Blog > ECG Analysis

ECG Analysis: Feature Extraction, Signal Processing, and Deep Learning

1. Introduction

Cardiovascular diseases are critical illnesses worldwide and cause 30% of deaths in the world according to the World Health Organization (WHO). Early detection of patients with this risk and a good understanding of the disease is very critical to improve diagnosis and treatment.

Electrocardiogram (ECG) recordings capture the propagation of the electrical signals of the heart on the body surface. ECG shows a complex of P, Q, R, S, and T waves corresponding to each heartbeat. The changes in waves, the appearance of different waves from regular ones, and the changes in the times between waves give technicians, analysts, and doctors indications about heart diseases.

Figure 1. ECG of a heart in normal sinus rhythm (Source)

On the other hand, manually reviewing large amounts of ECG data is time-consuming and sometimes almost impossible. Therefore, powerful computational methods are needed to extract information and insight from big ECG datasets. Different calculation and analysis techniques should be also designed and used for different ECG formats and clinical applications.

The most important phase of an ECG analysis is to transform an ECG record to a human-readable format that can be achieved using conventional computational, imaging, and drawing techniques. So, the technicians and analysts can inspect the heartbeats on a screen second by second. On the other side, detecting heartbeats and arrhythmias in an automatic way may not be so easy as supposed. First, all heartbeats should be identified using different features (morphologic, time interval, hybrid, etc.) which means robust feature extraction techniques are needed as a part of preprocessing steps. Some of the common feature extraction methods used are mentioned in Section 2.


2. Feature Extraction Using Traditional Signal Processing Methods

Method 1.1: Morphological features

The morphological features such as slopes, peaks, and amplitudes describe the shape of the ECG waveforms.

Method 1.2: Time interval features

They characterize the dynamics of ECG such as QRS duration, QT interval, or heart rate, defined as the number of beats per unit of time.

Method 2: Feature Extraction Using the State-of-the-art Deep Learning Models

With this approach, the raw signal data is used as input to the model and the model attempts to learn data to extract the features by itself.

Method 3: Features Extraction Using Hybrid Methods

In real-life scenarios where the ECG data is very noisy, it is very crucial to use a hybrid combination of the feature extraction methods mentioned above.


3. General ECG Analysis Overview

Figure 2 shows an overview outline of the ECG analysis pipeline of the technical side.

Figure 2. ECG analysis overview

If you are interested in ECG analysis in more detail, you can check the articles listed below.

(1) Shenda Hong, Yuxi Zhou, Junyuan Shang, Cao Xiao, and Jimeng Sun. "Opportunities and Challenges of Deep Learning Methods for Electrocardiogram Data: A Systematic Review." Computers in Biology and Medicine.

(2) Ma, Linhai, and Liang Liang. "Improve robustness of DNN for ECG signal classification: a noise-to-signal ratio perspective." arXiv preprint arXiv:2005.09134 (2020).

(3) Zhu, Hongling, Cheng Cheng, Hang Yin, Xingyi Li, Ping Zuo, Jia Ding, Fan Lin et al. "Automatic multilabel electrocardiogram diagnosis of heart rhythm or conduction abnormalities with deep learning: a cohort study." The Lancet Digital Health 2, no. 7 (2020): e348-e357.

(4) Attia, Zachi I., Peter A. Noseworthy, Francisco Lopez-Jimenez, Samuel J. Asirvatham, Abhishek J. Deshmukh, Bernard J. Gersh, Rickey E. Carter et al. "An artificial intelligence-enabled ECG algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction." The Lancet 394, no. 10201 (2019): 861-867.

(5) Van Steenkiste, Glenn, Gunther van Loon, and Guillaume Crevecoeur. "Transfer learning in ECG classification from human to horse using a novel parallel neural network architecture." Scientific reports 10, no. 1 (2020): 1-12.

(6) Mousavi, Sajad, Atiyeh Fotoohinasab, and Fatemeh Afghah. "Single-modal and multi-modal false arrhythmia alarm reduction using attention-based convolutional and recurrent neural networks." PloS one 15, no. 1 (2020): e0226990.