Automated Atrial Fibrillation Detection with ECG.

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Electrocardiography (ECG) systems record the heart's electrical activity.
• Cardiac arrhythmia, such as atrial fibrillation, requires accurate and timely diagnosis.
• Current diagnostic methods can be time-consuming and require expert interpretation.

Purpose of the Study:

• To develop a fast, automated deep-learning algorithm for predicting atrial fibrillation.
• To evaluate the performance of the proposed algorithm in classifying atrial fibrillation from ECG signals.

Main Methods:

• Pre-processing of ECG signals.
• Conversion of ECG signals into spectrograms for alternative signal representation.
• Utilizing a fine-tuned EfficientNet B0 (a pre-trained convolutional neural network) for classification.
• Employing a transfer learning approach for model optimization.

Main Results:

• The deep-learning algorithm achieved an F-1 score of 88.2% and an accuracy of 97.3%.
• The model demonstrated highly efficient and effective classification of atrial fibrillation.
• The automated approach offers a rapid prediction of cardiac arrhythmia.

Conclusions:

• The developed deep-learning algorithm provides a promising tool for the automated and accurate detection of atrial fibrillation.
• This approach can assist healthcare professionals in diagnosing cardiac arrhythmia more effectively.
• The study highlights the potential of advanced AI techniques in improving cardiovascular diagnostics.