A Literature Review of the Integration and Application of AI-ECG for Precision Clinical Practices By Jasmine Simone Polimadei A thesis submitted to the School of Radiologic Sciences in collaboration with a research agenda team In partial fulfillment of the requirements for the degree of MASTER OF SCIENCE IN RADIOLOGIC SCIENCES (MSRS) WEBER STATE UNIVERSITY Ogden, Utah April 25, 2025 2 THE WEBER STATE UNIVERSITY GRADUATE SCHOOL SUPERVISORY COMMITTEE APPROVAL of a thesis submitted by Jasmine Simone Polimadei This thesis has been read by each member of the following supervisory committee and by majority vote found to be satisfactory. ______________________________ Dr. Tanya Nolan, EdD Chair, School of Radiologic Sciences ______________________________ Christopher Steelman, MS Director of MSRS Cardiac Specialist ______________________________ Dr. Laurie Coburn, EdD Director of MSRS RA ______________________________ Dr. Robert Walker, PhD Director of MSRS 3 THE WEBER STATE UNIVERSITY GRADUATE SCHOOL RESEARCH AGENDA STUDENT APPROVAL of a thesis submitted by Jasmine Simone Polimadei This thesis has been read by each member of the student research agenda committee and by majority vote found to be satisfactory. Date April 25, 2025 ______________________ ____________________________________ Jasmine Simone Polimadei 4 Abstract ABSTRACT The progression and integration of technological advancements in modern-day medical practices continue to propel the diversity of treatment options while improving patient outcomes. Technology implanted with artificial intelligence (AI) generates quality assurance measures with large data collection supporting providers and allied healthcare professionals to implement greater measures with stronger evidence-based medical practices. This review focuses on the advancements of AI technology and algorithms applied to electrocardiogram (ECG) interpretation to support clinical decision processes, which aim to improve patient outcomes. The application of Machine Learning (ML) techniques with Deep Learning (DL) subsets such as Recurrent Neural Networks (RNN), and Convolutional Neural Networks (CNN) increases the recognition and identification levels in ECG sensitivity, specificity, accuracy, and precision. The individual use or clinical application of the AI-ECG algorithm demonstrates great potential to enhance emergency hospital response time management (e.g., door-to-balloon), support precision medical interventions, such as PCI or pharmacological treatments, and promote the appropriate allocation of expensive medical resources during time-sensitive cardiovascular emergencies like ST Elevation Myocardial Infarction (STEMI). Additional support from AI-ECG clinical integrations offers substantial improvements in patient risk assessments and early diagnosis. Although the implementation of AI technology utilized in everyday practice is slow, and with some institutions are more reluctant than others, an established standardized method can provide greater 5 understanding in the technological application that emphasizes a standard of care supporting clinical reasoning and human judgment. This systematic review abides by the guidelines outlined by PRISMA, ensuring a detailed academic review of documents, articles, and scholarly peer-reviewed journals that are sound resources contributing to a greater understanding of the technological expansion of artificial intelligence (AI) with electrocardiogram (ECG) interpretation. The resources reviewed are intended to summarize current and future applications of AI-ECG ability to provide clarification to improve clinical reasoning and human judgment, optimizing time response to cardiac emergencies, resource allocation, and improving patient outcomes. 6 Acknowledgments I would like to express my sincere gratitude to program director Christopher Steelman MS, R.T.(R)(CI)(ARRT), for his unwavering support and guidance. His wealth of knowledge, understanding, and counsel has been instrumental in helping me articulate my thoughts, ideas, and inquisitive nature, allowing me to convey my genuine passion on paper. I would also like to extend my appreciation to Weber State University and its Department of Radiological Sciences for fostering an environment where students can share their passion, expand their knowledge, and develop expertise to better serve the community. A heartfelt thank you to Intermountain Medical Center in Murray, Utah, for providing me with the invaluable opportunity to develop my skills as a student and aspiring RCIS. The clinical experience I have gained here is immeasurable, reinforcing the importance of teamwork in delivering optimal care and ensuring excellence in patient service. Lastly, to my family and friends, your unwavering support and encouragement have given me the strength to embrace each experience with an open heart and a resilient spirit. Your belief in me has allowed me to grow, adapt, and fully appreciate every opportunity that has come my way. 7 Table of Contents Chapter 1: Introduction ....................................................................................................... 1 Background ..........................................................................................................................2 Statement of the Problem .....................................................................................................4 Purpose of the Study ............................................................................................................5 Research Questions ..............................................................................................................5 Nature of the Study ..............................................................................................................6 Significance of the Study .....................................................................................................6 Definition of Key Terms ......................................................................................................7 Summary ..............................................................................................................................9 Chapter 2: Literature Review ............................................................................................. 10 Documentation ................................................................................................................... 31 Theme/Subtopic [repeat as needed] ................................................................................... 30 Summary ............................................................................................................................ 32 Chapter 3: Research Method .............................................................................................. 35 Research Methods and Design(s)....................................................................................... 36 Population .......................................................................................................................... 37 Assumptions ....................................................................................................................... 38 Limitations ......................................................................................................................... 38 Delimitations ...................................................................................................................... 39 Summary ............................................................................................................................ 39 Chapter 4: Implications, Recommendations, and Conclusions ......................................... 41 Implications........................................................................................................................ 41 Recommendations .............................................................................................................. 43 Conclusions ........................................................................................................................ 44 References .......................................................................................................................... 45 Appendices ......................................................................................................................... 53 Appendix A: Key Study Insights ....................................................................................... 53 Appendix B: Inclusion and Exclusion Criteria Used in Literature Review ....................... 54 Appendix C: Abbreviations ............................................................................................... 55 8 List of Figures Figure 1. Joynt Maddox et al. “Forecasting the Burden of Cardiovascular Disease and Stroke in the United States Through 2050—Prevalence of Risk Factors and Diseases: A Presidential Advisory From the American Heart Association”………..…..………,…...10 Figure 2. Joynt Maddox et al. “Forecasting the Burden of Cardiovascular Disease and Stroke in the United States Through 2050—Prevalence of Risk Factors and Diseases: A Presidential Advisory From the American Heart Association”………..………….…….,.11 Figure 3. Kazi et al. “Population-level economic burden of key cardiovascular risk factors in US adults, 2020 to 2050”.………………………………………………………..12 Figure 4. Kazi et al. “Population-level economic burden of cardiovascular disease and stroke in US Adults”.............…………………………………………………...…………..12 Figure 5. Roa et al.