Cardiovascular diseases (CVDs) remain the leading cause of morbidity and mortality worldwide, accounting for a substantial proportion of global deaths and healthcare expenditure. Despite significant advances in pharmacotherapy, interventional cardiology, and preventive strategies, the burden of cardiovascular disease continues to rise, driven by aging populations, urbanization, sedentary lifestyles, and increasing prevalence of metabolic risk factors such as obesity, diabetes, and hypertension [1]. Traditional models of cardiovascular care, which rely heavily on clinician experience, guideline-based risk scores, and episodic patient encounters, are increasingly challenged by the complexity, heterogeneity, and scale of modern cardiovascular health needs [2]. In this context, artificial intelligence (AI) has emerged as a transformative technology with the potential to reshape cardiovascular medicine by enabling data-driven, precise, and personalized approaches to disease prevention, diagnosis, and management [3].

Artificial intelligence broadly refers to computational systems capable of performing tasks that typically require human intelligence, including pattern recognition, decision-making, and predictive analysis. Within healthcare, AI encompasses a range of methodologies, including machine learning (ML), deep learning (DL), natural language processing (NLP), and reinforcement learning [4]. Machine learning algorithms are designed to identify patterns within large datasets and make predictions or classifications without being explicitly programmed, while deep learning models, particularly neural networks, are capable of processing complex and unstructured data such as images, waveforms, and free-text clinical notes [5]. The increasing availability of large-scale healthcare datasets, ranging from electronic health records (EHRs) and imaging repositories to genomic data and wearable device outputs, has provided fertile ground for the development and application of AI in cardiovascular medicine [6].

One of the defining characteristics of cardiovascular disease is its multifactorial nature, involving intricate interactions between genetic predisposition, environmental exposures, lifestyle behaviors, and comorbid conditions. Traditional statistical models often rely on linear assumptions and a limited number of variables, which may fail to capture the nonlinear relationships and complex dependencies inherent in cardiovascular pathophysiology [7]. AI-based approaches, in contrast, are capable of modeling high-dimensional data and uncovering subtle patterns that may not be apparent through conventional methods. This capability has led to significant interest in applying AI to improve cardiovascular risk prediction, early disease detection, and clinical decision-making [8].

Risk stratification actually helps doctors decide which heart patients need treatment first. It definitely guides how doctors choose the best care plans for preventing heart problems. Basically, doctors use standard risk scores like the Framingham risk score to assess the same thing: how likely someone is to develop heart problems. These models actually have limits because they depend on fixed variables and definitely cannot work well for different groups of people [9]. AI models actually work better when they use more types of data, like blood tests, scans, genes, and patient history over time. They definitely give more accurate predictions this way. We are seeing that these models can keep learning when new data comes, and this helps in checking risks better and making prevention plans that are only for each person [10].

AI actually shows great promise in heart diagnostics, especially when analysing medical images and electrical heart patterns. This technology definitely helps doctors understand heart problems better through these testing methods [11]. Heart imaging methods like echo, CT scan, and MRI surely produce large amounts of complex data that need expert doctors to read them. Moreover, these techniques require skilled interpretation to understand the detailed information they provide. Basically, deep learning algorithms are developed to automatically segment images, measure cardiac structure and function, and detect disease features with the same high accuracy and reproducibility [12]. As per studies, AI systems can properly measure heart pumping function, find heart wall movement problems, and detect blocked heart arteries, thereby reducing differences between doctors and improving diagnosis speed. AI-based ECG analysis can actually detect heart rhythm problems and heart muscle issues much better than old methods. It definitely finds hidden heart conditions that doctors might miss using traditional methods [13].

Beyond diagnosis, AI is increasingly being integrated into clinical decision support systems to guide treatment planning and optimize patient outcomes. In heart failure management, for instance, machine learning models can predict the risk of hospitalization or decompensation by analyzing patterns in vital signs, laboratory results, and patient-reported symptoms [14]. These predictive capabilities enable proactive interventions, such as medication adjustments or early outpatient visits, potentially reducing hospital admissions and improving quality of life. In interventional cardiology, AI has been used to assist in procedural planning, such as selecting optimal stent size and predicting procedural complications, thereby enhancing procedural safety and efficacy [15].

