An AI-based Predictor of Atrial Fibrillation Recurrence Following Pulmonary Vein Isolation
(2024) BMEM01 20241Department of Biomedical Engineering
- Abstract
- Atrial fibrillation (AF) is the most prevalent arrhythmia, increasing with each year. AF is characterized by periods of rapid and irregular beating of the heart and increases the risk of stroke and heart failure. The disease often originates from ectopic beats generated by the pulmonary veins, activating the atria, creating short periods of AF. Without treatment the disease evolves, remodels the atria, and eventually develops into a permanent condition of AF. Treatment with pulmonary vein isolation (PVI) through ablation is commonly prescribed. The treatment is effective, but AF recurrence post ablation occurs regularly. Clinical success is dependent on atrial health which can be interpreted through the P-wave of an electrocardiogram... (More)
- Atrial fibrillation (AF) is the most prevalent arrhythmia, increasing with each year. AF is characterized by periods of rapid and irregular beating of the heart and increases the risk of stroke and heart failure. The disease often originates from ectopic beats generated by the pulmonary veins, activating the atria, creating short periods of AF. Without treatment the disease evolves, remodels the atria, and eventually develops into a permanent condition of AF. Treatment with pulmonary vein isolation (PVI) through ablation is commonly prescribed. The treatment is effective, but AF recurrence post ablation occurs regularly. Clinical success is dependent on atrial health which can be interpreted through the P-wave of an electrocardiogram (ECG). With the help of this information, a more personalized treatment can be implemented, leading to increased arrhythmia freedom. Moreover, outcome prediction on ECG data using machine learning (ML) has proven efficient, but often uses extensive, inefficient manual delineation of the P-waves.
Therefore this study aims to evaluate AF recurrence prediction capabilities for two models trained on automatically extracted P-wave characteristics, raw ECG and coronary sinus (CS) catheter electrograms, respectively. Furthermore, one model is evaluated by training and testing on synthesized ECG data. The proposed method, applied to datasets of 281 subjects in total, automatically delineates ECG signals of adequate quality, recorded before, during and after PVI ablation. Annotations from delineation are used in beat classification, as well as to extract P-wave features, separately. All data is balanced based on beats per patient and outcome label and subsets with training and testing data are created. A Random Forest (RF) classifier is trained on the P-wave features and a Convolutional Neural Network (CNN) is trained on segments consisting of raw ECG and CS catheter electrograms. In order to further evaluate the CNN, a separate, synthesized dataset, to be used in training and testing, is created by inserting augmented P-wave characteristics in the ECG signals of AF recurrent subjects. The Area under the ROC Curve (AUC)-scores of the RF classifier and CNNs are 0.502 and 0.463. Furthermore, the AUC-score for the CNN training on synthesized ECG data is 0.793. Conclusively, neither of the RF classifier or the CNN can predict AF recurrence using this method and dataset. However, the result for the CNN training on synthesized ECG data illustrates the potential of a CNN to extract information from P-wave morphology. Moreover, validation on automatic delineation on this dataset is needed to investigate the potential of the RF classifier. Conclusively, more research is needed on ML models trained on automatically delineated data. (Less) - Popular Abstract
- Using AI to foresee the future, tackling the most common heart rhythm disease.
Atrial fibrillation is a heart disease affecting an increasing number of people around the world, posing a great challenge in modern healthcare. The focus of this work is to investigate the use of AI to predict which patients will get the disease again after treatment.
Atrial fibrillation affects 1 in 3 people over the age of 55, and the number only seems to be increasing. This heart disease is recognized by an irregular and fast heart rhythm. The most successful treatment is to ablate the part of the heart that most often causes atrial fibrillation; however, it often happens that the disease returns. Researchers can today, with the help of an ECG, guess... (More) - Using AI to foresee the future, tackling the most common heart rhythm disease.
Atrial fibrillation is a heart disease affecting an increasing number of people around the world, posing a great challenge in modern healthcare. The focus of this work is to investigate the use of AI to predict which patients will get the disease again after treatment.