“2025 ACC/AHA/ACEP/NAEMSP/SCAI Guideline for the Management of Patients With Acute Coronary Syndromes: A Report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines.”..………………………………………………………………………..……..……..13 Figure 6. Roa et al. “2025 ACC/AHA/ACEP/NAEMSP/SCAI Guideline for the Management of Patients With Acute Coronary Syndromes: A Report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines.”……………………………………………………………………………………….16 Figure 7. Yeh et al. “Artificial Intelligence-Enhanced Electrocardiography Improves the Detection of Coronary Artery Disease.”…..……………………………….………………17 Figure 8. Yeh et al. “Artificial Intelligence-Enhanced Electrocardiography Improves the Detection of Coronary Artery Disease.”….………………………..……………………….…22 9 Figure 9. Chang et al.“Usefulness of multi-labelling artificial intelligence in detecting rhythm disorders and acute ST-elevation myocardial infarction on 12-lead electrocardiogram”...................................................................................................……...26 Figure 10. Chang et al.“Usefulness of multi-labelling artificial intelligence in detecting rhythm disorders and acute ST-elevation myocardial infarction on 12-lead electrocardiogram”......……………………………………………………………………….…..28 Figure 11. Chang et al.“Usefulness of multi-labelling artificial intelligence in detecting rhythm disorders and acute ST-elevation myocardial infarction on 12-lead electrocardiogram”………………………...………………………………………………….28 1 Chapter 1: Introduction The electrocardiogram (ECG) and its application as a medical tool remain essential, providing key information for physicians and other allied health professionals to recognize the electrical signals suggestive of coronary artery disease. With coronary artery disease having a profound impact on global mortality rates, early recognition, diagnosis, and interventions must be facilitated promptly. Current ECG technology has a relatively moderate accuracy in identifying cardiac emergencies, specifically ST Elevation Myocardial Infarction (STEMI). With recent advances in artificial intelligence (AI), the potential to revolutionize the utility and accuracy of the ECG is promising. AI technology has and continues to be recognized for its direct impact on improving prognosis and diagnosis, which is creating new opportunities to improve modern medical practices. The application of machine learning (ML), deep learning (DL) algorithms, and central neural networks (CNN) led the way for precision medicine to take shape and improve all aspects of medicine. Artificial intelligence continues to contribute to the improvement of radiological imaging and diagnostics in various fields, including magnetic resonance imaging (MRI), computed tomography (CT), Cardiac MRI (CMRI), and X-ray, to name a few. The application of AI technology in clinical practice suggests the opportunity to place greater emphasis on personalized medicine, especially with ECG analysis. With increasing data and research, the application and utility of AI algorithms demonstrate promise to improve the accuracy and detection of cardiac emergencies, ensuring timely STEMI activation and providing physicians of varied degrees in experience and specialties (residents, interns, experienced physicians, and cardiologists) additional supportive guidance, reducing the rate of human error. As AI-ECG algorithms 2 are intended to provide improved insight and an enhanced ability to identify specific cardiac emergencies, with AI-ECG a patient's need for reperfusion therapy could be determined. The greatest factor in cardiac emergencies is time; therefore, additional support improving door-to-balloon (D2B) times will directly contribute to an improvement in patient outcomes. With a multitude of technological platforms, AI algorithms have the potential to facilitate improved ECG readings and screening, allowing for greater medical accessibility with the potential to enhance a pathway to precision medical practices. Background The ECG has been the standard technology to diagnose coronary artery disease (CAD), as well as identify changes in electrical conduction via waveform morphology. The identification of specific changes allows providers and supporting allied health professionals to initiate recommended protocols or standards for optimal outcomes concerning the cardiovascular system. Ensuring optimal patient outcomes is of great significance when the accuracy of ECG interpretation is the determining factor for STEMI activation. The identification of a STEMI requires accurate and time-sensitive verification from a provider, ensuring that prompt activation of catheterization lab personnel facilitates reperfusion therapies within the recommended 90-minute time frame. Coronary Vascular Disease (CVD) remains a leading global cause of mortality, responsible for approximately 17.9 million deaths (32%), with ischemic heart disease (IHD) accounting for 9.14 million (16.7%) of these fatalities10,30 Approximately 805,000 3 acute myocardial infarctions (AMI) occur annually, of this number, 605,000 are first-time occurrences, with the remaining 200,000 being recurrent AMI events48. The increasing rate of specific risk factors impacting American adults, including obesity (42%) and hypertension (47%), impedes the progress toward reducing the rate of these timesensitive cardiac emergencies that are the leading cause of global mortality rates28. It has been recommended that all communities should sustain STEMI care systems to ensure that continuous assessment and quality improvement efforts within hospital-based activities and EMS establish performance-based initiatives improving door-to-balloon (D2B) times62. Suspected STEMI responses outside of the hospital begin with EMS. The initial objective data is gathered from a 12-lead ECG and then read by EMS personnel. In the event of an in-hospital setting, initial interpretation is overseen by a provider with varying experience that often requires additional review from the expertise of a cardiologist. In both settings, ECG readings are communicated to a cardiologist for additional interpretation and confirmation, delaying time-sensitive medical decisions. Due to ambiguity and uncertainty with current ECG interpretation, artificial intelligence algorithms demonstrate strong potential to enhance clinician guidance with interpretation and verification to facilitate improvements in appropriate activation for time-sensitive cardiovascular emergencies. The progression and integration of technological advancements in modern-day medical practices propels the vast diversity of treatment options available to improve patient outcomes. Technology programmed with AI algorithms demonstrates promise to generate quality assurance measures by utilizing large data collection to support providers and allied healthcare professionals with greater accuracy and stronger evidence-based 4 medical practices4. This review focuses on the progressing advancements of AI algorithms and how they are applied to ECG interpretation to support clinical decision processes, improving patient outcomes. The application of Machine Learning (ML), Deep Learning (DL), and Convolutional Neural Networks (CNN) increases levels of sensitivity, specificity, accuracy, and precision. The implantation of these four components into current ECG algorithms has the potential to improve EMS clinical assessments and emergency hospital response time management (door-to-balloon, triage), while simultaneously promoting individualized patient care interventions (i.e. utilizing PCI or medication) to facilitate the appropriate allocation of resources during timesensitive cardiovascular emergencies, such as STEMI. While new medical innovations contribute to driving the rapid progression of clinical medical practices, the integration of AI technology into clinical settings remains limited. Greater understanding is needed to promote the development of standardized approaches that enhance and support the clinical reasoning and judgment of current and future practitioners, as well as allied healthcare professionals. Statement of the Problem Advances in cardiovascular medicine were revolutionized by the innovation and application of the ECG over 100 years ago. With the technological application of computer-based algorithms, the ECG expands the healthcare provider's ability to understand, analyze, and expand evidence-based medical practices concerning anatomical, physiological, and pathological cardiovascular changes. Current methods, programs, and algorithms are limited in their recognition of lethal arrhythmias, transient events, and precipitating morphological changes of STEMI and non-ST-elevation 5 myocardial infarction (NSTEMI). With increasing unforeseen diagnostic limitations, greater opportunities exist for human error and a delay in care. Integrating clinical medical practice with advancing technologies, such as AI algorithms, suggests a firm promise to reduce incidences of human error while supporting practitioners and allied medical staff with increased supportive guidance. Integrating AI technology with the 12lead ECG promises to drive improvements in STEMI activation, reducing D2B time, reducing medical expenses, and optimizing patient outcomes. Purpose of the Study This is a systematic review that aims to illustrate how advancements in AI algorithms and technology can enhance STEMI activation management, reducing mortality rates through improved response times in cardiovascular emergencies. Additionally, this study aims to highlight the potential of AI’s role in minimizing the economic burden that is associated with inappropriate catheterization lab activations, improving cardiovascular screening and emergency triage, and supporting continuous education for physicians and allied health professionals. Research Questions Q1. Can artificial intelligence algorithms generate ECG readings comparable to those of a cardiologist, ensuring accurate STEMI activation? Q2. Can implementing AI algorithms minimize human error and enhance ECG interpretation accuracy? Q3. Can AI algorithms improve training and educational comprehension of telemetry readings for allied health professionals? 