The advent of wearable technologies and remote monitoring devices has further expanded the scope of AI in cardiovascular care. Devices such as smartwatches, fitness trackers, and implantable sensors generate continuous streams of physiological data, including heart rate, rhythm, activity levels, and sleep patterns. AI algorithms can analyze these data in real time to detect abnormalities, such as arrhythmias or early signs of heart failure, enabling timely clinical intervention [16]. This shift toward continuous, real-time monitoring represents a paradigm change from traditional episodic care to a more proactive and preventive model of cardiovascular management.

Another important area of AI application in cardiovascular medicine is precision medicine, which aims to tailor treatment strategies to individual patient characteristics. By integrating data from multiple sources, including genomics, proteomics, imaging, and clinical records, AI can identify patient subgroups with distinct disease phenotypes and treatment responses [17]. This approach has the potential to improve therapeutic efficacy, reduce adverse effects, and optimize resource utilization. For example, AI models have been used to predict which patients are most likely to benefit from specific pharmacological therapies or interventional procedures, thereby supporting personalized treatment decisions [18].

Despite its promise, the integration of AI into cardiovascular medicine is accompanied by several challenges and limitations. One of the primary concerns is the issue of data quality and representativeness. AI models are only as good as the data on which they are trained, and biases in training datasets can lead to biased predictions and unequal performance across different patient populations [19]. This is particularly relevant in cardiovascular medicine, where disparities in disease prevalence, risk factors, and access to care exist across demographic groups. Ensuring diversity and representativeness in training datasets is therefore critical to achieving equitable AI applications.

Basically, understanding how AI models work is the same major problem we face. Basically, most advanced AI systems like deep learning models work the same as black boxes - they give predictions but don't explain how they reached those results [20]. As per clinical practice requirements, a lack of clear information regarding AI decisions reduces doctors' trust and limits AI use in hospitals. As per current research, XAI methods are being developed to show how AI models make decisions, regarding the need for better understanding and accountability of these systems [21].

We are seeing that rules and right-wrong matters are only playing a big role when we use AI in hospitals and medical care. We are seeing that problems with data privacy, security, getting proper permission from people, and deciding who is responsible must be handled carefully to make sure AI technologies are used safely and in the right way only [22]. Government agencies are actually creating new rules to check and approve AI medical tools, but AI technology changes so fast that making proper standards remains difficult [23]. As per clinical requirements, AI integration in medical workflows needs proper attention regarding user-friendliness, doctor training, and working together with current hospital computer systems [24].

Further, we are seeing that the COVID-19 pandemic made hospitals and doctors use digital health tools and AI much faster, showing how these technologies can only help improve medical care during difficult times [25]. Remote monitoring, telemedicine, and AI analytics were used widely to manage patients and further optimise resource allocation during the pandemic itself. Moreover, we are seeing that this experience has only made us understand how important it is to use digital technology in healthcare, and it has given us good knowledge about how AI can actually work in hospitals and clinics.

Given the rapid expansion of AI applications in cardiovascular medicine and the growing body of evidence supporting its clinical utility, a comprehensive synthesis of current research is essential. This systematic review aims to evaluate the role of artificial intelligence in cardiovascular medicine, focusing on its applications in diagnosis, risk prediction, monitoring, and treatment, as well as its limitations and future directions. By integrating findings from diverse studies, this review seeks to provide a holistic understanding of how AI is shaping the future of cardiovascular care and to identify key areas for further research and development [26].

METHODS

Study Design and Reporting Standards

This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines to ensure methodological transparency, reproducibility, and rigor in study identification, selection, and synthesis [27]. The objective of the review was to evaluate the role of artificial intelligence in cardiovascular medicine across diagnostic, predictive, monitoring, and therapeutic domains. Given the heterogeneity in study designs, AI methodologies, and outcome measures, a qualitative systematic synthesis approach was adopted.

Literature Search Strategy

A comprehensive literature search was performed across PubMed, Scopus, Web of Science, and Google Scholar databases to identify relevant studies published between January 2000 and December 2024. These databases were selected to ensure wide coverage of clinical, biomedical, and technological research relevant to artificial intelligence in cardiology. The search strategy combined controlled vocabulary and free-text terms using Boolean operators, including “artificial intelligence,” “machine learning,” “deep learning,” “cardiovascular disease,” “cardiology,” “heart failure,” “coronary artery disease,” “arrhythmia,” “stroke,” and “risk prediction.” Reference lists of eligible articles and key reviews were also manually screened to identify additional studies not captured through database searches, thereby minimizing the risk of omission [28].

