The Emergence of Artificial Intelligence in Cardiology: Current and Future Applications

Abstract

Artificial intelligence technology is emerging as a promising entity in cardiovascular medicine, potentially improving diagnosis and patient care. In this article, we review the literature on artificial intelligence and its utility in cardiology. We provide a detailed description of concepts of artificial intelligence tools like machine learning, deep learning, and cognitive computing. This review discusses the current evidence, application, prospects, and limitations of artificial intelligence in cardiology.

1. INTRODUCTION

Cardiovascular diseases (CVDs) are the primary cause of morbidity and mortality in the world. According to the World Health Organization (WHO), an estimated 17.9 million people die each year from CVDs, accounting for 31% of all deaths across the globe [1]. Mitigating strategies must be devised to provide optimal and cost-effective care for CVDs.

There has been an exponential increase in the generation of health care data over the past decade. According to some sources, the growth of new data is almost 48% each year, overwhelming the physician and changing healthcare services’ dynamics [2]. Therefore, this large volume of data must be harnessed for research to formulate preventive measures and provide efficient patient care [2]. This availability of unlimited data makes it difficult for the health care professional to assimilate it and take a prompt errorless decision [3]. Computer systems have the inherent capabilities to store big data, process this information intelligently, and arrive at quick conclusions [4]. Artificial intelligence (AI) and its tools like machine learning can aid cardiology in reducing the physician’s burden and dispense precise, rapid, and personalized care to the patients [5,6]. Machine learning can analyze extensive complex data and provide a solution that the clinician can use for making accurate prognostication [7].

The AI tools are making their impact in the rapidly evolving field of cardiology. Cardiovascular medicine services must appreciate the capabilities of this technology to make efficient use of it in the coming years. With the increased usage of AI for solving complex tasks and quantifying data, it is predicted to usher into improved outcomes for the patients, cardiologists, health services, insurance providers, and the executive powers worldwide [8]. AI has a huge potential to aid physicians in medical diagnosis, disease treatment, drug discovery, and risk prediction [9].

We present a literature review on the role of AI in cardiology, focusing on the basic concept of AI tools and their current application in cardiovascular sciences. Lastly, we discuss future directions and challenges in AI adoption.

2. ARTIFICIAL INTELLIGENCE

2.1. Concepts of Artificial Intelligence

Simply put, artificial intelligence is a science that enables human-like intellect into machines [10]. It comprises machine learning, deep learning, and cognitive computing to equip machines with intelligence [11]. The intelligence here is synthesized to mimic the natural human decision-making process and take action, as a human would do.

For a machine to clone human intelligence and behaviour, it needs:

1. Contextual knowledge

2. Variety of situational solutions

3. Enables one to make decisions based on the situation posed.

2.1.1. Machine Learning

Machine learning (ML) is the crux of AI. This technology solves complex big data tasks by recognizing variables and providing a model based on future accurate, predictable data [12]. Machine learning is categorized into supervised, unsupervised, and reinforcement types. In supervised learning, the machine algorithms utilize data labelled by humans

to predict and achieve the desired outcome [13]. Supervised learning is also vital in the operation of many biological and artificial neural networks [14].

In unsupervised learning, the machine finds the patterns from the hidden data without any human input. These learning algorithms have been applied to cardiovascular disease prediction, diagnosis, care, and cardiovascular image interpretation. The drawbacks of unsupervised learning are that the primary cluster pattern needs to be unbiased and validated with other cohorts.

Reinforcement learning is a combination of supervised and unsupervised learning. In this type of learning, the aim is to augment the precision of algorithms empirically.

2.1.2. Deep Learning

Deep learning is a powerful upgrade of machine learning technology [15, 16], which utilizes artificial neuronal networks and mimics the human brain’s operations to generate automated predictions from the inserted data. This technology has enhanced speech and visual object recognition, object detection, drug invention, and genomics, along with significant breakthroughs in processing image, video, voice, and audio data [17].

2.1.3. Cognitive Computing

In cognitive computing using the machine or deep learning algorithms, a system or a device creates automated computerized problem-solving models mimicking the human thought process.

