# CHDdECG
Scripts and modules for training and testing deep neural networks that conducts congenital heart disease detection using electrocardiogram (ECG).
Companion code to the paper "Congenital Heart Disease Detection by Pediatric Electrocardiogram Based Deep Learning Integrated with Human Concepts".
<!-- https://www.xxxxxxxxx.com/
--------------------
Citation:
```
Authors et al. Automatic Diagnosis of Congenital Heart Diseases from 9-Lead Pediatric Electrocardiogram Using Deep Neural Networks.
journal. doi
```
Bibtex:
```
@article{,
title = {Automatic Diagnosis of Congenital Heart Diseases from 9-Lead Pediatric Electrocardiogram Using Deep Neural Networks},
author = {},
year = {},
volume = {},
pages = {},
doi = {},
journal = {},
number = {}
}
```
-------------------- -->
## Abstract information
Early detection is critical to achieving improved treatment outcomes for child patients with congenital heart diseases (CHDs). Therefore, developing effective CHD detection techniques using low-cost and non-invasive pediatric electrocardiogram are highly desirable. We propose a deep learning approach for CHD detection, CHDdECG, which automatically extracts features from pediatric electrocardiogram signals and wavelet transformation characteristics, integrating them with human concept features. Trained with 65,869 cases, CHDdECG achieved ROC-AUC of 0.915 and 0.917 on real-world and external test sets containing 12,000 and 7,137 cases, respectively. Overall specificities were 0.881 and 0.937, respectively, with major CHD subtype specificities above 0.91. Notably, CHDdECG outperformed cardiologists in comparison, and the automatically extracted features from electrocardiogram were about 8 times more significant than human concept features, implying that CHDdECG find some knowledge beyond human cognition. Our study directly impacts CHD detection with pediatric electrocardiogram and highlights the potential of pediatric electrocardiogram for broader benefits.
--------------------
## Requirements
This code was tested on Python 3 with Tensorflow `2.6.0`
In addition, the packages we are calling now is as follows:
- [x] tensorflow2.0
- [x] sklearn
- [x] random
- [x] scipy
- [x] pandas
- [x] numpy
- [x] tabnet
- [x] tensorflow_addons
## Framework illustration
- **input**: `shape = (N, 5000, 9)`. The input tensor, a signal of 10 seconds should contain the 5000 points of the ECG tracings sampled at 500Hz both in the training and in the test set. The last dimension of the tensor contains points of the 9 different leads. The leads are ordered in the following order: `{I, II, III, AVR, AVL, AVF, V1, V3, V5}`. All signal are preprocessed with noise removal techniques before feeding it to the neural network model.
![example](https://github.com/shuaih720/CHDdECG/blob/main/Figures/ECG%20example.png)
- **framework illustration**: ``CHDdECG.py``: Auxiliary module that defines the architecture of the deep neural network. The internal module structure is in the following files:``layers_Resblock.py``,``layers_TemporalAttention.py``,``layers_Transformer.py``,``layers_Tabnet.py``.
![example1](https://github.com/shuaih720/CHDdECG/blob/main/Figures/An%20illustration%20of%20the%20deep%20learning%20based%20model.png)
- **train and test**: ``main.py``:Script for training the neural network and generating the neural network predictions on a given dataset.
- **output**: `shape = (N, 2)`. Each entry contains a probability between 0 and 1, and can be understood as the probability of a given abnormality to be present.
## Install from Github
```python
python
>>> git clone https://github.com/shuaih720/CHDdECG
>>> cd CHDdECG
>>> python setup.py install
```
(Typical install time on a "normal" desktop computer: very variable)
## Instructions for use
```python
python
>>> cd CHDdECG
>>> python import.py
>>> python layers_Tabnet.py
>>> python layers_Resblock.py
>>> python layers_TemporalAttention.py
>>> python layers_Transformer.py
>>> python CHDdECG.py
>>> python main.py
```
OR running the integrated version
```python
python
>>> cd CHDdECG
>>> python Merged_CHD.py
```
Training the neural network and generating the neural network predictions on given datasets.
## License
This project is covered under the Apache 2.0 License.
深度神经网络进行先天性心脏病(CHD)检测.zip
版权申诉
125 浏览量
2024-05-10
23:20:09
上传
评论
收藏 127.36MB ZIP 举报
小码蚁.
- 粉丝: 2584
- 资源: 4344
最新资源
- tomcat安装及配置教程.md
- 51单片机贴片心形灯原理图PCB程序 亚克力外壳.dwg资料
- 户型为单家独院式图纸编号D068-三层-20.60&13.40米-施工图.dwg
- 农村小别墅图纸编号D067-三层-10.00&11.50米- 施工图.dwg
- 三层独栋别墅图纸编号D066-三层-12.00&12.00米- 施工图.dwg
- 农村小别墅图纸编号D065-三层-14.40&18.55米-施工图.dwg
- 基于YOLOv8检测高铁吊弦缺陷实现的系统的Python源码+文档说明+训练源文件+模型.zip
- 三层农村小别墅图纸编号D064-三层-13.80&22.20米-施工图.dwg
- 三层别墅图纸编号D063-三层-13.57&17.40米- 施工图.dwg
- STC IAP15F2K61S2单片机i2c 接口PCF8591-ADC实验+DAC实验 KEIL例程源码+开发板硬件原理图
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