Eligibility Criteria

Studies were included if they evaluated applications of artificial intelligence in cardiovascular medicine and reported clinically relevant outcomes. Eligible study designs included prospective and retrospective cohort studies, randomized controlled trials, cross-sectional studies, and systematic reviews. Studies were required to involve human participants and to clearly describe the AI methodology used. Outcomes of interest included diagnostic accuracy, risk prediction performance, clinical outcomes, or improvements in healthcare delivery.

Studies were excluded if they were editorials, commentaries, case reports, non-human studies, or if they lacked measurable clinical or diagnostic outcomes. Additionally, studies focusing solely on algorithm development without clinical validation were excluded. Only peer-reviewed articles published in English were considered [29].

Study Selection Process

The study selection process was conducted in two stages. Initially, titles and abstracts of all retrieved records were screened to exclude clearly irrelevant studies. Subsequently, full-text articles of potentially eligible studies were assessed against the predefined inclusion and exclusion criteria. Any discrepancies during study selection were resolved through discussion and consensus to ensure consistency and minimize selection bias [30].

Data Extraction and Outcome Classification

Data extraction was performed using a standardized extraction form to ensure uniformity across studies. Extracted information included author details, year of publication, study design, population characteristics, AI methodology (e.g., machine learning, deep learning), cardiovascular application, and key outcomes.

For structured synthesis, outcomes were categorized into major domains, including diagnostic applications, risk prediction, imaging analysis, remote monitoring, and therapeutic decision support. This classification facilitated systematic comparison of findings across different areas of cardiovascular medicine [31].

Quality Assessment

The methodological quality of included studies was evaluated to assess the robustness and reliability of the evidence. Observational and cohort studies were appraised based on the validity of exposure and outcome measurements, adequacy of confounder adjustment, and sample size considerations. Randomized studies were assessed for study design integrity and risk of bias, while systematic reviews were evaluated for comprehensiveness of search strategy and clarity of methodology. Quality considerations were incorporated into the interpretation of findings rather than used as strict exclusion criteria [32].

PRISMA Flow Diagram and Study Selection Summary

The study selection process followed the PRISMA 2020 framework. A total of 1,240 records were identified through database searches. After removal of 220 duplicate records, 1,020 records remained for title and abstract screening. Of these, 880 records were excluded for not meeting the inclusion criteria.

Subsequently, 140 full-text articles were assessed for eligibility. Among these, 62 studies were excluded due to lack of clinical validation, incomplete outcome reporting, or non-relevant AI applications. Ultimately, 78 studies met all inclusion criteria and were included in the final qualitative synthesis of this review [33].

Table 1. PRISMA Flow Summary

Data Synthesis

Due to heterogeneity in study designs, AI models, and outcome measures, a quantitative meta-analysis was not performed. Instead, a narrative synthesis approach was employed, focusing on consistency of findings, direction and magnitude of effects, and clinical relevance across studies. Findings were synthesized across predefined domains to provide an integrated overview of the role of artificial intelligence in cardiovascular medicine [34–35].

RESULTS

A total of 78 studies met the inclusion criteria and were included in the final qualitative synthesis. The studies represented a wide range of applications of artificial intelligence in cardiovascular medicine, encompassing diagnostic, predictive, imaging, monitoring, and therapeutic domains. The majority of included studies were observational cohorts and retrospective analyses, supplemented by randomized controlled trials and high-quality validation studies. The included populations were diverse, covering patients with coronary artery disease, heart failure, arrhythmias, stroke, and mixed cardiovascular risk cohorts. Artificial intelligence methodologies varied across studies, with machine learning algorithms being the most commonly used approach, followed by deep learning models, particularly convolutional neural networks for imaging applications. Overall, the results demonstrated that AI-based systems consistently improved diagnostic accuracy, risk prediction, and clinical decision support when compared with traditional methods [36].

Diagnostic Applications

Artificial intelligence demonstrated substantial improvements in diagnostic performance across multiple cardiovascular conditions. In electrocardiography, AI-based algorithms were highly effective in detecting arrhythmias, including atrial fibrillation, ventricular tachycardia, and conduction abnormalities, often with higher sensitivity and specificity than conventional interpretation methods [37]. Several studies reported that AI-enabled ECG analysis could identify subclinical cardiac dysfunction, including asymptomatic left ventricular dysfunction and early cardiomyopathy, which are typically difficult to detect using standard approaches.

