Multiscaled Fusion of Deep Convolutional Neural Networks for Screening Atrial Fibrillation From Single Lead Short ECG...

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Atrial fibrillation (AF) is a common cardiac arrhythmia in the elderly, leading to significant mortality and morbidity.
• Early AF detection is vital to prevent severe health consequences like stroke and heart failure.
• The episodic nature of AF presents challenges for accurate and timely diagnosis using traditional methods.

Purpose of the Study:

• To propose a multiscaled fusion of deep convolutional neural network (MS-CNN) for effective atrial fibrillation screening.
• To evaluate the performance of MS-CNN in classifying AF from short, single-lead electrocardiogram (ECG) recordings.
• To demonstrate the capability of MS-CNN for potential use in daily monitoring with wearable devices.

Main Methods:

• Development of a novel MS-CNN architecture utilizing two-stream convolutional networks with varying filter sizes.
• Training and testing the MS-CNN model on single-lead short ECG recordings.
• Comparative analysis of MS-CNN performance against other machine learning models, including ANNs, shallow CNNs, and VGG networks.

Main Results:

• The MS-CNN achieved a classification accuracy of 96.99% on 5-second ECG recordings.
• Optimal classification accuracy of 98.13% was achieved with 20-second ECG recordings.
• MS-CNN demonstrated superior performance compared to traditional artificial neural networks and single-stream CNNs.
• Feature visualization confirmed MS-CNN's ability to extract relevant ECG features without manual engineering.

Conclusions:

• The proposed MS-CNN offers a highly accurate and efficient method for screening atrial fibrillation from short ECG recordings.
• MS-CNN's performance suggests its suitability for integration into wearable devices for continuous elderly monitoring.
• This deep learning approach overcomes limitations of traditional methods in detecting episodic AF.