Multi-label Classification on 12, 6, 3 and 2 Lead ECG Signals using Convolutional Recurrent Neural Networks

Niels Osnabrugge1, Felix Rustemeyer1, Francesca Battipaglia2, Christos Kaparakis1, Kata Keresztesi1, Joel Karel1, Pietro Bonizzi3
1Maastricht University, 2hearMAASters (Maastricht University), 3Department of Data Science and Knowledge Engineering, Maastricht University


Aim: Automatic identification of cardiac abnormalities through the ECG with a reduced lead system (less than the standard 12-lead) can provide a valuable easy to use and lower cost diagnostic alternative to standard 12-lead ECG devices. This study investigates the use of Convolutional Recurrent Neural Networks (CRNN) to identify cardiac abnormalities in 12, 6, 3 and 2 lead ECG data.

Methods: Multi-label classification with CRNNs relies on effective data pre-processing, model architecture and hyperparameter tuning. ECG signals were first pre-processed and then divided into equal-sized non-overlapping segments of 1000 data points. The class imbalance issue was addressed by a combination of random oversampling and undersampling. Additionally, ECG signals were transformed from the time domain to the frequency domain, and both time and frequency domain representations were concatenated and used as input to a single 2D-Convolutional Neural Network (CNN) consisting of ten layers.

Results: Preliminary results of the proposed method achieved an official score of -0.41 (team name: heartMAASters) and a cross-validation score of -0.40.

Conclusions: The results imply that possible overfitting occurs. Our plan is to address class imbalance and improve performance by employing an ensemble learning approach. In this respect, our plan is to extend the baseline CNN model with a CRNN ensemble model, while keeping both the time domain and frequency domain signals as input.