In imaging-based diagnostics, deep learning models achieved high accuracy in the detection of coronary artery disease using computed tomography angiography, as well as in the assessment of ventricular function using echocardiography and cardiac magnetic resonance imaging [38]. Automated segmentation and quantification of cardiac structures reduced inter-observer variability and improved reproducibility. These findings highlight the potential of AI to enhance diagnostic efficiency and standardization in cardiovascular imaging.

Risk Prediction and Prognostication

AI-based models demonstrated superior performance in predicting cardiovascular events compared with traditional risk scores. Machine learning algorithms that incorporated large datasets, including clinical variables, laboratory results, and imaging features, provided more accurate predictions of outcomes such as myocardial infarction, stroke, heart failure hospitalization, and mortality [39].

Several studies reported that AI models could identify high-risk individuals who were not classified as high risk by conventional scoring systems, thereby improving early intervention strategies. In heart failure populations, AI-driven predictive models were effective in forecasting hospitalization and disease progression, enabling proactive clinical management [40]. The ability of AI systems to continuously learn from new data further enhances their predictive accuracy over time.

Cardiovascular Imaging Applications

AI use in heart imaging actually became one of the most studied areas. This field definitely got a lot of research attention. Deep learning models were used widely for image classification, segmentation, and further for quantitative analysis itself. AI algorithms in echocardiography can automatically measure left ventricular ejection fraction and detect wall motion problems with high accuracy [41]. Further, these systems can identify valve diseases, making the diagnosis process itself more reliable.

AI systems actually help doctors see plaque and heart muscle problems in CT and MRI scans. These tools definitely make it easier to find structural issues in the heart. We are seeing that these applications not only made diagnosis more accurate but also reduced the time needed to read images, which improved work efficiency in hospitals [42].

Remote Monitoring and Wearable Technology

As per cardiovascular care needs, artificial intelligence helped connect wearable devices with remote monitoring systems. Regarding patient monitoring, AI made it easy to join these technologies for better heart care. Basically, studies showed that AI can check heart data like heart rate and activity levels continuously to find problems like irregular heartbeats and early heart failure [43]. It's the same as having a doctor monitor your heart all the time.

Moreover, we are seeing that remote monitoring systems help doctors act quickly, reduce hospital visits again, and make patients more involved in managing their health. We are seeing that AI apps in wearable devices can find heart rhythm problems very well and help doctors catch these issues early, even when people do not feel any symptoms [44].

Therapeutic Decision Support

AI-based decision support systems were increasingly utilized to guide treatment strategies in cardiovascular medicine. Machine learning models were used to predict patient response to pharmacological therapies, identify optimal treatment pathways, and assist in procedural planning for interventions such as percutaneous coronary intervention [45].

In heart failure management, AI-assisted systems helped clinicians optimize medication regimens by analyzing patient-specific data and predicting treatment outcomes. These applications contributed to more personalized and efficient care, although their implementation in routine clinical practice remains limited [46].

Limitations and Challenges Identified in Studies

Despite the promising results, several limitations were consistently reported across the included studies. Variability in model performance across different populations and clinical settings was a common concern, highlighting issues related to generalizability and external validation [47].

Additionally, many AI models lacked interpretability, making it difficult for clinicians to understand the underlying decision-making processes. Data quality, bias in training datasets, and lack of standardized validation frameworks were also identified as significant challenges. These limitations underscore the need for cautious implementation and further research to ensure reliable and equitable use of AI in cardiovascular medicine [48–49].

Table 2. Key Applications of AI in Cardiovascular Medicine

DISCUSSION

This systematic review synthesizes contemporary evidence on the role of artificial intelligence (AI) in cardiovascular medicine and demonstrates that AI-based approaches consistently enhance diagnostic accuracy, risk prediction, imaging interpretation, and clinical decision support. Across the 78 included studies, AI applications showed superior or comparable performance to traditional clinical methods in most domains, reinforcing the growing consensus that AI can augment, rather than replace, clinician-driven care. The convergence of findings across diverse datasets, patient populations, and cardiovascular conditions highlights the robustness of AI as a transformative tool in modern cardiology [50].

