AI and ML In Arrhythmias And Cardiac Electrophysiology
February 14, 2022
The uses of artificial intelligence (AI) and machine learning (ML) in medicine are growing all the time, offering the advantages of time-saving automation and, in some cases, performing tasks beyond the capabilities of humans. AI and ML are also changing clinical practice and research in cardiac electrophysiology, proving beneficial to the detection and diagnosis of diseases, patient outcome prediction, and the novel characterization of disease.
A review led by a team from Cleveland Clinic Lerner College of Medicine, Case Western Reserve University in Ohio, United States, was aimed at educating novice readers on AI and ML methods, discussing the impact of AI and ML on cardiac electrophysiology, and highlighting important considerations and challenges for adoption of AI into clinical practice.
AI and ML definitions
The review describes AI as “machine-based data processing to achieve objectives that typically require human cognitive function,” underlining its ability to mine dense data and classify complex patterns of data beyond the capabilities of humans.
ML is defined as “a subdiscipline of AI and employs algorithms to learn patterns empirically from data”. Deep learning (DL) was highlighted as a “powerful ML approach”; with its ability to leverage large datasets aligned to recent advances in “computational power”.
The advancement of AI and ML in cardiovascular imaging
Public awareness of AI is increasing in modern life, with ML and DL being used in a broad variety of areas; from language processing to gaming, engineering, manufacturing, and science. The review traced back AI’s involvement in cardiac electrophysiology to the 1970s when automated EKG interpretation was achieved. But the technology has gained a new significance in cardiovascular imaging with the more recent development of large databases, algorithm innovations, software tools, and hardware capabilities. Not only are AI tools showing potential for the automation and assistance of disease diagnosis, but they could also be used to improve disease prognosis prediction and offer a novel characterization of health and disease.
The current role of AI and ML in clinical cardiac electrophysiology
According to the review, AI applications in disease detection and diagnosis are transforming clinical cardiac electrophysiology:
Mobile technology for arrhythmia detection
Mobile technology is being used to detect arrhythmia, with wearable photoplethysmographic sensors transforming possibilities for atrial fibrillation (AF) screening by enabling long-term, passive assessment of pulse rate and regularity to detect an irregular pulse consistent with AF.
The Apple Heart Study evaluated the photoplethysmographic monitoring algorithm of the Apple smartwatch in 419,297 participants. Some 2161 participants received an irregular pulse notification when 5 out of 6 photoplethysmographic tachograms suggested AF. Of the 450 participants who then wore an external heart monitor for one week, AF was detected in 34% of participants. The positive predictive value between the photoplethysmographic and a concurrent external monitor was 84%.
Expanding the use of 12-lead EKG
The use of the 12-lead EKG has been expanded. In a 2019 study, an automated cardiologist-level classification of 12 different rhythms was obtained with single-lead EKGs using a deep neural network. AI interpretation of the EKG also provides information about diseases not typically diagnosed on EKG, such as subtle potassium changes, the reliable detection of left ventricular dysfunction, and identifying patients with paroxysmal AF. The review concludes that these capabilities “represent complementary applications of AI to medicine - scaling our current workflow and insuring quality but also adding value to a routine medical test”.
Predictive and prognostic models for response to therapy
Predictive and prognostic models for response to therapy have opened the door to a deeper understanding of individual variation in the prevention and treatment of disease. In 2015’s Precision Medicine Initiative, The Centers for Medicare and Medicaid Services authorized the use of evidence-based patient decision aids in primary prevention implantable cardioverter-defibrillator implant and left atrial appendage closure. This was intended to foster shared decision-making with patients. It is foreseen that ML algorithms using diverse data from the electronic medical record (EMR) may become central to this process.
The improvement of CRT response prediction
Several studies highlighted by the review have shown potential in developing models to meet the need for improved response prediction in cardiac resynchronization therapy (CRT). One model differentiated outcomes following CRT better than current clinical discriminators of bundle branch block morphology and QRS duration. In another study, echocardiographic response to CRT in a retrospective cohort from two institutions was predicted using an ML model that marginally outperformed current guidelines in predicting response and improved discrimination of event-free survival.
Novel characterization of disease
AI has also shown potential in providing a novel characterization of disease processes and phenotypes between individuals and the development of novel granular classifications that enable personalized medicine.
Computational modeling and ML have been used to study AF, offering “increasingly detailed data that could be used by AI to classify AF and personalize therapy for patients”. Among the reported approaches is multiscale computer modeling to identify specific ablation targets in individuals. Image-based characterization of AF for ablation has integrated imaging with cellular computational modeling. Late gadolinium-enhanced (LGE) magnetic resonance imaging (MRI) has emerged as an important imaging modality to visualize fibrosis and objectively assess scars.
An analysis showed that unsupervised clustering may improve phenotypic classification of AF to aid clinical evaluation and management, and in the MADIT-CRT trial, the integration of complex imaging data with standard clinical variables for unsupervised phenotyping was demonstrated on heart failure patients.
Future directions and conclusions
The review emphasized the “great preliminary promise” of AI in cardiac electrophysiology while underlining significant needs in “basic and translational research, and institution-level improvement in data collection and harmonization practices, and clinical validation and practical implementation”.
By performing “expert-level tasks” and extending “capabilities beyond human cognition”, AI is “in a position to alter the landscape of biomedical research and clinical practice”. The review sees the empowerment of researchers and clinicians with AI and ML literacy as beneficial to the increasing adoption of the methods into modern use.
To learn more about the advanced capabilities of machine learning in cardiovascular imaging, please read our article on ‘Can ML Help To Predict Long Term Mortality In TAVI Patients?’.