Skip to content

Latest commit

 

History

History
32 lines (31 loc) · 1.48 KB

readme.md

File metadata and controls

32 lines (31 loc) · 1.48 KB
Raw

ECG R-wave and P-wave localization in paper:

@InProceedings{Abrishami2018,
  author    = {H. {Abrishami} and M. {Campbell} and C. {Han} and R. {Czosek} and X. {Zhou}},
  title     = {P-{QRS}-T localization in {ECG} using deep learning},
  booktitle = {Proc. IEEE EMBS Int. Conf. Biomedical Health Informatics (BHI)},
  year      = {2018},
  pages     = {210--213},
  month     = mar,
  doi       = {10.1109/BHI.2018.8333406}
}

Since the code of this paper is not open, I implemented the code according this paper with keras framework.

Data preprocess

Data preprocessed in MATLAB. Download data files from https://www.physionet.org/content/qtdb/1.0.0/ with download_QTDB.m. PC will get xxxann.mat for Y and xxxdata.mat for X.
For input data to keras conveniently, Segmentor.m will segment all recording into complexes and position of P-wave and R-wave is also saved in segmentors.mat.
if you load segmentor.mat into matlab. You will get segs with 96863 by 300 and anns with dimention of 96863 by 2 in workspace. That mean there are 96863 complexes with length of 300 sampling points.
ann[:,1] presents position of P-wave. ann[:,2] presents position of R-wave. More detail can be found in paper.

models

for fully-connected net usage:

    python ./paper_models_codes/denseNet_P_R_localization.py

for 1D CNN usage:

    python ./paper_models_codes/ECGNet.py

for 1D CNN with dropout usage:

    python ./paper_models_codes/ECGNet_Dropout.py