Block Pyramid Network
包括心电信号预处理模块,一维卷积双交叉臂网络模块,房颤心拍风险提示模块以及外部接口模块。心电信号预处理模块用于对输入的心电信号进行预处理;
一维卷积双交叉臂网络模块用于对双导联特征进行特征融合和选择;
房颤心拍风险提示模块用于将输入的单导联心电信号特征输入到神经网络中,得到房颤心拍的置信度;
外部接口模块用于兼容不同心电采集设备和报警输出接口。本系统还有一种包括上述系统的房颤风险提示设备。
采用本系统的房颤风险提示系统,通过对两个导联的心拍进行融合并抽取特征,实现对双导联心拍进行高质量的房颤检测,有效提高双导联心拍的房颤检测准确率。
竞赛题目:The 4th China Physiological Signal Challenge 2021 (CPSC 2021) aims to encourage the development of algorithms for searching the paroxysmal atrial fibrillation (PAF) events from dynamic ECG recording
网址:(http://2021.icbeb.org/CPSC2021)
- Task I : Three classification tasks for the whole ECG signal
对于整条信号的三分类任务:全部房颤心拍的信号,全部正常心拍的信号,房颤阵法型心拍信号 - Task II : Endpoint recognition tasks for paroxysmal atrial fibrillation signal
对于阵法型房颤心拍产生的端点的识别
数据预处理 Data processing
Two ways for augmentation processing
1:Differential processing
a:The difference is greater
2:The center mirror image processing
a:The difference is greater
b:Reserve the noise component of the original signal
本项目的方案 Game ideas
- Part I : Cut out
- Part II : Sew
Normal:
No five consecutive heart beats are recognized as atrial fibrillation
Atrial Fibrillation:
Not five consecutive heart beats are recognized as normal
Paroxysmal Atrial Fibrillation:
There are five consecutive heart beats with atrial fibrillation and normal
网络模型 Neural network design
Fusion layer makes better use of the two leads data provided by the competition
其中一个模块实例 example network moudle:
系统输出效果演示 output figure:
code link :https://github.com/CEllmemichael/Block-Pyramid-Network-