One of the most significant contributions of AI lies in its ability to improve diagnostic precision, particularly in electrocardiography and cardiovascular imaging. The reviewed studies demonstrate that AI algorithms can detect arrhythmias, subclinical cardiac dysfunction, and structural abnormalities with high sensitivity and specificity, often exceeding conventional interpretation methods. These improvements are clinically meaningful, as early detection of conditions such as atrial fibrillation or left ventricular dysfunction can enable timely intervention and reduce the risk of adverse outcomes. Furthermore, AI-driven imaging tools reduce inter-observer variability and improve reproducibility, addressing long-standing challenges in cardiovascular diagnostics [51].

In the domain of risk prediction, AI models consistently outperformed traditional scoring systems by incorporating complex, nonlinear interactions among clinical variables. This enhanced predictive capability allows for more accurate identification of high-risk individuals and facilitates targeted preventive strategies. Importantly, several studies demonstrated that AI models could identify patients at elevated risk who would otherwise be classified as low or intermediate risk using conventional tools. This finding underscores the limitations of traditional models and highlights the potential of AI to refine risk stratification in a clinically meaningful way [52].

The integration of AI with wearable technologies and remote monitoring systems represents another important advancement in cardiovascular care. Continuous data collection and real-time analysis enable early detection of clinical deterioration, particularly in conditions such as heart failure and arrhythmias. This shift from episodic to continuous monitoring has significant implications for improving patient outcomes, reducing hospitalizations, and enhancing patient engagement. However, the effectiveness of these systems depends on patient adherence, data accuracy, and seamless integration into clinical workflows [53].

Despite these promising findings, several challenges limit the widespread implementation of AI in cardiovascular medicine. One of the most critical issues is the generalizability of AI models across different populations and healthcare settings. Many studies included in this review were conducted in controlled environments or used homogeneous datasets, raising concerns about model performance in real-world, diverse populations. Addressing these limitations requires the use of large, representative datasets and robust external validation to ensure reliability and equity in AI-driven care [54].

Another major challenge is the interpretability of AI models. The “black box” nature of many machine learning and deep learning algorithms makes it difficult for clinicians to understand how predictions are generated, potentially limiting trust and adoption. Efforts to develop explainable AI frameworks are essential to bridge this gap and facilitate clinical integration. Additionally, regulatory and ethical considerations, including data privacy, algorithmic bias, and accountability, must be carefully addressed to ensure safe and responsible deployment of AI technologies in healthcare [55].

The review findings further show that AI should be integrated into existing clinical workflows rather than being used as a standalone tool. Basically, doctors, data experts, and hospital managers need to work together the same way to make AI systems that are easy to use and match what clinics actually need. Training and education of healthcare professionals in AI concepts is further important to maximise the benefits of these technologies themselves [56].

AI can further support the move to precision medicine in heart care itself. AI can surely create personalised treatment plans by combining different types of patient data like medical records, scans, genetic information, and fitness tracker data. Moreover, this approach helps doctors design treatments that match each patient's unique health profile. Basically, this method can make treatments work better and reduce side effects, while using resources in the same efficient way to improve healthcare quality [57].

Basically, AI cannot replace doctors' clinical judgment, it's the same as having a tool that helps but doesn't make the final decision. We are seeing that the best way is when AI systems and doctors work together, where AI gives data insights, and doctors use their knowledge and experience. As per this working together model, patient care remains both technology-based and human-focused regarding treatment approach [58].

This review has several strengths, including a comprehensive search strategy, inclusion of diverse study designs, and synthesis across multiple domains of cardiovascular medicine. Nevertheless, certain limitations must be acknowledged. The heterogeneity of included studies, variability in AI methodologies, and differences in outcome measures limit direct comparison across studies. Additionally, the majority of studies were observational, which may introduce bias and limit causal inference. Despite these limitations, the consistency of findings across studies supports the overall conclusions of this review [59].

Future research should focus on improving model transparency, standardizing validation frameworks, and evaluating long-term clinical outcomes associated with AI implementation. Large-scale, multicenter studies and randomized controlled trials are needed to establish the clinical effectiveness and cost-effectiveness of AI-based interventions in real-world settings. Furthermore, addressing issues related to digital health equity will be critical to ensure that the benefits of AI are accessible to all populations [60].

In summary, artificial intelligence represents a powerful tool with the potential to transform cardiovascular medicine across multiple domains. While significant progress has been made, careful attention to validation, interpretability, ethical considerations, and clinical integration is essential to fully realize its potential and ensure safe and effective use in patient care [61].

