FADE: Forecasting for Anomaly Detection on ECG

TL;DR

FADE tackles the challenge of detecting ECG anomalies without relying on large labeled datasets by forecasting the normal ECG trajectory using a self-supervised SlowFast–U-Net architecture with a novel Split-MSE morphological loss. An anomaly is detected by comparing the forecasted normal signal to the actual measurement via the NMAE distance, with domain adaptation enabling transfer to new data sources. The approach achieves strong anomaly and normal-signal classification performance on MIT-BIH NSR and Arrhythmia, and ablation analyses show the combined slow and fast paths plus Split-MSE are key to performance. The work suggests practical viability for scalable, remote ECG monitoring and sets the stage for wearables deployment and future anomaly-type classification.

Abstract

Cardiovascular diseases, a leading cause of noncommunicable disease-related deaths, require early and accurate detection to improve patient outcomes. Taking advantage of advances in machine learning and deep learning, multiple approaches have been proposed in the literature to address the challenge of detecting ECG anomalies. Typically, these methods are based on the manual interpretation of ECG signals, which is time consuming and depends on the expertise of healthcare professionals. The objective of this work is to propose a deep learning system, FADE, designed for normal ECG forecasting and anomaly detection, which reduces the need for extensive labeled datasets and manual interpretation. FADE has been trained in a self-supervised manner with a novel morphological inspired loss function. Unlike conventional models that learn from labeled anomalous ECG waveforms, our approach predicts the future of normal ECG signals, thus avoiding the need for extensive labeled datasets. Using a novel distance function to compare forecasted ECG signals with actual sensor data, our method effectively identifies cardiac anomalies. Additionally, this approach can be adapted to new contexts through domain adaptation techniques. To evaluate our proposal, we performed a set of experiments using two publicly available datasets: MIT-BIH NSR and MIT-BIH Arrythmia. The results demonstrate that our system achieves an average accuracy of 83.84% in anomaly detection, while correctly classifying normal ECG signals with an accuracy of 85.46%. Our proposed approach exhibited superior performance in the early detection of cardiac anomalies in ECG signals, surpassing previous methods that predominantly identify a limited range of anomalies. FADE effectively detects both abnormal heartbeats and arrhythmias, offering significant advantages in healthcare through cost reduction or processing of large-scale ECG data.

Figures (8)

• Figure 1: Proposed System. General functionality of our proposed system for ECG forecasting with domain adaptation and heart anomaly detection.
• Figure 2: ECG signal slicing. ECG slices for training the forecasting model: $I$ represents the input data, $L$ is the label data and $AL$ shows the auxiliary label data.
• Figure 3: Proposed Model. The proposed model for ECG forecasting that combines both SlowFast and U-Net networks is shown. The U-Net based decoder is on the left side (blue). The Slow path can be observed in the middle (color pink), while the Fast path is on the right side (green). Note that orange arrows indicate concatenation operation and black arrows show connections between layers. More details in the main text.
• Figure 4: Split-MSE Loss Function. MSE is calculated separately in two different ranges: outer bands, designed to capture the R-peaks, and inner band, which seeks to seize the segment between R-peaks. Note that P represents P-wave (atrial depolarization), QRS complex represents the depolarization of the ventricles and T represents T-wave (ventricular repolarization).
• Figure 5: Domain Adaptation. For a given sample of a new domain, the top row shows the prediction (unseen temporal segment) using the original model, while the bottom row represents the prediction employing the domain-adapted model. Huge differences can be appreciated in the unseen part of the signal. More details in the main text.