基于改进U-Net模型的心电波形分割

基于U-Net框架提出一种新的算法用于心电波形的分割。该方法将一定长度的心电信号作为输入，输出P波、QRS波和T波的分割图像，同时定位各个特征波的起始点和终止点，创新性地提出了多通道空洞卷积加上注意力机制的模型结构，并设计了一种数据增强公式用于增加数据的多样性。本研究提出的方法在LUDB上进行训练测试，在QTDB上验证算法的泛化能力。实验结果表明，所提的算法在LUDB的平均灵敏度、平均阳性预测率、平均F1分数分别为99.41%、98.90%、98.75%；在QTDB的平均灵敏度、平均阳性预测率、平均F1分数分别为98.65%、98.43%、98.23%，这说明本文算法效果更好，并具有优异的泛化性能。

ECG waveform segmentation based on improved U-Net model

Abstract: A new algorithm based on U-Net framework is proposed for ECG waveform segmentation, taking the ECG signal of fixed length as the input, and then outputting the images of P wave, QRS wave and T wave. The method can locate the starting and ending points of each characteristic wave. A novel model structure of multi-channel dilated convolution with attention mechanism is put forward, and a data enhancement formula is designed to increase the diversity of data. The proposed method is trained and tested on LUDB, and the generalization ability of the algorithm is verified on QTDB. The experimental results show that the average sensitivity, average positive prediction rate, and average F1 score of the proposed algorithm are 99.41%, 98.90%, 98.75% on LUDB, and 98.65%, 98.43%, 98.23% on QTDB, indicating that the proposed algorithm performs better and has excellent generalization performance.