N-BEATS for Heart Dysfunction Classification

Abstract

Introduction: Recurrent Neural Networks are useful tools for the prediction and classification of ECG problems. The most commonly used network for such a solution is Long Short-Term Memory (LSTM) architecture. This study aims to assess if another state-of-the-art solution, Neural Basis Expansion Analysis for Interpretable Time Series (N-BEATS), can be adopted to diagnose the same cardiac problems. In addition, a comparison is conducted for a different number of electrocardiogram leads. Methods: Two architectures were tested for performance and dimension reduction problems, both in variants consisting of blended branches, allowing retaining accuracy while reducing the computational capacity needed. Results: Our team's (WEAIT) entry was scored incorrectly due to unexpected formatting in outputs; hence only results from cross-validation are presented. LSTM outperforms N-BEATS in terms of multi-label classification, data set resilience, and obtained challenge metrics. Still, N-BEATS can obtain acceptable results and outperforms LSTM in terms of complexity and speed. Conclusions: This paper features a novel approach of using the N-BEATS, which was previously used only for forecasting, to classify ECG signals with success. While N-BEATS multi-label classification capacity is lower than LSTM, its speed allows it to be used on wearable devices.