6 Nature of the Study This comprehensive literature review evaluates the current advancements, applications, and challenges concerning integrating artificial intelligence (AI) into electrocardiography (ECG) analysis for clinical practice. Rather than collecting primary data, this review should synthesize historical information with current peer-reviewed research, clinical studies, and technological reports that provide a concise understanding of how the application of AI algorithms, particularly those involving machine learning models and subsets, is leveraged to enhance ECG interpretation, diagnosis, and clinical decision processes. This study will focus on the application across various cardiovascular conditions, including common arrhythmias, myocardial infarctions (MI, STEMI, NSTEMI), and heart failure. Attention will also highlight diagnostic accuracy, the benefit of early detection, workflow efficiency, and ethical considerations. This review aims to identify current limitations and the potential for future research applications for potential clinical implementation. Significance of the Study This study demonstrates the importance of providing up-to-date medical practices, supporting technological advances, and ensuring optimal patient outcomes. With complications of cardiovascular disease leading the cause of the global rates of mortality, healthcare professionals require supportive tools for rapid, accurate, and non-invasive diagnostic measures to optimize initial assessments. Currently, false positive rates for STEMI unnecessarily consume resources that should be readily available for emergent 7 cardiovascular conditions. The integration of AI-ECG demonstrates promise to provide supportive measures with informed data reflecting greater accuracy, sensitivity, and precision, facilitating the most appropriate treatment decisions for patient-centered medicine. Definition of Key Terms Artificial Intelligence (AI): Artificial intelligence refers to the development of computer systems that can perform tasks typically requiring human intelligence, such as pattern recognition, learning, and decision-making.41, 62 Machine Learning (ML): A subset of AI where systems learn patterns from data and improve their performance without explicit programming. In ECG analysis, ML techniques enhance the detection of subtle features linked to cardiac abnormalities.11,41 Deep Learning (DL): A specialized form of ML that employs multi-layered neural networks to extract complex features from data. DL models have shown strong performance in ECG-based diagnosis. 11,30,41 Convolutional Neural Network (CNN): A type of DL model designed for analyzing spatial or time-series data, such as ECG waveforms. CNNs effectively capture morphological features important for classifying arrhythmias. 30,41,62 Recurrent Neural Network (RNN): RNNs are deep learning models ideal for analyzing sequential data, such as ECG signals, where time dependency and signal continuity are critical. 11,41, 8 Electrocardiogram (ECG): A diagnostic test that records the heart’s electrical activity over time. ECGs are widely used in diagnosing cardiac conditions such as arrhythmias and myocardial infarction.7,27 AI-Enhanced ECG (AI-ECG): The use of AI models, such as ML, DL, CNNs, and RNNs to analyze ECG data and support clinicians in making faster and more accurate diagnoses.27 Arrhythmia: A condition characterized by abnormal heart rhythms, which may be benign or life-threatening. AI-ECG aims to improve early detection and classification of arrhythmias.11,41,62 STEMI/NSTEMI/AMI/OMI: Types of myocardial infarction identified through ECG patterns: ● STEMI: ST-elevation myocardial infarction, indicating a complete coronary artery blockage. ● NSTEMI: Non-ST-elevation myocardial infarction, representing partial blockage. ● AMI: Acute myocardial infarction, encompassing both STEMI and NSTEMI. ● OMI: Occlusion myocardial infarction, focusing on coronary artery occlusion regardless of ECG ST elevation.6,11,15 Precision Medicine: A healthcare model that customizes treatment to individual patient profiles. AI-ECG contributes to precision medicine by offering tailored diagnostic insights based on unique electrical signatures.43,44,47 9 Summary This systematic academic review intends to provide greater insight into future applications of artificial intelligence algorithms with electrocardiogram diagnostic measures, improving the decision process during cardiovascular emergencies, like STEMI activation. It is the intention of this review to provide greater depth and understanding of the technology available to physicians and allied healthcare providers that demonstrates strong supporting evidence to facilitate greater diagnostic accuracy, minimize human error, improve positive cardiac catheterization laboratory (CCL) activation, and reduce the impact of medical waste contributing to the global growing economic burden. Over the course of eight months, a total of 62 peer-reviewed articles have been reviewed using methods outlined by PRISMA. Information, data, and materials collected and cited within this review were accessed through a variety of academic resources and are relevant within the last 10 years. All materials prior to 2015 provide historical support in nature and are acknowledged within the references section for their continuing contributions to academia. 10 Chapter 2: Literature Review Coronary Vascular Disease (CVD) remains a leading global cause of mortality, responsible for approximately 17.9 million deaths (32%), with ischemic heart disease (IHD) accounting for 9.14 million (16.7%) of these fatalities10, 30. In the United States, coronary artery disease (CAD) is a significant cause of death, contributing to approximately 610,000 fatalities annually (1 in 4) and generating $200 billion in medical costs for a largely treatable and preventable condition10. A complication of CVD is myocardial infarction (MI), which occurs due to a significant reduction or complete blockage of blood flow in the coronary arteries, resulting in impaired oxygen delivery to the myocardial tissue. Approximately 805,000 acute myocardial infarctions (AMI) occur annually, of this number, 605,000 are first-time occurrences, with the remaining 200,000 being recurrent AMI events48. Although the statistical value in AMI events is reducing, the increasing rate of specific risk factors impacting American adults, including obesity (42%) and hypertension (47%), is impeding the progress to reduce the rates of the leading cause in global mortality28. The estimated economic burden of MI alone is estimated to be 84.9 billion dollars in annual expenses and loss of productivity31, 48. Populations impacted by AMI more frequently are disproportionately non-Hispanic Black males and women who often experience a delay in care due to atypical signs and symptoms48. 11 Figure 1: Joynt Maddox et al. (2024). Number of US adults with adverse levels of cardiovascular health factors, health behaviors, and cardiovascular disease and stroke, 2020 to 2050 12 Figure 2: Joynt Maddox et al. (2024). “Number of US adults with adverse levels of cardiovascular health factors, health behaviors, and cardiovascular disease and stroke, 2020 to 2050”. 13 Figure 3: Kazi et al (2024). Population-level economic burden of key cardiovascular risk factors in US Adults, 2020 to 2050. Figure 4: Kazi et al. (2024). Population-level economic burden of cardiovascular disease and stroke in US Adults, 2020 to 2050. Management of the signs and symptoms of MI frequently includes reperfusion therapy, which involves the assessment of coronary vessels via a minimally invasive percutaneous catheterization procedure. Interventional cardiologists utilize angiographic imaging, producing 2-D x-ray images and films to visualize coronary perfusion and potential lesion burden using TIMI flow. However, prior to emergency angiography, the patient undergoes multiple evaluations and assessments by emergency medical services (EMS), 14 emergency department (ED) nursing staff, and physicians to determine if the 12-lead electrocardiogram (ECG) accurately indicates STEMI. Figure 5: Rao et al. (2025): “Initial Assessment of Patients With Suspected ACS”. Since its inception, the ECG has served as a vital tool in medical diagnostics. The initial design and development of the ECG began with the string galvanometer by the Dutch physician and physiologist Willem Einthoven7,51,54. The foundation of the string galvanometer generated the first images of the rhythmic cardiac cycle, mapping the synchronous electrical impulses and pathways within the heart 51,54. In 1901, Willem Einthoven revolutionized the world of cardiovascular medicine with the electrocardiogram (ECG), providing visual electrical tracings that describe “extra systole, 15 complete heart block, auricle and ventricular hypertrophy, atrial fibrillation, and flutter, the U wave, the effects of heart rate and