Using AI To Predict Atrial Fibrillation Earlier and More Accurately

Axel Loewe?PhD and colleagues at the?Institute of Biomedical Engineering at Karlsruhe Institute of Technology in Germany are developing new ways to predict cardiovascular diseases earlier and more accurately. Dr. Loewe leads an interdisciplinary team that is developing computer models of the human heart using software engineering, algorithmics, numerics, signal processing, data analysis, and machine learning. The group applies the models in simulation studies and brings them into clinical application by creating individualized digital twins of patients. They use digital twins to optimize diagnostic approaches and personalize therapies. They use AI methods based on simulated data and clinical information to help decipher disease mechanisms.

Axel Loewe?PhD at the Institute of Biomedical Engineering at Karlsruhe Institute of Technology. Image source Karlsruhe Institute of Technology

Atrial Fibrillation

In a healthy heart, a small cluster of cells sends out an electrical signal. The signal travels through the atria causing them to contract and pump out blood. A healthy heart pumps blood with a steady rhythm and 60-100 beats per minute. In atrial fibrillation (AF) electrical signals fire from multiple locations in the atria in a chaotic way. This causes the heart to beat irregularly and very fast with 100-175 beats per minute.

Some people with AF?may have no symptoms but others may experience a pounding heartbeat, chest pain, dizziness, fatigue, shortness of breath, and weakness. Risk factors for AF include aging, high blood pressure, obesity, European ancestry, diabetes, heart failure, ischemic heart disease, hyperthyroidism, chronic kidney disease, moderate alcohol use, and smoking.

AF is the most common type of arrhythmia. It is a serious medical condition that can lead to stroke, hospitalization, and death. AF is the leading cause of embolic stroke and increases the risk for heart failure 5-fold and mortality 2-fold. Strokes caused by AF are more severe than strokes caused by other conditions. Over 37 million people worldwide suffer with AF today and that number is expected to increase to 60 million people by 2050.

When studying how cardiac motion affects ECGs, Dr. Loewe and his team found amplitude differences in leads close to the heart. Source Axel Loewe?PhD

"This is?the first study that used only simulated data to train a machine learning classifier that then showed very good performance on real-world clinical data."

Axel Loewe?PhD, Head of Computational Cardiac Modeling at Karlsruhe Institute of Technology

Using Machine Learning To Predict AF

Doctors diagnose AF by analyzing electrocardiogram?(ECG) signals. An?ECG?can show if the heart is beating too fast or too slow. Dr. Loewe and his team in Germany recently collaborated with scientists at universities in Italy, Spain, Brazil, and the UK in an important study using machine learning to analyze ECG signals. This study is significant because it is the first study that used only simulated data to train a machine learning classifier that then showed very good performance on real-world clinical data. The study demonstrates that using machine learning to analyze ECG signals can help predict AF.?This could help millions of people avoid strokes and provide them with personalized treatment.

The training data were simulated using mechanistic multi-scale models of cardiac electrophysiology from ion channel via cell, tissue and organ up to the ECG on the body surface.?The study is described in a paper entitled Machine learning enables noninvasive prediction of atrial fibrillation driver location and acute pulmonary vein ablation success using the 12-lead ECG published in the Cardiovascular Digital Health Journal.

Study Overview

• AF drivers were induced in 2 computerized atrial models and combined with 8 torso models, resulting in 1,128 12-lead ECGs.
• a total of 103 features were extracted from the signals.
• features were extracted from the signals using several biosignal processing methods including Hjorth descriptors, recurrence quantification analysis on the?vectorcardiogram, spatial reduced RQA, individual component RQA, principal component analysis, organization index, and spectral entropy.
• a binary decision tree classifier was trained on the simulated data and evaluated using hold-out cross-validation.
• the PVs were subsequently isolated in the models to assess PVI success.
• the classifier was tested on a clinical dataset.
• there were 46 patients: 23 PV-dependent AF and 23 with additional extra-PV sources.

A.1 - Example of simulated atrial fibrillation driver located near the?pulmonary veins.?

B.1 -?Example of simulated atrial fibrillation driver located in an extra-pulmonary vein region. The red arrows show the AF driver position and propagation direction.?

A.2, B.2 - Body surface potential?maps on a magnetic resonance imaging–derived torso model. The torso potentials were obtained by solving the forward problem of electrocardiography from the simulated transmembrane voltages on the?atria.?

A.3, B.3?- f-waves for leads I, II, and V1?from the 12-lead electrocardiogram signals extracted from the BSPMs.

Results and Conclusions

1. the results presented in this study show that a machine learning classifier to distinguish between AF drivers located in the?PVs?vs in other atrial regions is feasible
2. the classifier yielded 82.6% specificity and 73.9% sensitivity for detecting PV drivers on the clinical data.
3. consistency analysis on the 46 patients resulted in 93.5% results match.
4. classifiers trained on simulated data for which the ground truth is known can generalize well to unseen clinical data
5. AF drivers identified in the PVs could be considered to be treated with cryoballoon?PVI?without prior time-consuming and costly electroanatomic mapping
6. AF drivers identified in extra-PV areas using the ECG are unlikely to terminate or convert upon PVI.
7. applying PVI on the simulated AF cases terminated AF in 100% of the cases in the PV class.
8. machine learning–based classification of 12-lead-ECG allows discrimination between patients with PV drivers vs those with extra-PV drivers of AF.
9. the novel algorithm may aid to identify patients with high acute success rates to PVI.
10. a machine learning–based classifier leveraging the routinely available noninvasive ECG signal could prove to be valuable for clinical decision-making and increase personalization of therapy.

References

Machine learning enables noninvasive prediction of atrial fibrillation driver location and acute pulmonary vein ablation success using the 12-lead ECG , Cardiovascular Digital Health Journal

Authors: Giorgio Luongo MSc, LucaAzzolin MSc, SteffenSchuler MSc, Massimo W. Rivolta PhD, Tiago P. Almeida PhD, Juan P. Martínez PhD, Diogo C. Soriano PhD, Armin Luik MD, Bj?rn Müller Edenborn MD, Amir Jadidi MD, Olaf D?ssel PhD, Roberto Sassi PhD, Pablo Laguna PhD, Axel Loewe PhD

Affiliations: Institute of Biomedical Engineering, Karlsruhe Institute of Technology, Karlsruhe, Germany, Dipartimento di Informatica, Università degli Studi di Milano, Milan, Italy, Department of Cardiovascular Sciences, University of Leicester, Leicester, United Kingdom, Universidad de Zaragoza, and CIBER-BNN, Zaragoza, Spain, Engineering, Modelling and Applied Social Sciences Centre, ABC Federal University, S?o Bernardo do Campo, Brazil, Medizinische Klinik IV, St?dtisches Klinikum Karlsruhe, Karlsruhe, Germany, Department of Electrophysiology, University-Heart-Center Freiburg-Bad Krozingen, Bad Krozingen Campus, Bad Krozingen, Germany

Subscribe and Comment

I'm interested in your feedback - please leave your comments. To subscribe please click subscribe at the top of this article.

Copyright ? 2022 Margaretta Colangelo. All Rights Reserved.

This article was written by?Margaretta Colangelo. ?Margaretta is Co-founder of Jthereum. She serves on the advisory board of the AI Precision Health Institute at the University of Hawai?i?Cancer Center. She's based in San Francisco?

Twitter?@realmargaretta