2.2. The Pillars of the AI Ecosystem

The AI tools, i.e., machine learning, deep learning, and cognitive computing, form the AI ecosystem’s pillars [18] (Fig. 1). The various outcomes by the integration of these methods in AI are enumerated as follows:

Fig. (1).

Pillars of AI ecosystem.

2.2.1. Data

In order to program the machine to arrive at an accurate decision, many data sets are required to make better predictions.

a. Volume- comprises data from electronic health records (EHR), i.e., historical clinical data, patient data, demographic data, images, audios, scan videos, doctor prescriptions, pharmaceutical reference & drug data.

b. Velocity - requires real-time feeds from individuals and geographical conditions, news on epidemics, and regular updates from clinical references.

c. Variety- the variable formats of the data sets based on the provider, source of data, and the stream.

d. Veracity- While there is a flood of data flowing around, it is essential to ascertain that the data being used is of high quality, catering to the dimensions of accuracy, completeness, latency, and sufficiency parameters.

2.2.2. Training the Data Set

The crux of AI, i.e., machine learning, lies in the training of models, which analyzes this multi-stream data and classifies and prepares it for the models to consume. The technologies of image processing, natural language processing, text mining are used to arrive at meaningful synopsis of the data sets. Many classifications/supervised and clustering/unsupervised techniques are tested for accuracy before arriving at the suitable model applicable for use.

2.2.3. Engine

The heart of AI is the engine based on situational data, utilizes the knowledge gained and applies algorithms to arrive at the best possible outcome. After every prediction, it should personalize the results to the patient and the doctor before concluding on the outcome.

2.2.4. Action

For precision action, the AI system communicates to humans using alerts (sounding off threshold attainment) and provides a prescriptive analysis (situation-based options available for the doctor to choose), resulting in analytical guidance on the physician’s case and informed decision.

The technological advancement available today enables all these functions of AI to work in tandem:

- Big data technologies and memory based processing helps in managing the data sets efficiently.

- Image recognition (IR) technologies, natural language processing (NLP), and text mining help data preparation.

- Off-the-shelf models in a multitude of languages assist in quick availability in the market.

- Communication methods such as message-based alerting, chat-bots, live assistance, and links to multiple health apps can be utilized to make a direct connection to the hospital. It can also be used for auto-dialling of ambulance services in case of a medical emergency.

Lastly, building an efficient AI ecosystem needs the integrated expertise of the doctor, technologist, and data scientist.

3. AI IN CARDIOLOGY

Cardiology is where precision, accuracy, and quick response times are critical for the patient’s survival. Apart from being a lifesaver, the physician’s responsibility is to understand his patient aptly and guide him for his overall well-being. All these can be achieved with a well-rounded solution of artificial intelligence which assists the cardiologist to be adept, nimble, and empathetic to the larger social community.

The critical applications of AI in the field of cardiology are summarized below, (Fig. 2).

Fig. (2).

AI in cardiology.

3.1. Prediction of Fatality

Based on the streaming data of the individuals from health apps, wearable devices, and patient scan images, AI is used to rightly predict the event of a fatality. These utilities are possible by applying algorithms to machine learning and big data analytics.

Researchers from Cedars-Sinai Medical Center in the USA conducted a 15-year prospective ESINER trial in 1912 asymptomatic subjects and used machine learning to assess the risk of myocardial infarction and cardiac death [19]. In this study, the ML integration of clinical parameters with coronary artery calcium and automated epicardial adipose tissue quantification was done, and 76 subjects presented an event of myocardial infarction or cardiac death during the follow-up period of 15 years. The researchers concluded that compared to standard risk assessment, machine learning integration of clinical risk factors and imaging measures could accurately predict the patient’s risk of suffering an adverse event, such as heart attack or cardiac death [19].

3.2. Personalized Prescription and Precision Medicine

AI can be utilized to provide prescriptive analytics to the physician based on the patient’s unstructured EHR dataset, i.e., age, gender, ethnicity, family history, demographic location, lifestyle, past treatment history, and health vitals. The cardiologist could utilize AI to diagnose the disease and implement an accurate, personalized medical plan for each patient. Thus, AI can reduce the physician’s burden and help him provide more precise care to the patient. Health care professionals should get familiarized and trained in AI technology, as they can apply AI to analyze big data and leverage this to improve cardiovascular disease diagnosis and treatment. Preferably precision medicine should utilize more AI to provide tailor-made health solutions for each patient.

