Automated heart disease detection using Swin Transformer and ECG signal processing: a high-accuracy approach.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Cardiovascular diseases (CVDs) are a primary cause of global mortality, underscoring the need for early and accurate detection.
• Electrocardiography (ECG) is crucial for diagnosing cardiac conditions, but traditional interpretation and machine learning (ML) methods face limitations in feature extraction and handling long-range dependencies.
• Deep learning (DL) advancements, particularly transformer architectures, show promise for enhanced ECG classification.

Purpose of the Study:

• To propose and evaluate the Swin Transformer, a hierarchical vision transformer, for automated ECG-based heart disease detection.
• To compare the Swin Transformer's performance against conventional ML models (Random Forest, Gradient Boosting, SVM, Neural Networks).

Main Methods:

• Utilized a Swin Transformer model with shifted window self-attention mechanisms to capture local and global ECG dependencies.
• Evaluated the model on benchmark ECG datasets.
• Compared performance metrics including accuracy, precision, recall, and AUC against established ML algorithms.

Main Results:

• The Swin Transformer achieved superior performance, reaching 99.8% accuracy, 99.72% precision, 99.91% recall, and 99.99% AUC.
• Demonstrated significant outperformance compared to Random Forest, Gradient Boosting, SVM, and Neural Networks.
• Eigenvalue analysis confirmed the model's ability to retain essential features and generalize across diverse ECG patterns.

Conclusions:

• The Swin Transformer presents a highly effective, scalable, and clinically viable solution for automated ECG-based cardiac disease detection.
• This DL approach establishes a new benchmark in ECG classification accuracy and reliability over traditional methods.
• Future research should address computational complexity and interpretability using Explainable AI (XAI) and explore real-time optimization.