respiration and examples of various heart diseases”7, 54. ECG data illustrate how physiological changes affect electrical impulse conduction and ion movement, enabling physicians and clinical staff to identify cardiac dysrhythmias, myocardial infarction, electrolyte imbalances, cardiac chamber enlargement, and other secondary physiological or pathophysiological conditions impacting cardiac function7,54. As a crucial diagnostic tool, the standard ECG algorithm's ability to rapidly and accurately interpret data is under scrutiny, as its accuracy directly impacts response times, the appropriate allocation of medical resources, diagnosis, and treatment planning that impact patient outcomes. Building on more than 100 years of medical diagnostic—leveraging machine learning (ML), deep learning (DL), recurrent neuronal networks (RNN) and convolutional neuronal networks (CNN)—provides a transformative opportunity to enhance the ECG’s ability to collect a more comprehensive data for clinical support, while minimizing human error during rapid, and timely decisions reliant on accuracy. ML, DL, and CNN provide greater depth and recognition of a multitude of individualized patient factors, thus outperforming the previous conventional scoring systems for generalized risk assessment, determinants for major adverse cardiac events (MACE), early diagnosis from non-invasinve biomarker for cardiovascular disease via subtle electrical patterns that are not apparent to the human eye19,39,46,50,62. The introduction of automated algorithms for the ECG began in the 1960s. This technological advancement solidified the ECG's performance capabilities, establishing its role as a standard tool in cardiac care diagnostics19. The evolving progression and 16 application of automated algorithms continue to broaden clinical interpretation through supportive technological equipment. Clinical assessments with programmed AI-ECG demonstrate a strong potential to reduce false-positive activation and misdiagnosis, with continued educational support measures providing real-time interpretation feedback, thus building on initiatives to reduce resource misallocation6,19,62. Recommendations supported by the American Heart Association and the European Society of Cardiology (ESC) for STEMI diagnosis state that obtaining a 12-lead ECG in less than 10 minutes or at the first point of medical contact is essential26, 30. From that point forth, all clinical decisions are impacted by diagnostic data and clinical interpretations collected concerning electrical pathway conduction from bipolar leads I, II, III, augmented leads aVL, aVF, and precordial leads V1, V2, V3, V4, V5, and V6. Signals transmitted typically consist of a P wave, QRS complex, and T wave, which are reflective of anatomical, physiological, and mechanical influences on the heart. Institutions, physicians, EMS, and paramedics have classically defined STEMI diagnoses through the observable changes within the ST segment prior to repolarization activity indicated by the T wave. Commonly characterized as either ST elevation or depression, the limited sensitivity and specificity with standard ECG algorithms for STEMI characteristics and detection have demonstrated overall interpretive rates of 65% and 76.9%, as described by Kavak et al. Additional studies concerning field activation have reported false-positive rates of up to 34%62, false-positive rates of 10-36% in emergency departments6, and a general STEMI detection range of 0.62-0.93 for sensitivity and 0.89-0.99 for specificity11. 17 Figure 6: Rao et al. (2025): “Types and Classification of Acute Coronary Syndromes”. NSTEMI indicates non-ST-segment elevation myocardial infarction, and STEMI, ST-segment elevation myocardial infarction. Given that time to reperfusion during an acute STEMI event is a critical factor in determining patient outcomes62, the interpretive abilities of AI-ECG to assist in rhythm interpretation can facilitate greater incidences of appropriate positive activation, reduce door-to-ballon times (D2B) and improve access for accurate AMI diagnosis in rural or impoverished locations where cardiologist are scarce18,46. The ease and application of the 12-lead ECG are ideal for all healthcare settings, with information generated immediately, cost-effectively, and non-invasive. Software developers, biomedical engineers, medical informatics specialists, and many others are expanding the potential of various machine learning subfields, such as deep learning 18 (DL), convolutional neural networks (CNN), and long short-term memory (LSTM), to name a few. AI with cognitive reasoning capabilities equal to human intelligence is designed to analyze and identify potential solutions to complex problems using algorithms that may or may not involve explicit programming, which is designed to continuously improve with increased data exposure41. Figure 7: Yeh et al. 2025 (a) ECGs included within retrospective study under specified criteria. (b)Training cohort composed of four groups. The rapid and multifaceted advancements of AI programming, particularly in its ability to replicate aspects of human reasoning, provide unique, significant, and nuanced enhancements across several domains and subsets within machine learning (ML)41. Many aspects of everyday living are simplified or enhanced by the program capabilities of ML, 19 from the electric car auto-pilot features to speech recognition and increasing application in precision medicine practices53. ML allows computer systems to expand their knowledge base on previous experiences from imported data, improving and processing observations with “Support Vector Machines (SVM), decision trees, Bayes learning, kmeans clusters, association rule learning, regression, and neural networks”41. The subset of ML, DL has allowed algorithms to overcome many of the limitations of earlier shallow networks53, processing large datasets, which enable a computer to work off programmable assumptions6,11,30,39,47,50,62. DL, as well as convolutional neuronal networks (CNN) and recurrent neuronal networks (RNN), apply multilayer units based on the model of the human brain, improving accuracy and operational speed, while reducing the time needed for critical training45,53. Algorithms for DL can be grouped based on specific problem-solving tasks. These groupings include supervised for prediction parameters, unsupervised for categorizing clusters, density estimation, dimensionality reduction, general adversarial networks or problems, and reinforcement learning for design algorithms for situational-based data41. The subfields CNN, DNN, and RNN are again inspired by the structure of neural networks within the brain, mimicking a layered network (input, hidden/middle, and output) that consists of connected units called artificial neurons41. In the 2D-CNN study by Kavak et al., the 12-lead ECG model is trained using a binary classification to improve STEMI detection and localization. Data gathered for algorithmic programming from 540 ECG images, including 270 STEMI images and 270 other ECG images not STEMI30. The 2D-CNN model achieved 96.3% accuracy, 96.2% sensitivity, 89.4% precision, 0.926 F1-score, and 0.962 ROC-AUC scores for 537 ECG 20 images30. This data demonstrates a strong correlation in supporting a model that assists physicians and allied healthcare professionals in making fast and accurate decisions within a limited time. Two comparative studies by Kavak et al. provide definitive insight that assesses the accuracy of physician interpretation of ECG data. The first survey included 99 physician interpretations and demonstrated an overall sensitivity and specificity to identify STEMI of 76.9% and 65%30. The second survey with 124 physicians to interpret 4392 ECGs reported findings for overall sensitivity and specificity in detecting STEMI at 65% and 79%30. This data suggests that the current interpretive abilities likely lead to increased rates of misdiagnosis, as well as a likely culprit for falsepositive or false-negative STEMI activation for timely treatment. A direct correlation with delays in treatment and appropriate activation of emergency resources is likely increasing mortality rates. For example, data collection reflective of ST-elevation myocardial infarction (STEMI) contributes to “in-hospital mortality of approximately 5-6% in the United States…[with] 42.6% of all-cause mortality in rural and urban areas” (62) of CVD in China. STEMI mortality rates are generating greater attention in questioning how to optimize ECG interpretation to improve door-to-balloon times, which has a defined start time in recognizing “symptom onset to hospital arrival” 62 . In a study in China, time-lapse for treatment received was notably delayed up to 4 hours, while data within some areas of the United States suggested a delay in interventional time of up to 2 hours. A delay in treatment has a cumulative effect in D2B concerning patient outcomes; for example, “15 min to 180 min…increase in in-hospital mortality of 3% to 8.5%...