ECG Arrhythmia Detection Using Disease-specific Attention-based Deep Learning Model (2407.18033v1)

Abstract: The electrocardiogram (ECG) is one of the most commonly-used tools to diagnose cardiovascular disease in clinical practice. Although deep learning models have achieved very impressive success in the field of automatic ECG analysis, they often lack model interpretability that is significantly important in the healthcare applications. To this end, many schemes such as general-purpose attention mechanism, Grad-CAM technique and ECG knowledge graph were proposed to be integrated with deep learning models. However, they either result in decreased classification performance or do not consist with the one in cardiologists' mind when interpreting ECG. In this study, we propose a novel disease-specific attention-based deep learning model (DANet) for arrhythmia detection from short ECG recordings. The novel idea is to introduce a soft-coding or hard-coding waveform enhanced module into existing deep neural networks, which amends original ECG signals with the guidance of the rule for diagnosis of a given disease type before being fed into the classification module. For the soft-coding DANet, we also develop a learning framework combining self-supervised pre-training with two-stage supervised training. To verify the effectiveness of our proposed DANet, we applied it to the problem of atrial premature contraction detection and the experimental results shows that it demonstrates superior performance compared to the benchmark model. Moreover, it also provides the waveform regions that deserve special attention in the model's decision-making process, allowing it to be a medical diagnostic assistant for physicians.

• The paper introduces a novel disease-specific attention mechanism that improves ECG arrhythmia detection by emphasizing clinically relevant waveform regions.
• It employs a two-pronged approach using soft-coding and hard-coding attention strategies to balance interpretability with computational efficiency.
• Experimental results on the Aliyun-Tianchi ECG Challenge Database demonstrated a 91.72% accuracy, outperforming baseline CNN models.

Disease-specific Attention-based Deep Learning Model for ECG Arrhythmia Detection

The paper introduces a novel approach to arrhythmia detection using a disease-specific attention-based deep learning model (DANet), specifically contextualized for ECG signal interpretation. Unlike existing frameworks that predominantly leverage generic attention mechanisms, DANet employs domain-specific insights to guide its learning process, thereby addressing the challenges of model interpretability and arrhythmia detection accuracy.

Central to the paper is the argument that traditional deep learning models inadequately distinguish between different waveform regions in ECG signals, such as the P wave, QRS complex, and T wave, which are essential for accurate arrhythmia diagnosis. This oversight often results in suboptimal focus on clinically relevant ECG components, particularly problematic in conditions like atrial premature contraction (APC) where the P wave carries specific diagnostic value. The proposed model incorporates a disease-specific attention mechanism which aligns the automatic attention weights with established cardiological diagnostic criteria, enhancing both precision and interpretability.

The research outlines a two-pronged framework for ECG analysis: the soft-coding and hard-coding attention mechanisms. The soft-coding approach employs a self-supervised pre-training mechanism to allow existing deep neural networks to amend ECG signals before classification. This effectively simulates the cognitive diagnostic processes of cardiologists, ensuring the model's focus on relevant ECG features. In this framework, the feature extraction process is facilitated by a dilated convolutional network, enabling the identification of critical waveform regions without unduly escalating computational complexity.

Alternatively, the hard-coding approach, termed DANet-h, simplifies computational demands by directly utilizing traditional fiducial methods like ECGPuWave to predefine attention weights. Although this method eschews deeper data-driven learning for efficiency, it mandates external tools for attention weight assignment and lacks adaptive learning capabilities provided by the deep learning model in the soft-coding approach.

In the experimental setup, the proposed models were evaluated on the Aliyun-Tianchi ECG Challenge Database for APC detection. The findings demonstrated improved performance over baseline CNN models, particularly in specificity and accuracy. Specifically, the stage-3 DANet attained an accuracy rate of 91.72% with improved F-scores, underscoring the efficacy of disease-specific attention mechanisms in refining diagnostic processes for arrhythmias linked to non-QRS complexes.

The introduction of disease-specific attention in deep learning models confers significant theoretical and practical enhancements, notably in aligning model interpretability with clinical intuition. Automatic attention weights provide medical insights into the model's decision-making process, vital for clinical adoption. Future adaptations of these models could encompass multivariate arrhythmia detection, integrating broader disease frameworks to capitalize on cross-pathological diagnostic insights. The paper serves as a foundational leap towards blending computational rigor with clinical interpretability in ECG analysis, fostering more robust AI tools in healthcare diagnostics.