Atrial fibrillation affects 1 in 3 people over the age of 55, and the number only seems to be increasing. This heart disease is recognized by an irregular and fast heart rhythm. The most successful treatment is to ablate the part of the heart that most often causes atrial fibrillation; however, it often happens that the disease returns. Researchers can today, with the help of an ECG, guess who will get the disease again. With the help of this, before surgery, a more accurate decision can be made about where and how much to ablate, which leads to an increased number of healthy patients. With the help of AI, patients that have a higher risk of relapse can be found faster.
This study presents an automated method for extracting the most important part of an ECG for this purpose, the P wave. This is used to develop AI models. These models then predict which patients will relapse after treatment and which will not, automating the process and alleviating the clinical staff.
The resulting AI models fail to predict which patients will get atrial fibrillation again. However, if looking at previous research and the limitations in the method, the conclusion is: More research is needed on the implementation of automatic analysis of ECG. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9161335
- author
- Botvidsson, Jakob LU
- supervisor
- organization
- alternative title
- AI-baserad Prediktering av Återfall av Förmaksflimmer efter Lungvensisolering
- course
- BMEM01 20241
- year
- 2024
- type
- H2 - Master's Degree (Two Years)
- subject
- language
- English
- additional info
- 2024-10
- id
- 9161335
- date added to LUP
- 2024-06-18 15:18:09
- date last changed
- 2024-06-18 15:18:09
@misc{9161335,
abstract = {{Atrial fibrillation (AF) is the most prevalent arrhythmia, increasing with each year. AF is characterized by periods of rapid and irregular beating of the heart and increases the risk of stroke and heart failure. The disease often originates from ectopic beats generated by the pulmonary veins, activating the atria, creating short periods of AF. Without treatment the disease evolves, remodels the atria, and eventually develops into a permanent condition of AF. Treatment with pulmonary vein isolation (PVI) through ablation is commonly prescribed. The treatment is effective, but AF recurrence post ablation occurs regularly. Clinical success is dependent on atrial health which can be interpreted through the P-wave of an electrocardiogram (ECG). With the help of this information, a more personalized treatment can be implemented, leading to increased arrhythmia freedom. Moreover, outcome prediction on ECG data using machine learning (ML) has proven efficient, but often uses extensive, inefficient manual delineation of the P-waves.
Therefore this study aims to evaluate AF recurrence prediction capabilities for two models trained on automatically extracted P-wave characteristics, raw ECG and coronary sinus (CS) catheter electrograms, respectively. Furthermore, one model is evaluated by training and testing on synthesized ECG data. The proposed method, applied to datasets of 281 subjects in total, automatically delineates ECG signals of adequate quality, recorded before, during and after PVI ablation. Annotations from delineation are used in beat classification, as well as to extract P-wave features, separately. All data is balanced based on beats per patient and outcome label and subsets with training and testing data are created. A Random Forest (RF) classifier is trained on the P-wave features and a Convolutional Neural Network (CNN) is trained on segments consisting of raw ECG and CS catheter electrograms. In order to further evaluate the CNN, a separate, synthesized dataset, to be used in training and testing, is created by inserting augmented P-wave characteristics in the ECG signals of AF recurrent subjects. The Area under the ROC Curve (AUC)-scores of the RF classifier and CNNs are 0.502 and 0.463. Furthermore, the AUC-score for the CNN training on synthesized ECG data is 0.793. Conclusively, neither of the RF classifier or the CNN can predict AF recurrence using this method and dataset. However, the result for the CNN training on synthesized ECG data illustrates the potential of a CNN to extract information from P-wave morphology. Moreover, validation on automatic delineation on this dataset is needed to investigate the potential of the RF classifier. Conclusively, more research is needed on ML models trained on automatically delineated data.}},
author = {{Botvidsson, Jakob}},
language = {{eng}},
note = {{Student Paper}},
title = {{An AI-based Predictor of Atrial Fibrillation Recurrence Following Pulmonary Vein Isolation}},
year = {{2024}},
}