SViT-ECG: Spectrogram Vision Transformer for Detection of Short-Term Atrial Fibrillation from ECG Signals

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Atrial fibrillation (AF) is a common arrhythmia with serious complications like stroke and heart failure.
• Early and accurate diagnosis of AF is crucial for timely intervention and risk reduction.

Purpose of the Study:

• To develop a novel deep learning approach for detecting short atrial fibrillation episodes in ECG signals.
• To evaluate the performance of a transformer-based model for AF detection.

Main Methods:

• ECG segments were converted into spectrograms.
• A fine-tuned Vision Transformer model (SViT-ECG) was utilized for classification.
• The model underwent 5-fold cross-validation and validation on an unseen dataset.

Main Results:

• The SViT-ECG model achieved 98.14% accuracy and 95.81% F1-score during cross-validation.
• On an unseen dataset, the model obtained 95.97% accuracy and 91.14% F1-score.
• The proposed method demonstrates high efficacy in identifying AF episodes.

Conclusions:

• The SViT-ECG model shows significant potential for real-time atrial fibrillation detection.
• This deep learning approach represents an advancement over existing state-of-the-art algorithms.
• The findings support the clinical utility of AI in improving AF diagnosis and management.