在通过分析分流杂音来自动识别分流功能。在传统的方法中，分流函数是通过支持向量机或随机森林识别的。然而，由于这些研究没有考虑第二类误差，因此识别精度较低。因此，我们的目标是通过改进整个识别系统来提高识别精度，该系统包括以下三个部分：数据生成部分、特征计算部分和识别部分。为了改进数据创建部分，我们考虑分割一个对应于一个脉冲的分流杂音。为了改进特征计算过程，我们考虑了使用样本移动和扩张收缩过程来增强数据。结果表明，该方法提高了识别精度。
Problems such as stenosis and occlusion are observed owing to the aging of the hemodialysis patients. Therefore, patients must check shunt function daily. However, analyzing shunt murmurs is difficult because there are individual differences and judging the shunt function requires experience. This study aims to automatically identify shunt function by analyzing shunt murmurs. In conventional methods, the shunt function is identified via support vector machine or random forest. However, the identification accuracy is low as these studies do not consider error of the second kind. Therefore, we aim to enhance identification accuracy by improving the entire identification system, which has the following three parts: data-creation part, feature-calculation and identification parts. To improve the data-creation part, we consider segmenting a shunt murmur that corresponds to one pulse. To improve the feature-calculation process, we consider data augmentation by using sample-movement and expansion-and-contraction processes. The results confirm that the identification accuracy was improved.