Counting 2,129 Big Data & Machine Learning Frameworks, Toolsets, and Examples...
Suggestion? Feedback? Tweet @stkim1

Last Commit
Dec. 11, 2017
Nov. 5, 2017


CardIO is designed to build end-to-end machine learning models for deep research of electrocardiograms.

Main features:

  • load and save signal in various formats (wfdb, blosc, etc)
  • resample, crop, flip and filter signals
  • detect PQ, QT, QRS segments
  • calculate heart rate and other ECG characteristics
  • apply complex transformations like fft and wavelets, as well as custom functions
  • recognize heart diseases (e.g. atrial fibrillation)
  • efficiently work with large datasets that do not even fit into memory
  • perform end-to-end ECG processing
  • build, train and test neural networks and other machine learning models.

For more details see the documentation and tutorials.

About CardIO

CardIO is based on Dataset. You might benefit from reading its documentation. However, it is not required, especially at the beginning.

CardIO has three modules: batch, models and pipelines.

batch module contains EcgBatch class which defines how ECG are stored and includes actions for ECG processing. These actions might be used to build multi-staged workflows that can also involve machine learning models.

models module provides several ready to use models for important problems in ECG analysis:

  • how to detect specific features of ECG like R-peaks, P-wave, T-wave, etc;
  • how to recognize heart diseases from ECG, for example, atrial fibrillation.

pipelines module contains predefined workflows to

  • train a model to detect PQ, QT, QRS segments
  • calculate heart rate
  • train a model to find probabilities of heart diseases, in particular, atrial fibrillation.

Under the hood these methods contain actions that load signals, filter it and do complex calculations.

Basic usage

Here is an example of pipeline that loads ECG signals, makes preprocessing and train a model over 50 epochs:

train_pipeline = (
        .init_model("dynamic", DirichletModel, name="dirichlet",
        .init_variable("loss_history", init=list)
        .load(components=["signal", "meta"], fmt="wfdb")
        .load(components="target", fmt="csv", src=LABELS_PATH)
        .replace_labels({"N": "NO", "O": "NO"})
        .random_resample_signals("normal", loc=300, scale=10)
        .random_split_signals(2048, {"A": 9, "NO": 3})
        .train_model("dirichlet", make_data=make_data, fetches="loss", save_to=V("loss_history"), mode="a")
        .run(batch_size=100, shuffle=True, drop_last=True, n_epochs=50)


CardIO module is in the beta stage. Your suggestions and improvements are very welcome.

CardIO supports python 3.5 or higher.

Installation as a python package

With pipenv:

pipenv install git+

With pip:

pip3 install git+

After that just import cardio:

import cardio

Installation as a project repository:

When cloning repo from GitHub use flag --recursive to make sure that Dataset submodule is also cloned.

git clone --recursive

Citing CardIO

Please cite CardIO in your publications if it helps your research.

Khudorozhkov R., Illarionov E., Kuvaev A., Podvyaznikov D. CardIO library for data science research of heart signals. 2017.

Latest Releases
Initial release
 Nov. 22 2017