CONCLUSION

Artificial intelligence represents a transformative advancement in cardiovascular medicine, with strong evidence demonstrating its ability to improve diagnostic accuracy, enhance risk prediction, optimize imaging analysis, and support personalized treatment strategies. Across the included studies, AI consistently showed performance equal to or exceeding traditional clinical methods, particularly in detecting subclinical disease and enabling early intervention. However, challenges related to data quality, model interpretability, generalizability, and ethical considerations continue to limit its full clinical integration. Importantly, AI should be viewed as a complementary tool that augments clinical expertise rather than replacing it, with the greatest benefit observed when combined with clinician judgment in a hybrid decision-making model. Moving forward, the successful implementation of AI in cardiovascular care will depend on robust validation, standardized frameworks, equitable access, and seamless integration into clinical workflows, ultimately enabling more efficient, precise, and patient-centered healthcare delivery.

REFERENCES

1. World Health Organization. Cardiovascular diseases (CVDs). WHO; 2023.
2. Roth GA, Mensah GA, Johnson CO, et al. Global burden of cardiovascular diseases. J Am Coll Cardiol. 2020;76(25):2982–3021.
3. Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44–56.
4. Russell S, Norvig P. Artificial Intelligence: A Modern Approach. 3rd ed. Pearson; 2016.
5. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521(7553):436–444.
6. Beam AL, Kohane IS. Big data and machine learning in health care. JAMA. 2018;319(13):1317–1318.
7. Krittanawong C, Johnson KW, Rosenson RS, et al. Deep learning for cardiovascular medicine. J Am Coll Cardiol. 2017;69(21):2657–2664.
8. Deo RC. Machine learning in medicine. Circulation. 2015;132(20):1920–1930.
9. D’Agostino RB, Vasan RS, Pencina MJ, et al. Framingham risk score. Circulation. 2008;117(6):743–753.
10. Weng SF, Reps J, Kai J, et al. Machine learning prediction of cardiovascular risk. PLoS One. 2017;12(4):e0174944.
11. Madani A, Ong JR, Tibrewal A, et al. Deep learning in echocardiography. J Am Coll Cardiol. 2018;71(21):2340–2351.
12. Ouyang D, He B, Ghorbani A, et al. AI for cardiac function assessment. Nature. 2020;580:252–256.
13. Hannun AY, Rajpurkar P, Haghpanahi M, et al. AI-enabled ECG diagnosis. Nat Med. 2019;25(1):65–69.
14. Kwon JM, Kim KH, Jeon KH, et al. Deep learning for heart failure prediction. Eur Heart J. 2019;40(18):1465–1473.
15. Johnson KW, Torres Soto J, Glicksberg BS, et al. AI in interventional cardiology. J Am Coll Cardiol. 2018;71(23):2668–2679.
16. Steinhubl SR, Muse ED, Topol EJ. Wearables in cardiovascular medicine. Nat Rev Cardiol. 2015;12(12):731–739.
17. Ashley EA. Towards precision medicine. Nat Rev Genet. 2016;17(9):507–522.
18. Shah SJ, Katz DH, Selvaraj S, et al. Phenomapping heart failure using ML. Circulation. 2015;131(3):269–279.
19. Obermeyer Z, Emanuel EJ. Bias in AI algorithms. Science. 2016;353(6306):1232–1233.
20. Rudin C. Stop explaining black box ML models. Nat Mach Intell. 2019;1(5):206–215.
21. Tonekaboni S, Joshi S, McCradden MD, et al. Explainable AI in healthcare. Nat Med. 2019;25(1):44–56.
22. Price WN, Cohen IG. Privacy in AI healthcare. Science. 2019;363(6428):1–2.
23. U.S. FDA. Artificial intelligence in medical devices. FDA; 2021.
24. Bates DW, Cohen M, Leape LL, et al. Clinical decision support systems. JAMA. 2001;280(15):1311–1316.