[for a span of] 6-month mortality [with total percentage increases] from 10% to 20%” 62. AI-ECG improvements in D2B can be facilitated 21 through early detection of STEMI with AI-enhanced ECG monitors to increase margins in sensitivity and specificity that are not always observable to the human eye39, 62. Implementation of deep convolutional neural networks is achieving satisfactory results due to continuous interactions with a high capacity of “raw ECG data” 62. Model data from Zhao et al. demonstrated that the area under the curve (AUC) for AI 12-lead ECG is 0.9954 (95% Confidence interval (CI), 0.9885 to 1), sensitivity 96.75%, specificity 99.20%, accuracy 99.01%, precision 90.86%, and F1 score 0.9372. Comparative data collected from this study evaluating the standard ECG algorithm and clinician interpretation (non-cardiologist) provided the following data: standard ECG algorithms values for sensitivity 32.0 %, specificity 90.0%, accuracy 61.0%, precision 76.19%, and F1 score 0.4702. Respectively, data collection from the 15 physicians who participated in this study demonstrated the following values: 71.73% sensitivity, 89.33% specificity, 80.53% accuracy, 87.05% precision, and an F1 score of 0.7865. With a high CI and significant contribution to performance improvements in data analysis, AI-equipped ECG technology provides more substantial support measures during time-sensitive events and continues to demonstrate a valid purpose in future medical practice. With the variety of multiple subsets in AI to process high volumes of complex data, various professionals can gain fast, reliable insight, promoting informed decision processes. With the potential of known improvements in accuracy and rapid analysis, AIECG has the ability to positively impact the fieldwork process for EMS during suspected STEMI and other cardiac emergencies. With an average AI-ECG interpretation performing at the same degree as a cardiologist or better, EMS personnel could effectively communicate accurate ECG readings in addition to an incoming patient's 22 status and condition. With the increased confidence in the diagnostic capabilities of AIECG, the clinical support role of EMS can excel, as they will be able to better aid inhospital emergency staff with improved triage measures. As well, a reduction in the rate of out-of-hospital false-positive STEMI activations by EMS will contribute to a substantial decrease in medical waste from unnecessary cardiac catheterization lab (CCL) activation costs and reducing avoidable fatigue and stress placed on CCL clinical staff60. In one US study over a span of 10 months, 23 activations were field-related, while another 33 STEMI activations were from the ED61. Performance from this study demonstrated that with additional support from a physician, 9% of ED activations were false, compared to 39% of field activations. This indicates that a 30% higher activation with conventional ECG technology does not provide efficient data or factors to appropriately define the incidence of STEMI. Additional studies discussing the strength in data from AI-ECG acknowledge comparable abilities to those of highly trained cardiologists or superior11,30,39,50,60,62, expanding its role in supporting fieldwork during out-of-hospital events. Due to significant knowledge gaps, and without the guided expertise of a cardiologist, EMS relies solely on standard ECG readings as the primary point-of-care tool for real-time identification of acute coronary occlusion6, likely contributing to a higher rate than necessary activation of the cardiovascular lab (CVL). In another US study, 10-36% of false positive rates occur within the emergency department (ED), while an additional 25% of field settings contribute to misutilization of resources and inappropriate cardiac catheterization lab (CCL) activation6. With the rising number of emergency field calls for suspected STEMI, the growing responses to non-ST elevation myocardial infarction (NSTEMI) also 23 contribute to false-positive and false-negative rates, impacting emergency response times. It is estimated that 500,000 NSTEMI cases occur annually in the United States, and approximately 125,000 NSTEMI cases meet the criteria of STEMI requiring emergent reperfusion therapy6. With AI-ECG having the ability to differentiate the details of OMI, AMI, STEMI/NSTEMI, and many other cardiovascular dysrhythmias6,11, improving diagnostic accuracy in the field will minimize delays before reperfusion therapy, as well as reduce false positive activations. Figure 8: Yeh et al. (2025) (a) Flow chart of ML (b)AI-algorithm enhances performance of the ECGs compared with conventional ECGs in the training cohort, (c) Prediction rates of AI-ECG in subjects with normal sinus rhythm (d) ROC curves calculated for CAD detection. (e) adjusted odds ratios of features associated with the frequency of CAD in subjects with normal ECG tracings. 24 In recent years, AI has demonstrated remarkable success in the interpretation of various medical images of multiple specialties that include dermatologic conditions, pathological slides, ophthalmic images, the classification of abnormalities on plain radiographs, computed tomographic (CT) scans, magnetic resonance imaging (MRI) scans, and ECG interpretation providing diagnosis with more informed treatment planning and decision4,9,49. With growing approval for the application of AI algorithms with computer technology, government regulating institutions like the Food and Drug Administration (FDA) in the United States have approved the use and application of 200 radiology products49. Due to the significant impact of the COVID-19 pandemic, commercial AI algorithm expansion is successfully taking place in more than 20 countries, with 2% of this influence on the U.S. market49. These applications directly impact the workflow and triage efficiency across diverse clinical settings, including hospitals, small or large practices, and point-of-care centers with enabled AI-ECG. With AI-ECG serving as a clinical “co-pilot”49, clinicians' precision-guided medical practices are empowered to make faster decisions with greater accuracy in patient care. Previously stated in the study led by Youngquist et al. brings attention that an upwards of ~34% of misutilization of CCL resources directly results from false-positive activation. Current STEMI criteria outlined by the American College of Cardiology Foundation, American Heart Association, and Heart Rhythm Society recommend that measurements to determine ST elevation (STE) be determined from the J-point57 on a telemetry reading. This morphologic detail is noted to assist cardiologists in differentiating influences that potentially contribute to changes in ST elevation and/or influences resulting from electrical conduction pathways (Left bundle branch blocks), stress/cardiac exertion, or 25 chronic cardiomyopathies, such as Takotsubo cardiomyopathy (TC)57. This single parameter defines the importance of accurate telemetry interpretation, as well as underscores the increased time required to appropriately analyze STE changes, which often dictate a course of either emergent or non-emergent medical response. Considering that STEMI is one of the most severe types of myocardial infarction, the accuracy of data collection for the ECG is paramount. The implementation of ML, DL, DNN, CNN, or a combination39,47 of algorithm capabilities expands the traditional ECG model with the diagnostic abilities of AI, providing a higher degree of interpretation of morphological nuances. As in the case of the 2D-CNN model, cardiac rhythm analysis interpretation extends beyond waveforms but also gauges analysis on the basis of programmable images. The additional utility in image analysis allows for 2D-CNN artificial intelligence, a platform for a high degree of identifying the localization of STEMI signals30. The performance of the 2D-CNN was compared alongside additional systems and was reportedly able to outperform its 10 counterparts. This system demonstrates potential benefits by allowing users to observe its analytical methodology during ECG interpretation, addressing the 'black box' concern. The 2D-CNN model identifies and highlights specific regions within the data corresponding to STEMI signals, providing a higher degree of transparency in its decision-making process. With greater insight on the algorithm’s deductive reasoning, the system allows clinicians to develop greater confidence in the results, rather than relying on them without critical evaluation30. With the CCL leading interventional cardiovascular practices with reperfusion therapy to coronary vessels, additional procedures that facilitate life-saving/changing interventions are likely to benefit from the clinical application of AI-ECG. Extending beyond the 26 reduction of false-positive activation rates of STEMI, AI-ECG demonstrates strong potential with individualized pre/post procedural risk assessment (mortality risk/rate), as well as the identification of lethal arrhythmias that could occur during mitral valve replacements, transaortic valve replacement (TAVR), ablations, coronary total occlusion (CTO) and many more. Mahmoud et al. illustrate the effect AI-ECG facilitates when performing mortality assessment on patients anticipating transaortic valve replacement (TAVR). Current modalities in assessing the various preoperative components and risk factors require tedious and time-consuming calculations from STS and TAV12 scores to provide interventionists with the necessary insight into procedural risk that could follow within the first year39. The strength and accuracy of AI-ECG calculations are able to be quantified, generating faster assessment with the potential to provide stratification for a 5-year long-term prognosis39. Similar to the methods to objectively assess pre-/post-procedural risk, the study by Sau et al. was able to apply AI-ECG algorithms with a DL model that generates statistical data for prediction risk assessment of CVD mortality rates and potential future occurrence of the disease. In the AI-ECG Risk Estimator (AIRE) study, a programmed risk estimator utilizes a data set of 189,539 ECGs from 1,163,401 patients, allocating greater focus and detail to individual patient factors. The data collected by the AIRE model assembles greater “explainability, or biological plausibility”52 with data validation generated from diverse international cohorts from the USA, Brazil, and the United Kingdom52,35. Data collected offers clinicians valuable insights that support informed navigation of high-risk 27 associated with specific procedures, with a common goal in enhancing patient survival trajectories52. The growing expansion of the conventional 12-lead ECG with AI includes developing algorithms that apply long-short-term memory (LSTM), a type of recurrent neural network (RNN) that excels at processing sequential data. With a specific ability to process data within a cell and three gates (input gate, output gate, and forget gate), neural networks provide memory for a greater period of time, as well as greater control of influencing data feeds from previous findings. In a study led by Chang et al., algorithm specifications developed and trained on 60,537 clinical ECGs were derived from 35,981 patient records collected between 15 January 2009 and 31 December 2018. Figure 9: Chang et al. (2021), “Algorithm processing and developing a bidirectional, four-layer LSTM model. This approach produced enhanced morphological specification and diagnostic accuracy via a multi-labeling system to identify more than one programmed rhythm. Comparative testing using this model weighted the performance of interpretive data outcomes from board-certified physicians against the outcomes of the LSTM ECGs. To ensure external 28 validation of the data acquired, the AI-ECG’s model is then trained to recognize 13 specific cardiac rhythms, similar to conventional models within real-world clinical settings that contain recognition features of both electrical and structural information. Specific rhythms programmed for identification include normal sinus rhythm, atrial fibrillation (AFIB), atrial flutter (AFL), atrial premature beat, ventricular bigeminy (BIGEMINY), ectopic atrial rhythm (EAR), paroxysmal supraventricular tachycardia (PSVT), sinus tachycardia (ST), and ventricular premature beat. Conduction defects included in dysrhythmia identification included the following: complete heart block (CHB), first-degree AV block (FRAV), and second-degree AV block (SAV). Within this study, the multi-labeling system was able to determine classification of the specific number 13 rhythms, the following results identified that 55,108 (91.032%) of 60,537 ECG tracings with one of thirteen learned rhythms, 4262 (7.040%) with two, 176 (0.291%) with three, and four types of the 13 ECG diagnosis were identified in 0.0007%11. Overall performance metrics for the LSTM model identification of the 13 rhythms (including STEMI) provided an accuracy of 0.942-0.998. Data outcomes in identifying acute STEMI were the following: accuracy 0.983, AUC 0.957, precision 0.818, recall 0.652, and F1 score 0.726. Specific identification for BIGEMINY, CHB, EAR, and SAV was>0.990. Subsequent external testing with this LSMT model demonstrates continued consistency in comparative testing alongside commercial algorithms and board-certified physicians (cardiologist, internist, emergency physicians) with AUC 0.942, 0.799, 0.780, and 0.650 results. 29 Figure 10: Chang et al. (2021) External testing ECG (a) correct classification of SAV and AMI from LSTM, four cardiologists, one of three ED physicians, and a commercial algorithm correctly identified. (b) LSTM correctly identifies BIGEMINY and FAV, with 8 out of 10 physicians only annotating BIGEMINY. Figure 11: Chang et al. (2021) Performance of LSTM and board-certified doctors in detecting STEMI and different rhythms. (a) STEMI (b) atrial fibrillation, (c) complete heart block, (d) paroxysmal supraventricular tachycardia. 30 Rather than being reliant on the gold standard 12-lead ECG, AI-ECG models are reducing lead requirements, while producing quality data that is not inferior in performance or accuracy to its conventional counterpart. With the increasing expansion of wearable digital health technologies, the application of the 12-lead ECG can extend into the homes of everyday people. The capabilities of remote monitoring of cardiovascular disease provide real-time updates of cardiac events, as well as the progression of the disease to remote care teams. With continuous observations, data collected offer greater insight into the signs and undetected symptoms39,55 frequently experienced by individuals who are likely to obtain a delayed diagnosis. The ECG data for most consumer devices like Smartwatches (Apple Watch, Samsung Galaxy, KardiaMobile, etc.) often provide a 30-second rhythm strip with lead I on either the right or left wrist. Single-lead devices are noted to demonstrate promising detection of atrial fibrillation with a sensitivity of 78-88% and specificity of 80-86%12. Influences such as artifact and uninterpretable artificial noise require attention and should be taken into account, as this impacted 2 to 15% of the results gathered55. The use of remote patient monitoring promotes effective implementation of supportive care measures while improving the management of cardiovascular risk. Through continuous updates concerning physiological data, better preventative measures for sudden cardiovascular exacerbation can be implemented to reduce unnecessary hospitalization55. Although no specific data provided by Spatz et al. directly addressed the capabilities of wearable ECG devices in determining STEMI, the FDA continues its approval of indefinite guidance to utilize patient-facing ECGs55 that could support improved 31 measures to reduce misallocation of medical resources, false-positive activation, and improve patient outcomes. With more than a century of clinical application, the ECG remains a cornerstone in cardiovascular diagnostics. The increasing integration of artificial intelligence is advancing the deeper analysis of both modifiable and non-modifiable factors influencing cardiac electrical conduction. The literature reviewed indicates that AI-ECG— particularly through machine learning (ML) subsets such as, deep learning (DL), convolutional neuronal networks (CNN), and recurrent neuronal networks (RNN)— demonstrates considerable promise improving supportive diagnostic measures with enhanced speed, accuracy, and consistency of arrhythmia detection and risk stratification. As current findings highlight the strong potential for clinical integration, ongoing validation and standardization that includes diverse populations within clinical settings will be essential to identify the full transformative impact AI-ECG can provide for future cardiovascular care. Documentation References and material gathered for this literature review included usage of search engines such as Google Scholar, Elicit, PubMed, Perplexity, and Weber State Stewart Library. Search strategies for online sources included the following key terms “AI+ECG, AI+STEMI, Deep Learning+ECG, Cath Lab+STEMI activation, AI+Machine learning+Convolutional Neuronal Networks, AI-ECG+STEMI activation, and AIECG+D2B” as well as specific searches with Elicit an AI research platform (AI vs 32 Human: Diagnosing STEMI Errors, AI EKG Interpretation Accuracy, and Standardized Techniques in AI-ECG Systems). Theme: Diagnostic Accuracy and Early Detection Using AI-Enhanced ECG Artificial intelligence (AI) enhanced with ML, DL, CNN, and RNN models has advanced the diagnostic capabilities of the ECG, ensuring early detection of various cardiovascular conditions. The theme of this literature review focuses on how AI-ECG systems improve diagnostic precision during acute coronary syndromes, lethal arrhythmias, and subtle morphological abnormalities. Several studies have demonstrated that CNNs and RNNs facilitate enhanced detection of patterns in ECG data that are often imperceptible to the human eye. For example, Muzammil et al (2024) bring attention to how the training of CNN ECG models identified and classified 13 rhythm types with a level of accuracy superior to boardcertified cardiologists. Similarly, Zhao et al. (2020) demonstrated that a DL model capable of predicting changes in ST elevation, improving the recognition to facilitate faster interventions during STEMI. AI models also demonstrate great promise in the identification of subtle biomarkers linked to myocardial infarction, left ventricular dysfunction, and hypertrophic cardiomyopathies. The ability of the diverse systems to process large datasets for continuous learning supports more timely interventions and risk-stratification measures. Change et al. (2021) demonstrated that multi-label deep learning approaches increase the specificity of ECG interpretation, reducing false-positive activation in STEMI diagnosis. 