3.3. A Personal Assistant to the Physician

AI can provide guidance and alerts to the doctor about the patient, such as knowledge about possible drug interactions and allergies. The physician can use this for planning rapid pre-emptive interventions on the patient. AI has the potential to reduce undue variation in clinical practice, improve deliverance and prevent avoidable medical errors that affect most patients during their lifetime [20]. Many AI concept validation studies have aimed to improve the clinical workflow, including automatic extraction of logical information from transcripts [21], predicting failure of the patient to attend hospital appointments [22], summarizing patient’s doctor consultations [23], and even recognizing speech in doctor-patient interactions [24].

3.4. Police Individual Health

AI can be used for real-time monitoring of individual data and co-relate it clinically with possible outcomes. Personalized messages to the patient’s wearable sensors and devices can be sent with guidance on medications, preventive measures and recommend specialized treatment facilities in the geographical vicinity of the individual in case of a health emergency. This technology can help detect arrhythmias in susceptible individuals [25, 26]. Advances in consumer technology can be leveraged to plan studies; an example is the ongoing prospective study to detect atrial fibrillation in 419,093 individuals owning a commercially available smartwatch [27].

3.5. Cardiovascular Imaging

Cardiac imaging is one area in which machine learning and deep learning methods have been extensively harnessed and demonstrated promising outcomes [28]. AI technology has been used to automate image quality control, acquire and reconstruct an image more efficiently, and image segmentation. These algorithms are also used for myocardial motion, coronary artery blood flow analysis, and computer-aided diagnosis [29, 30]. Deep learning technology has been used for analyzing coronary calcium scoring from computer tomography scans [31]. In the CONFIRM registry, an analysis of 13 054 participants using an ML algorithm incorporating clinical variables in addition to coronary artery calcium score (CACS) derived from coronary computed tomography angiography (CCTA) could accurately estimate the pre-test probability of obstructive coronary artery disease (CAD) on CCTA [32]. ML technology has been shown to accurately detect obstructive CAD on CCTA better than the clinician’s visual assessment of significant CAD on the reconstructed CCTA images [33, 34]. Radiomic based ML algorithms can be used for phenotyping of coronary lesions on CCTA [34]. High risk coronary atheromatous plaque features and coronary micro-calcification and inflammation on CCTA can be more accurately identified by radiomic based ML platforms than the visual interpretation by the clinician [35 - 41].

Hemodynamic assessment of a coronary lesion is done by the invasive fractional flow reserve (FFR) on cardiac catheterization or non-invasively by estimating myocardial flow reserve on ¹³N-Ammonia positron emission tomography (PET) [42,43]. In contrast to FFR and PET, ML-derived algorithms accurately detected hemodynamically significant obstructive coronary lesions [42,43] and are comparable with a complex computation fluid dynamic modelling method to detect obstructive lesions [44,45].

AI-enabled CT scan can be used to detect myocardial scarring due to myocarditis and myocardial infarction accurately and left ventricular dilation and systolic function in patients with ventricular tachycardia [46-49].

Deep learning algorithms have enabled cardiac magnetic resonance imaging (CMR) for automated image segregation and evaluated myocardial motion analysis for human survival prediction [50]. Deep learning technology is adept at image recognition and used in cardiovascular imaging modalities like two- and three-dimensional echocardiography (2D and 3D), especially in speckle-tracking echocardiography and strain imaging data. Commercially available major imaging companies are equipping the ultrasound system with AI to help analyze echocardiogram images for better output and quality. One vendor has used ML 3D echo dataset acquisition algorithms to automatically analyze the variation in the acquired cardiac anatomy image slices, optimize them and choose the best standards views for presentation. It would take many years for an echocardiographer to gather similar information.

3.6. Robotic Procedures

AI has massive potential in cardiac procedures like percutaneous coronary intervention (PCI) and electrophysiology (EP) for catheter ablation of cardiac arrhythmias. AI models can be used to perform high precision interventions by auto manoeuvring, improving patient safety, and reducing radiation exposure to health care professionals.

