AI In The Diagnosis And Management Of Arrhythmias

Electrophysiology was one of the first areas of cardiac medicine to start using artificial intelligence (AI) methodologies. In recent times, deep learning (DL) techniques have offered new possibilities in electrocardiography including the signature identification of diseases. AI advances in 3D cardiac imaging have given birth to the concept of virtual hearts and cardiac arrhythmias simulation.

In this review, a team of British and American researchers discussed the impact of AI and recent technological advances in all aspects of arrhythmia care. The article offered an overview of milestones achieved and considered the future integration of AI into arrhythmia care.

Optimizing personalized treatment

AI has made great strides in healthcare in recent years. In electrocardiography and image interpretation, machine learning (ML) approaches are helping healthcare professionals to optimize personalized treatment. The adoption of AI methodologies has also opened the door for more patient care using minimally or non-invasive treatment modalities.

In electrophysiology, the majority of AI applications are used in the analysis of signals that represent cardiac electrical activity, with the signal varying according to the type of sensor and underlying technology.

The machine learning process

Typically in ML approaches, after the data is collected and pre-processed, it is feature engineered and classified. The ML methodology is then trained using training data with appropriate labels, before being assessed using a ‘validation data set’, allowing the algorithm to be fine-tuned. After these steps, the ML algorithm is ready to be used with a third test set. The performance of the algorithm can be expressed using values such as sensitivity, specificity, accuracy, receiver operating curve (ROC), and area under ROC (AUC).

The article segmented its review of AI in arrhythmia care into the following areas:

Arrhythmia detection using AI
AI and cardiac devices
ML in a multimodal integrative approach to predicting sites of arrhythmia origins
Personalized virtual heart modeling
ML in atrial fibrillation
Robotics in electrophysiology and the potential role of ML

Arrhythmia detection using AI

Since electrocardiograms (EKG) have been digitized, AI has been used for computerized interpretation. Enhanced ML algorithms have enabled noise reduction techniques, advanced feature extraction, and selection and reduction methods that have improved arrhythmia detection to a near 95% accuracy. Unsupervised DL algorithms have also been shown to be capable of identifying hidden diseased state signatures using a 12-lead ECG, detecting hyperkalemia, heart failure, hypoglycemia, and even changes in emotional states.

The article highlighted the “rapid growth of handheld and wearable cardiac monitoring systems,” and presented a table that contains the diagnostic accuracy of AI-aided devices in identifying atrial fibrillation in various studies. The accuracy of using AI-aided devices with EKG or photoplethysmography (PPG) signals for atrial fibrillation detection was no less than 90% in any of the six studies cited.

AI and cardiac devices

Building on the rule-based algorithms which most pacemaker and defibrillator functions already use, the article highlighted ML methods for the detection of shockable and non-shockable rhythms, with the shock advice algorithm representing a “considerable improvement over existing algorithms and in keeping with the standards set by the American Heart Association guidelines”.

ML and DL methods have also been used for:

The identification of cardiac device models from a chest radiograph
The improvement of cardiac resynchronization therapy (CRT) outcomes prediction
Predicting stroke using continuous remote monitoring data in patients with a cardiac implantable electronic device

ML in a multimodal integrative approach to predicting sites of arrhythmia origins

The contribution of ML-aided advances in cardiac 3D imaging to the non-invasive characterization of arrhythmia prior to attempting ablative therapy was highlighted. The article picked out the ability of DL methods such as convolutional neural networks to improve the speed of acquisition, time efficiency, reconstruction quality of images, and accuracy of cardiac magnetic resonance (CMR) segmentation.

Also outlined was the application of ML techniques to:

Improve myocardial tissue characterization and texture analysis
Define the heterogeneous nature of the scarred myocardium in late gadolinium enhancement (LGE) CMR images in patients post-myocardial infarction
Quantify hypertrophic cardiomyopathy

ML-aided imaging has been credited as paving the way for several developments in electrophysiology, including non-invasive localization of arrhythmia foci with high precision, personalized virtual heart modeling including simulation of cardiac arrhythmias, and the concept of non-invasive ablation.

Personalized virtual heart modeling

AI has played its part in the development of virtual heart modeling – an important platform for drug cardiotoxicity screening - including a computerized human cardiac myocyte with cellular modeling technology. Virtual heart modeling was used for non-invasive risk assessment of sudden cardiac death in a high-risk population undergoing cardioverter defibrillator implantation. This virtual heart arrhythmia risk predictor approach (VARP) was found to be superior to other predictors including left ventricular ejection fraction.

ML in atrial fibrillation

ML methodologies in conjunction with atrial computational models were used to define reentrant driver locations in AF. The first-in-human clinical study of virtual heart models to guide ablation in patients with persistent AF was detailed, as well as a study that combined ML and personalized computational modeling to predict AF recurrence post-pulmonary vein isolation in patients with paroxysmal AF.

These types of methodologies have also been applied to the large quantities of intracardiac data recorded during electrophysiology procedures. Using intracardiac electrograms and electro anatomical mapping data, ML methods have been used with a view to defining extra pulmonary ablation sites in AF.

Robotics in electrophysiology and the potential role of ML

Robotic navigation has been used in electrophysiology for over two decades. The article described robotic ablation as a “valuable tool to help improve access to the same level of accuracy in a reproducible fashion”.

Two of the recent developments in this area that were picked out by the review include; a mechanical robotic system that uses acoustic energy for both imaging and lesion deployment, and a magnetic navigation system that combines with 3D EAM systems and 3D image integration, and can be applied to all arrhythmias.

Conclusion

While acknowledging typical ML limitations such as overfitting, the article concluded that “AI has considerable impact on all aspects of patient management in the field of cardiac electrophysiology from the identification of arrhythmia to therapy (invasively and non-invasively)”.

The review singled out the AI-enabled EKG as “an innovation that may have a significant impact on the community-based diagnosis of latent cardiac conditions, even more so in the underprivileged parts of the world”.

Sources:

https://academic.oup.com/eurheartj/article/42/38/3904/6352590

https://pubmed.ncbi.nlm.nih.gov/28869899/

https://academic.oup.com/view-large/304793090

https://pubmed.ncbi.nlm.nih.gov/31122379/

https://pubmed.ncbi.nlm.nih.gov/29326129/

https://pubmed.ncbi.nlm.nih.gov/31610712/

https://pubmed.ncbi.nlm.nih.gov/30636630/

https://pubmed.ncbi.nlm.nih.gov/24053280/

https://pubmed.ncbi.nlm.nih.gov/31714190/

https://pubmed.ncbi.nlm.nih.gov/21637795/

https://pubmed.ncbi.nlm.nih.gov/27164184/