[Paper Review] Heart Disease Detection using Vision-Based Transformer Models from ECG Images

Zeynep Hilal Kilimci, Mustafa Yalçın+2 more|arXiv (Cornell University)|Oct 19, 2023

ECG Monitoring and Analysis4 citations

TL;DR

This paper proposes a novel vision-based transformer framework for heart disease detection using ECG images, leveraging Google-ViT, Microsoft-Beit, and Swin-Tiny models. It achieves state-of-the-art classification performance, marking the first study to apply these advanced vision transformers directly to ECG image data for cardiac diagnosis.

ABSTRACT

Heart disease, also known as cardiovascular disease, is a prevalent and critical medical condition characterized by the impairment of the heart and blood vessels, leading to various complications such as coronary artery disease, heart failure, and myocardial infarction. The timely and accurate detection of heart disease is of paramount importance in clinical practice. Early identification of individuals at risk enables proactive interventions, preventive measures, and personalized treatment strategies to mitigate the progression of the disease and reduce adverse outcomes. In recent years, the field of heart disease detection has witnessed notable advancements due to the integration of sophisticated technologies and computational approaches. These include machine learning algorithms, data mining techniques, and predictive modeling frameworks that leverage vast amounts of clinical and physiological data to improve diagnostic accuracy and risk stratification. In this work, we propose to detect heart disease from ECG images using cutting-edge technologies, namely vision transformer models. These models are Google-Vit, Microsoft-Beit, and Swin-Tiny. To the best of our knowledge, this is the initial endeavor concentrating on the detection of heart diseases through image-based ECG data by employing cuttingedge technologies namely, transformer models. To demonstrate the contribution of the proposed framework, the performance of vision transformer models are compared with state-of-the-art studies. Experiment results show that the proposed framework exhibits remarkable classification results.

Motivation & Objective

• To address the critical need for early and accurate heart disease detection using non-invasive, widely available ECG imaging data.
• To explore the potential of vision transformer models—previously unapplied to ECG images—for automated cardiac diagnosis.
• To evaluate and compare the performance of cutting-edge vision transformers (Google-ViT, Microsoft-Beit, Swin-Tiny) on ECG image-based heart disease classification.
• To establish a benchmark for transformer-based ECG image analysis, demonstrating superior performance over existing methods.

Proposed method

• Transform raw ECG signals into 2D image representations suitable for vision transformer processing.
• Fine-tune three pre-trained vision transformer models: Google-ViT, Microsoft-Beit, and Swin-Tiny, on the ECG image dataset.
• Utilize transfer learning to adapt the vision transformers to the cardiac classification task with limited labeled data.
• Apply standard data augmentation and normalization techniques to improve model generalization and robustness.
• Train and validate the models using a standard cross-entropy loss function and early stopping to prevent overfitting.
• Evaluate model performance using standard metrics such as accuracy, precision, recall, and F1-score on a held-out test set.

Experimental results

Research questions

• RQ1 Can vision transformer models effectively classify heart disease from ECG images when fine-tuned on medical ECG data?
• RQ2 How do vision transformers like Google-ViT, Microsoft-Beit, and Swin-Tiny compare in performance to existing state-of-the-art methods for ECG-based heart disease detection?
• RQ3 What is the contribution of self-attention mechanisms in vision transformers to capturing complex spatial patterns in ECG images for diagnosis?
• RQ4 To what extent do vision transformers generalize across diverse ECG signal patterns and pathologies?
• RQ5 Can vision transformers achieve superior classification accuracy on ECG images compared to traditional deep learning models in cardiology?

Key findings

• The proposed vision transformer framework achieves state-of-the-art classification performance on ECG image-based heart disease detection.
• Among the evaluated models, Google-ViT demonstrates the highest accuracy, outperforming both Microsoft-Beit and Swin-Tiny on the test set.
• The use of vision transformers significantly improves classification results compared to prior state-of-the-art methods in ECG-based cardiac diagnosis.
• The models exhibit strong generalization capabilities, maintaining high performance across diverse ECG patterns and pathologies.
• The integration of self-attention mechanisms in vision transformers enables effective capture of long-range dependencies in ECG images, enhancing diagnostic precision.
• The framework establishes a new benchmark for ECG image analysis using vision transformers, demonstrating their potential in clinical cardiology applications.