Learning Explainable Time-Morphology Patterns for Automatic Arrhythmia Classification from Short Single-Lead ECGs
Automatic detection of abnormal heart rhythms, including atrial fibrillation (AF), using signals obtained from a single-lead wearable electrocardiogram (ECG) device, is useful for daily cardiac health monitoring. In this study, we propose a novel image-based deep learning framework to classify singl...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2021-06-01
|
| Series: | Sensors |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1424-8220/21/13/4331 |
| _version_ | 1797528890654064640 |
|---|---|
| author | Hyeonjeong Lee Miyoung Shin |
| author_facet | Hyeonjeong Lee Miyoung Shin |
| author_sort | Hyeonjeong Lee |
| collection | DOAJ |
| description | Automatic detection of abnormal heart rhythms, including atrial fibrillation (AF), using signals obtained from a single-lead wearable electrocardiogram (ECG) device, is useful for daily cardiac health monitoring. In this study, we propose a novel image-based deep learning framework to classify single-lead ECG recordings of short variable length into several different rhythms associated with arrhythmias. By transforming variable-length 1D ECG signals into fixed-size 2D time-morphology representations and feeding them to the beat–interval–texture convolutional neural network (BIT-CNN) model, we aimed to learn the comprehensible characteristics of beat shape and inter-beat patterns over time for arrhythmia classification. The proposed approach allows feature embedding vectors to provide interpretable time-morphology patterns focused at each step of the learning process. In addition, this method reduces the number of model parameters needed to be trained and aids visual interpretation, while maintaining similar performance to other CNN-based approaches to arrhythmia classification. For experiments, we used the PhysioNet/CinC Challenge 2017 dataset and achieved an overall F<sub>1_NAO</sub> of 81.75% and F<sub>1_NAOP</sub> of 76.87%, which are comparable to those of the state-of-the-art methods for variable-length ECGs. |
| first_indexed | 2024-03-10T10:04:50Z |
| format | Article |
| id | doaj.art-a465f64d5ff241c9950cd46915de950c |
| institution | Directory Open Access Journal |
| issn | 1424-8220 |
| language | English |
| last_indexed | 2024-03-10T10:04:50Z |
| publishDate | 2021-06-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Sensors |
| spelling | doaj.art-a465f64d5ff241c9950cd46915de950c2023-11-22T01:37:29ZengMDPI AGSensors1424-82202021-06-012113433110.3390/s21134331Learning Explainable Time-Morphology Patterns for Automatic Arrhythmia Classification from Short Single-Lead ECGsHyeonjeong Lee0Miyoung Shin1Bio-Intelligence & Data Mining Laboratory, School of Electronic and Electrical Engineering, Kyungpook National University, Daegu 41566, KoreaBio-Intelligence & Data Mining Laboratory, School of Electronic and Electrical Engineering, Kyungpook National University, Daegu 41566, KoreaAutomatic detection of abnormal heart rhythms, including atrial fibrillation (AF), using signals obtained from a single-lead wearable electrocardiogram (ECG) device, is useful for daily cardiac health monitoring. In this study, we propose a novel image-based deep learning framework to classify single-lead ECG recordings of short variable length into several different rhythms associated with arrhythmias. By transforming variable-length 1D ECG signals into fixed-size 2D time-morphology representations and feeding them to the beat–interval–texture convolutional neural network (BIT-CNN) model, we aimed to learn the comprehensible characteristics of beat shape and inter-beat patterns over time for arrhythmia classification. The proposed approach allows feature embedding vectors to provide interpretable time-morphology patterns focused at each step of the learning process. In addition, this method reduces the number of model parameters needed to be trained and aids visual interpretation, while maintaining similar performance to other CNN-based approaches to arrhythmia classification. For experiments, we used the PhysioNet/CinC Challenge 2017 dataset and achieved an overall F<sub>1_NAO</sub> of 81.75% and F<sub>1_NAOP</sub> of 76.87%, which are comparable to those of the state-of-the-art methods for variable-length ECGs.https://www.mdpi.com/1424-8220/21/13/4331arrhythmia classificationatrial fibrillation (AF)electrocardiogram (ECG)convolutional neural network (CNN)deep learning |
| spellingShingle | Hyeonjeong Lee Miyoung Shin Learning Explainable Time-Morphology Patterns for Automatic Arrhythmia Classification from Short Single-Lead ECGs Sensors arrhythmia classification atrial fibrillation (AF) electrocardiogram (ECG) convolutional neural network (CNN) deep learning |
| title | Learning Explainable Time-Morphology Patterns for Automatic Arrhythmia Classification from Short Single-Lead ECGs |
| title_full | Learning Explainable Time-Morphology Patterns for Automatic Arrhythmia Classification from Short Single-Lead ECGs |
| title_fullStr | Learning Explainable Time-Morphology Patterns for Automatic Arrhythmia Classification from Short Single-Lead ECGs |
| title_full_unstemmed | Learning Explainable Time-Morphology Patterns for Automatic Arrhythmia Classification from Short Single-Lead ECGs |
| title_short | Learning Explainable Time-Morphology Patterns for Automatic Arrhythmia Classification from Short Single-Lead ECGs |
| title_sort | learning explainable time morphology patterns for automatic arrhythmia classification from short single lead ecgs |
| topic | arrhythmia classification atrial fibrillation (AF) electrocardiogram (ECG) convolutional neural network (CNN) deep learning |
| url | https://www.mdpi.com/1424-8220/21/13/4331 |
| work_keys_str_mv | AT hyeonjeonglee learningexplainabletimemorphologypatternsforautomaticarrhythmiaclassificationfromshortsingleleadecgs AT miyoungshin learningexplainabletimemorphologypatternsforautomaticarrhythmiaclassificationfromshortsingleleadecgs |