Multiple Cardiac Disease Detection from Minimal-Lead ECG Combining Feedforward Neural Networks with a One-vs-Rest Approach

Although standard 12-lead ECG is the primary technique in cardiac diagnostic, detecting different cardiac diseases using single or reduced number of leads is still challenging. The purpose of our team, itaca-UPV, is to provide a method able to classify ECG records using minimal lead information in the context of the 2021 PhysioNet/Computing in Cardiology Challenge, also using only a single-lead. We resampled and filtered the ECG signals, and extracted 109 features mostly based on Hearth Rhythm Variability (HRV). Then, we used selected features to train one feed-forward neural network (FFNN) with one hidden layer for each class using a One-vs-Rest approach, thus allowing each ECG to be classified as belonging to none or more than one class. Finally, we performed a 3-fold cross validation to assess the model performance. Our classifiers received scores of 0.34, 0.34, 0.27, 0.30, and 0.34 (ranked 26th, 21th, 29th, 25th, and 22th out of 39 teams) for the 12, 6, 4, 3 and 2-lead versions of the hidden test set with the Challenge evaluation metric. Our minimal-lead approach may be beneficial for novel portable or wearable ECG devices used as screening tools, as it can also detect multiple and concurrent cardiac conditions. Accuracy in detection can be improved adding more disease-specific features.

Automated summary

The team aimed to classify ECG records using minimal lead information and a single-lead approach. By resampling and filtering signals, extracting features based on HRV, training neural networks, and conducting cross-validation, they achieved good performance in detecting cardiac diseases. The method may be beneficial for portable or wearable ECG devices as screening tools.