33 Although the accuracy of this model is highly dependent on the quality and diversity of training data. Studies have recognized that AI-ECG models could potentially underperform when applied to external datasets or across populations with varied clinical characteristics, underscoring the concerns of generalizability35. In summary, the literature continues to support the role of AI-ECG-based diagnostics as a supportive measure, improving early detection of critical cardiac events. Greater implementation of real-world clinical settings is required, furthering validation, standardization, and clinical training. Summary Unreliant on the predefined criteria of traditional ECG algorithms, it is strongly suggested that AI-ECG offers faster and more accurate interpretations of cardiac events while leveraging machine learning to analyze large quantities of data, including subtle patterns and details that commonly go unnoticed by conventional methods39,62. Beyond the boundaries of merely providing rapid data interpretation, AI-ECG promotes opportunities to improve triage and decision-making processes that empower clinicians and emergency responders to effectively prioritize patient care needs. The adaptive abilities of AI-ECG demonstrate promise to continuously improve as data is continually gathered, refining diagnostic skills30,41,55. With ongoing data collection, AI-ECG will have capabilities to mitigate false positive STEMI interpretation, reducing unnecessary cardiac cath lab activation to ensure that resource utilization is reserved for emergencies requiring immediate intervention. Comparative data results against board-certified physicians and conventional ECG methodology demonstrate that, with statistically sound results, AI-ECG can perform consistently with high rates in accuracy, sensitivity, and 34 precision30,47,62. This technology aims to provide greater opportunity to mitigate ECG interpretation delays to improve clinical diagnostic measures to meet the needs of a growing global population with CVD and related emergencies. 35 Chapter 3: Research Method The electrocardiogram (ECG) remains an indispensable diagnostic tool for the foundation of clinical practice in all modalities of medicine. Its ability to provide critical insight with essential data enables physicians and allied health professionals to identify electrical abnormalities suggestive of coronary artery disease (CAD). With an increasing global burden of CAD impacting mortality rates, the promise of AI-ECG supports early detection and improved accuracy. As conventional models in ECG technology provide only moderate accuracy in detecting cardiac emergencies, AI-ECG with great potential can lead the pathway toward accessible precision-based care, transforming cardiovascular medicine practices. This literature review underscores the importance that contemporary medical practices recognize technological innovations with the potential to elevate patient outcomes. With CVD remaining a leading global cause of mortality, healthcare professionals require equipment that is efficient, accurate, non-invasive, and supportive of clinical judgement for early targeted intervention. High false-positive rates and delays for timely interventions for ST-Elevation Myocardial Infarction (STEMI) contribute to the misallocation of emergent resources and poor patient outcomes. The integration of AI technology with ECG demonstrates potential to improve diagnostic support with enhanced accuracy, sensitivity, and precision while optimizing patient outcomes with individualized precision-based medicine. With a detailed review utilizing the guidelines outlined by PRISMA, a total of 61 peerreviewed journals are the basis of this research. Each resource provided greater insight 36 into the specific question concerning the impact of AI-ECG, and how its performance compares to that of a board-certified physician and current ECG algorithm interpretation. Q1. Can artificial intelligence algorithms generate ECG readings comparable to those of a cardiologist, ensuring accurate STEMI activation? Q2. Can implementing AI algorithms minimize human error and enhance ECG interpretation accuracy? Q3. Can AI algorithms improve training and educational comprehension of telemetry readings for allied health professionals? Research Methods and Design(s) The research method and design used in this literature review are guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Selected for its ability to ensure structure, transparency, and a methodologically rigorous process, this literature review includes a comprehensive selection and synthesis of relevant studies. The search strategy developed for this review included a combination of specific keywords tailored to the research questions' specifications, various academic databases, including PubMed and GoogleScholar, scholarly online journal resources (New England Journal of Medicine, Heart Rhythm Society, JAMA Cardiology), and educational institutions (American College of Cardiology, European Society of Cardiology). Search terms were controlled with vocabulary including “AI+ECG,” “AI+STEMI,” “STEMI+false-positive,” “Machine learning+Deep Learning+Convolutional Neuronal Networks+ECG,” “ECG+AI-ECG.” 37 To ensure the relevance and quality of studies selected for review, materials required the following: peer-reviewed articles published in English, studies focused on the application of ECG or AI-enhanced ECG or AI in cardiovascular diagnostics with human subjects. Criteria for exclusion included non-peer-reviewed sources, studies not directly addressing the research focus, and data subject to potential biases. The included studies were screened after a review of the subject title and abstract. Eligibility was then determined by a full-text review, with data extraction and quality conducted systematically for consistency while maintaining reliability across the included studies. Population The population within each study consisted of large datasets from electronic health records (EHR), typically gathered within emergency department settings. Data collected was obtained from a diverse group within each region individual studies were performed. The majority of studies included male subjects with ages often >60. Of the many studies reviewed, each consisted of either several hundred or hundreds of thousands of ECG data. Although studies that combine EHR from human subjects with computer-generated data provided valuable insight and supported objective findings, this information was excluded from this review. The objective in each study was to determine the accuracy, sensitivity, and precision that AI-ECG could perform equal to or better than conventional ECG systems, as well as board-certified physicians. Assessments were based on identification of patterns, trends, and identifiable correlations concerning cardiac electrical tracings associated with cardiac arrhythmias, myocardial infarction (STEMI/NSTEMI), heart failure, and other 38 cardiovascular diseases. The studies selected for this literature review examined the capabilities and limitations of various AI methodologies—including machine learning (ML), deep learning (DL), convolutional neural networks (CNN, deep neural networks (DNN), and recurrent neural networks (RNN)—in analyzing complex waveforms. With either single subset application or combination, these approaches have demonstrated potential in enhancing clinical understanding of subtle morphological features and relevant biomarkers within electrical tracing that improve interventions. Assumptions This literature review assumes that the development and application of AI-ECG are based on algorithms developed on high-quality, accurately labeled datasets representative of the surveyed patient population. It is also believed that the AI models discussed have been trained and validated with the appropriate clinical standards and performance metrics. Additionally, it is assumed that the integration of AI-ECG interpretation is intended to support, rather than replace, clinical decision processes established by highly trained and certified healthcare providers. Limitations This literature review is subject to a variety of limitations. Due to the nature in the rapid evolution of AI technology, some recent developments may not be reflected in this published literature, subject to limiting the currency of this review. In addition, variability in study design, such as data quality, variable standardization methodology, population diversity, and clinical setting, poses challenges for comparative outcomes. Studies were 39 conducted in both controlled and clinical environments via EHR, which may only replicate a fraction of real-world clinical performance. These limitations bring attention to the need for ongoing research with clinical validation of AI-ECG integration. Delimitations This literature review is delimited to exploring the potential clinical advancements associated with the implementation of AI-ECG. This primary focus addresses how machine learning (ML), deep learning (DL), convolutional neuronal networks (CNN), and recurrent neuronal networks (RNN) improve in the detection, identification, and classification of lethal cardiac arrhythmias to improve patient outcomes. To ensure AI-ECG relevance to clinical practice, this review concentrates on studies that included adult human subjects, assessing the diagnostic accuracy of AI-ECG in identifying critical cardiac events, including STEMI, NSTEMI, AMI, and OMI. Priority is focused on peer-reviewed publications to provide a credible and evidence-based foundation for clinical analysis. The delimitations of this study ensured a focus on the practical synthesis of sound and clinically relevant applications of AI-ECG for clinical support interpretation. Summary This literature review is subject to several limitations, often regarding the variability in data gathering for generalizability within the demographics of locations within each study. Interpretations based on the data collected from the population majority restrict diversity when attempting to establish findings that are varied within population 40 demographics. Additional limitations within this review include the inability to transparently understand the methods in data interpretation for AI-ECG models operating off “black box” systems. The absence of standardized frameworks to evaluate AI algorithms generates complications for cross-study comparisons, leading to potential inconsistencies in performance reporting. With notable gaps in the literature, the limited number of prospective, real-world validations impacts the external validation reported findings. Publication bias is another concern that exists due to the exclusion of nonEnglish literature, gray literature, and unpublished or preprint studies; the potential for an overrepresentation of favorable results is possible. 