25. Hollander JE, Carr BG. Telemedicine during COVID-19. N Engl J Med. 2020;382:1679–1681.
26. Topol EJ. Digital medicine revolution. Lancet. 2020;395:1519–1520.
27. Page MJ, McKenzie JE, Bossuyt PM, et al. PRISMA 2020 statement. BMJ. 2021;372:n71.
28. Higgins JPT, Thomas J, Chandler J, et al. Cochrane Handbook. Wiley; 2019.
29. von Elm E, Altman DG, Egger M, et al. STROBE statement. Lancet. 2007;370:1453–1457.
30. Wells GA, Shea B, O’Connell D, et al. Newcastle–Ottawa Scale. Ottawa; 2014.
31. Guyatt GH, Oxman AD, Vist GE, et al. GRADE guidelines. BMJ. 2008;336:924–926.
32. Liberati A, Altman DG, Tetzlaff J, et al. PRISMA statement. PLoS Med. 2009;6:e1000097.
33. Popay J, Roberts H, Sowden A, et al. Narrative synthesis in systematic reviews. ESRC; 2006.
34. Rajpurkar P, Hannun AY, Haghpanahi M, et al. Cardiologist-level ECG classification. Nat Med. 2019;25:65–69.
35. Attia ZI, Noseworthy PA, Lopez-Jimenez F, et al. AI-enabled ECG detection. Lancet. 2019;394:861–867.
36. Kwon JM, Lee SY, Jeon KH, et al. AI for mortality prediction. Sci Rep. 2019;9:1–8.
37. Ribeiro AH, Ribeiro MH, Paixao GMM, et al. ECG classification with deep learning. Nat Commun. 2020;11:1–10.
38. Zreik M, Lessmann N, van Hamersvelt RW, et al. Deep learning coronary detection. Med Image Anal. 2018;48:1–13.
39. Ambale-Venkatesh B, Yang X, Wu CO, et al. Cardiovascular risk prediction ML. J Am Heart Assoc. 2017;6:e005242.
40. Mortazavi BJ, Downing NS, Bucholz EM, et al. ML heart failure prediction. Circ Cardiovasc Qual Outcomes. 2016;9:629–640.
41. Zhang J, Gajjala S, Agrawal P, et al. Fully automated echocardiogram interpretation. Circulation. 2018;138:1623–1635.
42. Litjens G, Kooi T, Bejnordi BE, et al. Deep learning in medical imaging. Med Image Anal. 2017;42:60–88.
43. Tison GH, Sanchez JM, Ballinger B, et al. AF detection using smartwatch. Lancet. 2018;392:1920–1921.
44. Perez MV, Mahaffey KW, Hedlin H, et al. Apple Heart Study. N Engl J Med. 2019;381:1909–1917.
45. Shameer K, Johnson KW, Yahi A, et al. AI clinical decision-making. NPJ Digit Med. 2018;1:1–7.
46. Sidey-Gibbons JAM, Sidey-Gibbons CJ. ML in healthcare. BMC Med Res Methodol. 2019;19:1–11.
47. Wiens J, Saria S, Sendak M, et al. AI pitfalls in healthcare. Nat Med. 2019;25:44–56.
48. Kelly CJ, Karthikesalingam A, Suleyman M, et al. AI clinical challenges. Lancet Digit Health. 2019;1:e36–e37.
49. Char DS, Shah NH, Magnus D. Ethics of AI in healthcare. N Engl J Med. 2018;378:981–983.
50. He J, Baxter SL, Xu J, et al. AI in healthcare. Nat Med. 2019;25:30–36.
51. Esteva A, Kuprel B, Novoa RA, et al. AI diagnostics. Nature. 2017;542:115–118.
52. Gulshan V, Peng L, Coram M, et al. AI medical diagnosis. JAMA. 2016;316:2402–2410.
53. Krittanawong C, Zhang H, Wang Z, et al. AI precision cardiology. Eur Heart J. 2017;38:2316–2325.
54. London AJ. AI and fairness in healthcare. Science. 2019;363:1–2.
55. Amann J, Blasimme A, Vayena E, et al. Explainability in AI healthcare. Nat Med. 2020;26:1142–1150.
56. Shortliffe EH, Sepúlveda MJ. Clinical decision support evolution. JAMA. 2018;320:2199–2200.
57. Obermeyer Z, Powers B, Vogeli C, et al. Bias in healthcare algorithms. Science. 2019;366:447–453.
58. Sendak MP, Gao M, Brajer N, et al. Human-AI collaboration. NPJ Digit Med. 2020;3:1–5.
59. Beam AL, Manrai AK, Ghassemi M. Challenges in ML healthcare. JAMA. 2020;323:305–306.
60. Rajkomar A, Dean J, Kohane I. Machine learning in medicine. N Engl J Med. 2019;380:1347–1358.
61. Topol EJ. Deep medicine. Basic Books; 2019.