Convolutional neural network and rule-based algorithms for classifying 12-lead ECGs

What is it about?

This study is part of the PhysioNet/Computation in Cardiology (CinC) Challenge 2020. Our objective was to classify 27 cardiac abnormalities based on a provided dataset of 43101 ECG recordings. Wedevelopedahybridmodelcombiningarule-basedalgorithmwithdifferentDeepLearningarchitectures. We compared two different Convolutional Neural Networks (FCN and Encoder), a combination of both, and with the addition of another neural network. Two of these combinations were finally combined with a rule-based model using derived ECG features. We evaluated the performance of the models on validation data during model development using hold-out validation. Finally, we deployed the models to a Docker image, trained the model on the provided development data before the models were tested on a hidden dataset, giving a performance based on a particular Physionet Challenge score. Our team, TeamUIO, found that the FCN in parallel with an Encoder without any rule-based model performed best on the validation data with a score of 0.412. However, the best score on the test set was the Encoder in parallel with a Fully Convolutional Network with the rule-based model added, receiving a score of 0.377.

Featured Image

Photo by Towfiqu barbhuiya on Unsplash

Photo by Towfiqu barbhuiya on Unsplash

Why is it important?

The study is part of a larger challenge, the PhysioNet/Computation in Cardiology Challenge 2020. Participating in such challenges encourages the development and evaluation of advanced techniques in the field of computational cardiology. The models are tested on a hidden dataset, providing an unbiased evaluation of their performance. This step is important for assessing the generalization capabilities of the developed models beyond the training and validation datasets.