A Personalized Zero-Shot ECG Arrhythmia Monitoring System: From Sparse Representation Based Domain Adaption to Energy Efficient Abnormal Beat Detection for Practical ECG Surveillance

TL;DR

The paper tackles personalized zero-shot arrhythmia detection for wearable ECGs by integrating sparse representation with null-space projections, domain adaptation via morphology transformation matrices, and an ensemble CNN with probabilistic gating. Key contributions include a fast left-null-space classifier (NPE/LAE), a sparse representation–based domain adaptation (SR-DA) to transfer normal/abnormal patterns across users, and an energy-efficient monitoring scheme that classifies many normal beats with lightweight computation. The approach achieves high accuracy and F1-scores on MIT-BIH data (e.g., up to $98.2\%$ accuracy and $92.8\%$ F1) while substantially reducing energy use (up to $40\%$ of beats processed with low-cost methods). This work enables practical, privacy-conscious, personalized ECG surveillance on wearables with strong performance and reduced power consumption.

Abstract

This paper proposes a low-cost and highly accurate ECG-monitoring system intended for personalized early arrhythmia detection for wearable mobile sensors. Earlier supervised approaches for personalized ECG monitoring require both abnormal and normal heartbeats for the training of the dedicated classifier. However, in a real-world scenario where the personalized algorithm is embedded in a wearable device, such training data is not available for healthy people with no cardiac disorder history. In this study, (i) we propose a null space analysis on the healthy signal space obtained via sparse dictionary learning, and investigate how a simple null space projection or alternatively regularized least squares-based classification methods can reduce the computational complexity, without sacrificing the detection accuracy, when compared to sparse representation-based classification. (ii) Then we introduce a sparse representation-based domain adaptation technique in order to project other existing users' abnormal and normal signals onto the new user's signal space, enabling us to train the dedicated classifier without having any abnormal heartbeat of the new user. Therefore, zero-shot learning can be achieved without the need for synthetic abnormal heartbeat generation. An extensive set of experiments performed on the benchmark MIT-BIH ECG dataset shows that when this domain adaptation-based training data generator is used with a simple 1-D CNN classifier, the method outperforms the prior work by a significant margin. (iii) Then, by combining (i) and (ii), we propose an ensemble classifier that further improves the performance. This approach for zero-shot arrhythmia detection achieves an average accuracy level of 98.2% and an F1-Score of 92.8%. Finally, a personalized energy-efficient ECG monitoring scheme is proposed using the above-mentioned innovations.

Figures (12)

• Figure 1: The schema of ABS kiranyaz2017personalized: The estimated linear degrading system of the existing user $j$ can be applied to the healthy ECG beats of a new user $p$ in order to synthesize possible abnormal beats for user $p$.
• Figure 2: Example of normal, S-type, and V-type ECG signals from patient 100, and their sparse approximation errors. From \ref{['fig:beat_types_and_errors']}d it is clear that both types of abnormal beats have noticeably large SAE compared to a normal beat.
• Figure 3: Sparse codes calculated through $\ell_1$ and $\ell_2$ minimization for a normal ECG beat and a dictionary with $n=20$ atoms.
• Figure 4: Linear Morphology Transformation System.
• Figure 5: The histogram of sparse approximation errors of normal beats of users $\bm{p}$ and $\bm{l}$ on user $\bm{p}$'s dictionary, $\bm{D^p}$. $\bm{\widetilde{e}_{\ell_1}}$ for $\bm{l}$ is high even for normal beats on $\bm{D^p}$ (upper), whereas after domain adaptation the error energy more closely resembles the ones of $\bm{p}$ (lower).