用于心電疾病診斷的深度學習模型庫
github: https://github.com/hzzhangqf0558/ECG_Nets
Baseline model collection of deep learning applied into ECGs. Those baseline models include 1D-ResNet, 1D-DenseNet, 1D-SE_ResNet, 1D-ResNext,1D-SE_ResNetV2, 1D-SE_ResNext and 1D-Top1Net(the champion model in Tianchi competition).
The F1-scores of those baseline models are in range of 0.83-0.90. you could fine-tune the parameters to reach a better level.
Step 1:
downloading datasets. The datasets from Tianchi competition includes dataset A and dataset B. The dataset A consists of 24106 ECGs, of which each has 8 leading records(I，II，V1，V2，V3，V4，V5, V6). while the dataset B consists of 20036 ECGs. Dataset can be downloaded from website.
https://pan.baidu.com/s/1fmCuV5i9oifnUNOsFhV0sA pwd: 8hs2
Please put the datasets into the fold: all_data.
Finally, unzip the data packages.
step 2: build environment
python 3.7 is required. And the requirements can be used directly.
pip install -r requirements.txt
step 3: create csv file.
config the dataset path in the data_preparing.py. Then
python data_preparing.py
step 4: choose a baseline model you like.
7 models are presented in the models fold and configs fold. Those
python main.py --config configs/ResNet50.yaml
if you want ensemble, please config the file configs/ensemble.
python ensemble.py --config configs/ensemble.yaml
Reference:
https://tianchi.aliyun.com/competition/entrance/231754/information
https://www.hindawi.com/journals/cmmm/2019/7095137/
https://ieeexplore.ieee.org/abstract/document/9113436
轉載請註明出處，本文鏈接：https://www.uj5u.com/houduan/147923.html
標籤：python
上一篇：使用(VAE)生成建模,理解可變自動編碼器背后的數學原理
下一篇：推薦 5 款牛逼的代碼編輯器