In December 2018, the world’s first-in-human telerobotic PCI was performed in India utilizing vascular robotic technology, and this intervention was conducted from a remote location outside the catheterization laboratory [51]. This technology has a vast potential to deliver precision cardiac interventional procedures to patient populations with poor access to cardiovascular services [51]. Other commercially available cardiovascular robotic systems have created a clinical database of successfully performed robotic PCI interventions across Africa and Europe. This registry allows subsequent devices to provide safer and precision vascular interventional procedures [52].

Robotic electrophysiology uses Stereotaxis technology that employs magnetic catheter tips and imaging for precise and safer navigation during EP ablation interventions for ventricular arrhythmias and atrial fibrillation [53]. Soon fully automated robotic process in electrophysiology would be available for catheter ablation procedures [54,55].

4. FUTURE APPLICATIONS

AI potentially is the next revolution in cardiovascular sciences [56]. The constraints of current clinical medicine tools and methods should act as a catalyst to spur the increased usage of AI tools in patient care. This technology is bound to grow exponentially, as many large corporates are investing in AI health care start-ups, estimated to be worth USD 25 billion in 2020 [57]. AI is part of the digital transformation occurring across all sections of the economy. The availability of increased digital data sourced from EHR and rapid progress in computer technology and the internet have created a fertile ground for the growth of AI [57]. One such example of harnessing AI is using a commercially available smartwatch sensor to monitor atrial fibrillation in the population, and the findings were comparable to the traditional insertable cardiac monitor arm in the study [58]. Another example of technological advancement of AI is the remote navigation of robotic EP systems for catheter ablation of cardiac arrhythmias, which allows the physician to control the catheter navigation from a comfortable location protected from radiation and reduces operator fatigue. The major areas of cardiovascular care in which AI would soon play a substantial role are cardiac imaging, prediction of risk scores and outcomes, daily decision making, such as diagnosis and treatment, and developing algorithms from large databases mapped with general guidelines. The novel AI approaches in cardiology would help provide rapid, precise, and less erroneous patient care, with substantial clinical and economic implications. The physicians should embrace AI for improving workflow dynamics without the undue fear of being replaced by AI technology [59].

5. CURRENT CHALLENGES WITH AI ADOPTION

The application of AI in health care has specific challenges to overcome [60].

1. AI is still evolving: The solutions in the marketplace have not reached a maturity where the predictions and the outcomes can be relied on with no manual intervention by the physician. These limitations have a bearing on the widespread adoption of AI with confidence. The maturity of the technology, iterating cycle of machine learning, and quality of data sets are some of the contributing factors [60].

2. Substantial investment: Building an AI solution requires extensive infrastructure, various data sets, and related technologies. These require a substantial capital investment which could be challenging for the institution to substantiate the investment. The continuous improvement and enrichment of the AI ecosystem also demand operational expenditure. In many countries, the hospital’s expenditure to enable AI systems could be exponentially more than the patient’s disease treatment expense.

3. Accuracy of the prediction and analysis: For example, images of benign moles were misdiagnosed as malignant by adding adversarial noise or just rotation in one study. Any wrong prediction is going to have a substantial bearing on physician and patient care.

4. Organizational readiness: The extensive quality data to incorporate into AI algorithms may not be readily available. In addition, investments in technology, medical and technical experts need to be available to run a program of such a scale.

5. Privacy concerns: There are ethical concerns that applying AI technologies in research might infringe on an individual’s privacy, and guidelines should be formulated to protect human rights [61].

CONCLUSION

AI has a tremendous potential to redefine cardiology practice. This technology is already showing its impact in the field of cardiovascular imaging and electrophysiology. Cardiologists need to embrace AI to improve workflow efficiency and deliver precision care to patients. At the same time, physicians should be aware of the challenges of AI implementation in clinical practice and research. AI algorithms should undergo vigorous prospective clinical trials before implementing them into cardiology practice. Foremost considerations should be given to the patient’s right to privacy, and ethical policies should be drafted before AI adoption.

CONSENT FOR PUBLICATION

Not applicable.

FUNDING

None.

CONFLICT OF INTEREST

The authors declare no conflict of interest, financial or otherwise.