SS-SWT and SI-CNN: An Atrial Fibrillation Detection Framework for Time-Frequency ECG Signal

Hongpo Zhang, Renke He, Honghua Dai, Mingliang Xu, Zongmin WangJournal of Healthcare Engineering 2020 · DOI

TL;DR

This paper introduces a novel framework, SS-SWT and SI-CNN, for accurately and rapidly detecting atrial fibrillation (AF) from ECG signals. By intelligently selecting specific time-frequency features using a specialized wavelet transform and processing them with a custom convolutional neural network, the method achieves high performance with simplified feature extraction.

What Problem Does This Paper Address?

Atrial fibrillation (AF) is a prevalent heart rhythm disorder linked to serious health risks like stroke and heart failure. Early detection is crucial for effective management, but AF can be intermittent and asymptomatic, making it challenging to diagnose. Traditional methods often rely on manual interpretation or complex feature engineering, which can be time-consuming and error-prone, especially for short AF episodes. This paper aims to develop an automated, efficient, and accurate system for AF detection that overcomes these limitations.

Methods

The proposed framework consists of two main components:

  • Specific-Scale Stationary Wavelet Transform (SS-SWT): This method first decomposes a 5-second ECG signal into multiple frequency bands (scales) using the Stationary Wavelet Transform (SWT). Crucially, instead of using all coefficients, SS-SWT selectively picks coefficients from specific scales that are deemed most informative for AF detection, discarding redundant information. This selective approach simplifies feature extraction and reduces data redundancy. The selected coefficients are then arranged into a 2D matrix.
  • Scale-Independent Convolutional Neural Network (SI-CNN): This custom CNN architecture is designed to process the 2D time-frequency coefficient matrices generated by SS-SWT. A key innovation is the use of a 1x3 convolutional kernel. This design preserves the independence between different frequency scales (rows in the matrix) while effectively fusing time-domain features within each scale. This approach is tailored to the specific structure of the wavelet coefficient matrix, aiming to extract both time and frequency domain characteristics more effectively than standard CNNs.

Datasets

The study utilizes the MIT-BIH Atrial Fibrillation Database (MIT-BIH AFDB) for both training and evaluation. This is a well-established benchmark dataset for AF research. The authors segmented the data into 5-second intervals and partitioned it into training, validation, and testing sets with a ratio of 7:1:2.

Key Insights and Ideas

  • Intelligent Feature Selection: The SS-SWT approach highlights that not all wavelet decomposition scales are equally important for AF detection. By focusing on specific scales, the method effectively filters out noise and redundancy, leading to a more efficient and robust feature representation.
  • Tailored CNN Architecture: The SI-CNN's 1x3 convolutional kernel is a clever adaptation to the nature of time-frequency representations from wavelet transforms. It respects the inherent independence of different frequency bands while still allowing for effective feature learning, a departure from standard 2D convolutions that might treat all spatial dimensions equally.
  • Simplification of Preprocessing: The SS-SWT method aims to reduce the need for extensive preprocessing steps like denoising or baseline drift removal, as the selective feature extraction inherently handles some of these issues.

Key Results

The proposed SS-SWT and SI-CNN framework achieved impressive results on the MIT-BIH AFDB dataset:

  • Sensitivity (Se): 99.03%
  • Specificity (Sp): 99.35%
  • Overall Accuracy (Acc): 99.23%

These metrics were obtained using 5-second ECG segments. The authors also demonstrated that their method simplifies feature extraction and reduces redundancy compared to other wavelet-based approaches.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that while their method is effective, it still relies on the quality of the input ECG signal.
  • The performance of other methods is noted to be highly dependent on accurate R-wave peak detection, which their proposed method aims to avoid.
  • Future Work:
    • The authors suggest that their method has high clinical potential and will be a focus of their future work.
  • Further exploration of optimizing the selection of specific wavelet scales and the SI-CNN architecture could lead to even better performance.

Summary updated

Abstract

Atrial fibrillation is the most common arrhythmia and is associated with high morbidity and mortality from stroke, heart failure, myocardial infarction, and cerebral thrombosis. Effective and rapid detection of atrial fibrillation is critical to reducing morbidity and mortality in patients. Screening atrial fibrillation quickly and efficiently remains a challenging task. In this paper, we propose SS-SWT and SI-CNN: an atrial fibrillation detection framework for the time-frequency ECG signal. First, specific-scale stationary wavelet transform (SS-SWT) is used to decompose a 5-s ECG signal into 8 scales. We select specific scales of coefficients as valid time-frequency features and abandon the other coefficients. The selected coefficients are fed to the scale-independent convolutional neural network (SI-CNN) as a two-dimensional (2D) matrix. In SI-CNN, a convolution kernel specifically for the time-frequency characteristics of ECG signals is designed. During the convolution process, the independence between each scale of coefficient is preserved, and the time domain and the frequency domain characteristics of the ECG signal are effectively extracted, and finally the atrial fibrillation signal is quickly and accurately identified. In this study, experiments are performed using the MIT-BIH AFDB data in 5-s data segments. We achieve 99.03% sensitivity, 99.35% specificity, and 99.23% overall accuracy. The SS-SWT and SI-CNN we propose simplify the feature extraction step, effectively extracts the features of ECG, and reduces the feature redundancy that may be caused by wavelet transform. The results shows that the method can effectively detect atrial fibrillation signals and has potential in clinical application.
BibTeX
@article{zhang2020ss,
  author = {Hongpo Zhang and Renke He and Honghua Dai and Mingliang Xu and Zongmin Wang},
  title = {{SS-SWT and SI-CNN: An Atrial Fibrillation Detection Framework for Time-Frequency ECG Signal}},
  journal = {Journal of Healthcare Engineering},
  year = {2020},
  doi = {10.1155/2020/7526825},
}
Figure