Abstract:Background and objective: Electrocardiogram (ECG) is one of the most important tools for assessing cardiac function and detecting potential heart problems. However, most of the current ECG report records remain on the paper, which makes it difficult to preserve and analyze the data. Moreover, paper records could result in the loss significant data, which brings inconvenience to the subsequent clinical diagnosis or artificial intelligence-assisted heart health diagnosis. Taking digital pictures is an intuitive way of preserving these files and can be done simply using smartphones or any other devices with cameras. However, these real scene ECG images often have some image noise that hinders signal extraction. How to eliminate image noise and extract ECG binary image automatically from the noisy and low-quality real scene images of ECG reports is the first problem to be solved in this paper. Next, QRS recognition is implemented on the extracted binary images to determine key points of ECG signals. 1D digital ECG signal is also extracted for accessing the exact values of the extracted points. In light of these tasks, an automatic digital ECG signal extraction and normal QRS recognition from real scene ECG images is proposed in this paper. Methods: The normal QRS recognition approach for real scene ECG images in this paper consists of two steps: ECG binary image extraction from ECG images using a new two-layer hierarchical method, and the subsequent QRS recognition based on a novel feature-fusing method. ECG binary image extraction is implemented using sub-channel filters followed by an adaptive filtering algorithm. According to the ratio between pixel and real value of ECG binary image, 1D digital ECG signal is obtained. The normal QRS recognition includes three main steps: establishment of candidate point sets, feature fusion extraction, and QRS recognition. Two datasets are introduced for evaluation including a real scene ECG images dataset and the public Non-Invasive Fetal Electrocardiogram Database (FECG). Results: Through the experiment on real scene ECG image, the F-1 score for Q, R, S detection is 0.841, 0.992, and 0.891, respectively. The evaluation on the public FECG dataset also proves the robustness of our algorithm, where F-1 score for R is 0.992 (0.996 for thoracic lead) and 0.988 for thoracic S wave. Conclusions: The proposed method in this article is a promising tool for automatically extracting digital ECG signals and detecting QRS complex in real scene ECG images with normal QRS. (C) 2019 Elsevier B.V. All rights reserved.

Deep learning for digitizing highly noisy paper-based ECG records

A Novel Method for ECG Paper Records Digitization.

DeepECG: Image-based Electrocardiogram Interpretation with Deep Convolutional Neural Networks

Multi-label Classification of Abnormalities in 12-Lead ECG Using Deep Learning.

Auto Lead Extraction and Digitization of ECG Paper Records using cGAN

Opportunities and Challenges of Deep Learning Methods for Electrocardiogram Data: A Systematic Review

DENS-ECG: A Deep Learning Approach for ECG Signal Delineation

Interpretable Deep Learning for Automatic Diagnosis of 12-lead Electrocardiogram

A deep learning approach to cardiovascular disease classification using empirical mode decomposition for ECG feature extraction

Paper-Recorded ECG Digitization Method with Automatic Reference Voltage Selection for Telemonitoring and Diagnosis

Combining Hough Transform and Deep Learning Approaches to Reconstruct ECG Signals From Printouts

ECG_SegNet: An ECG delineation model based on the encoder-decoder structure

Automatic Digital ECG Signal Extraction and Normal QRS Recognition from Real Scene ECG Images

Automated Detection of Cardiovascular Disease by Electrocardiogram Signal Analysis: a Deep Learning System

Image based deep learning in 12-lead ECG diagnosis

Semantic ECG Interval Segmentation Using Autoencoders

Research of Artificial Intelligence Algorithm for Identifying Atrial Fibrillation from Paper-Based ECG

ECG noise classification using deep learning with feature extraction

An automated ECG-based deep learning for the early-stage identification and classification of cardiovascular disease

Enhancing Deep Learning-based 3-lead ECG Classification with Heartbeat Counting and Demographic Data Integration