41 Chapter 4: Implications, Recommendations, and Conclusions The integration of AI into ECG interpretation suggests significant advancement in cardiovascular diagnostics, intending to provide enhanced clinical decision processes, reduce diagnostic errors, and improve patient outcomes. The findings of this review indicate that AI-ECG with ML and various subsets can provide accurate and rapid identification of critical cardiac events. Yet, the transition from research to clinical implementation remains challenging and complex. This section will explore the broader implications of the findings and conclude with key insights focused on the transformative potential and various considerations to promote clinical integration of AI-ECG. Implications The integration of AI-ECG into clinical workflows presents transformative implications, improving the diagnosis and management of CVD. Literature reviewed suggests that AI models, particularly those that employ ML, DL, CNN, and RNN, have demonstrated an ability to identify complex waveform morphologies that are often undetected by the human eye. Studies such as Zhao et al. bring attention to how CNN-based models are able to identify the subtle morphological patterns associated with MI, as a result of high specificity and sensitivity, which significantly improves early diagnosis in acute settings. Kavek et al. support these findings, demonstrating that AI-ECG models not only increase the accuracy of arrhythmia interpretation but also enhance clinician confidence, reducing the time required during emergency department decisions. This demonstrates how technological tools enhanced with AI are able to serve as effective support systems when high-stakes decisions are needed. 42 In the studies by Muzammil et al., emphasis on the role of AI in risk stratification is illustrated in the application of DL and CNN models trained on a large dataset. Training models for this study demonstrated AI-ECGs' effective predictive capabilities concerning MACE, compared to conventional scoring methods. This work emphasizes the value and opportunity available for AI in preventative care, which enables earlier interventions, promoting precision medical treatment strategies based on patient-specific risk factors. These findings collectively support that applying AI-ECG in clinical practice promises to improve diagnostic precision, while reshaping clinical pathways, from triage to emergency response to long-term risk management. Successful implementation of AIECG methods will require robust validation across diverse populations, as well as an application to integrate with existing health IT infrastructure and continuous clinician training. The ability to address each of these considerations will be essential in an effort to implement AI-ECG for measurable improvement of patient outcomes and healthcare system efficiency. Recommendations Key expert contributions, including Faramand et al., Baker et al., and Lekadir et al, propose the following to support the responsible development and clinical integration of AI-ECG technologies: 43 1. Ensure Clinical Relevance and Contextual Adaptability AI models should be designed to function within specific clinical workflows, ensuring practical utility and usability with real-time clinical support measures.19 2. Promote Human-Centered and Trustworthy AI The development and application of AI-ECG should be guided by transparency, explainability, and clinician oversight, reinforcing the supportive tool measure role within clinical judgment 35 3. Advanced Multicenter and Diverse Data Validation The application of prospective multicenter studies should be of great focus, ensuring generalizability within the dataset, reducing potential algorithmic bias6. 4. Adopt FAIR and Reproducible AI Practices Adherence to FAIR data principles (Findable, Accessible, Interoperable, and Reusable) promotes responsible data stewardship that ensures reproducibility in the development of AI-ECG system integration. 5. Enhance Interoperability and Integration AI-ECG models should fit within the parameters of clinician workflows, regarding their application to EHR and monitoring systems.6 6. Support Continuous Monitoring and Ethical Oversight Ongoing performance monitoring and regulation of the AI-ECG framework should address safety, equity, and accountability as outlined in FUTURE-AI.19 7. Invest in Clinician Training and Multidisciplinary Collaboration Clinician engagement and cross-disciplinary collaboration will ensure AI literacy for the 44 responsible adoption of AI-ECG systems. Collaborative training programs should be implemented, promoting successful integration.19, 35 Conclusions The collective evidence reviewed emphasized the transformative potential of AI-ECG technologies in improving cardiovascular diagnostics, risk stratification, and clinical decision-making. As highlighted through the implications and supported by strategic recommendations grounded in current literature, including Faramand et al, Baker et al., and the FUTURE-AI framework that represents the advancement in precision medicine. The successful clinical integration requires coordinated efforts to ensure fairness, transparency, interoperability, and validation across diverse patient populations. Forward progressions are dependent on the collaborative efforts and engagement in the best regulatory practices to harness the full potential of AI-ECG for precision cardiovascular medical practice. 45 References 1. AI-Enabled ECGs Reduce Time to Cath Lab in Patients with Suspected STEMI. https://www.healio.com/news/cardiology/20231121/aienabled-ecgs-reduce-time-tocath-lab-in-patients-with-suspected-stemi. Accessed 5 Mar. 2025. 2. AlGhatrif, Majd, and Joseph Lindsay. “A Brief Review: History to Understand Fundamentals of Electrocardiography.” Journal of Community Hospital Internal Medicine Perspectives, vol. 2, no. 1, Jan. 2012, p. 14383. DOI.org (Crossref), https://doi.org/10.3402/jchimp.v2i1.14383. 3. Anderson, Kelley P. “Artificial Intelligence‐augmented ECG Assessment: The Promise and the Challenge.” Journal of Cardiovascular Electrophysiology, vol. 30, no. 5, May 2019, pp. 675–78. DOI.org (Crossref), https://doi.org/10.1111/jce.13891. 4. Assadi, Azadeh, et al. “An Integration Engineering Framework for Machine Learning in Healthcare.” Frontiers in Digital Health, vol. 4, Aug. 2022. Frontiers, https://doi.org/10.3389/fdgth.2022.932411. 5. Avram, Robert, and William F. Fearon. “AI-RISE to the Challenge — Artificial Intelligence Reduces Time to Treatment in STEMI.” NEJM AI, vol. 1, no. 7, June 2024. DOI.org (Crossref), https://doi.org/10.1056/AIe2400472. 6. Baker, Peter O., et al. “Artificial Intelligence Driven Prehospital ECG Interpretation for the Reduction of False Positive Emergent Cardiac Catheterization Lab Activations: A Retrospective Cohort Study.” Prehospital Emergency Care, Sept. 2024, pp. 1–9. DOI.org (Crossref), https://doi.org/10.1080/10903127.2024.2399218. 7. Barold, S. Serge. “Willem Einthoven and the Birth of Clinical Electrocardiography a Hundred Years Ago.” Cardiac Electrophysiology Review, vol. 7, no. 1, Jan. 2003, pp. 99–104. Springer Link, https://doi.org/10.1023/A:1023667812925. 8. Benenati, Stefano, et al. “Invasive ‘in the Cath-Lab’ Assessment of Myocardial Ischemia in Patients with Coronary Artery Disease: When Does the Gold Standard Not Apply?” Cardiovascular Revascularization Medicine, vol. 19, no. 3, Part B, Apr. 2018, pp. 362–72. ScienceDirect, https://doi.org/10.1016/j.carrev.2018.01.005. 46 9. Borris, Mercer. AI in the Cath Lab: Implications of Clinical AI-Enabled Assistance for Intravascular Ultrasound Procedures. 2023. Massachusetts Institute of Technology, Thesis. dspace.mit.edu, https://dspace.mit.edu/handle/1721.1/151469. 10. Brown, Jonathan C., et al. “Risk Factors for Coronary Artery Disease.” StatPearls, StatPearls Publishing, 2025. PubMed, http://www.ncbi.nlm.nih.gov/books/NBK554410/. 11. Chang, Kuan-Cheng, et al. “Usefulness of Multi-Labelling Artificial Intelligence in Detecting Rhythm Disorders and Acute ST-Elevation Myocardial Infarction on 12-Lead Electrocardiogram.” European Heart Journal - Digital Health, vol. 2, no. 2, June 2021, pp. 299–310. DOI.org (Crossref), https://doi.org/10.1093/ehjdh/ztab029. 12. Choi, Jina, et al. “Smartwatch ECG and Artificial Intelligence in Detecting Acute Coronary Syndrome Compared to Traditional 12-Lead ECG.” IJC Heart & Vasculature, vol. 56, Feb. 2025, p. 101573. DOI.org (Crossref), https://doi.org/10.1016/j.ijcha.2024.101573. 13. Chung, Cheuk To, et al. “Clinical Significance, Challenges and Limitations in Using Artificial Intelligence for Electrocardiography-Based Diagnosis.” International Journal of Arrhythmia, vol. 23, no. 1, Oct. 2022, p. 24. Springer Link, https://doi.org/10.1186/s42444-022-00075-x. 14. Cox, David A., et al. “Comparative Early and Late Outcomes After Primary Percutaneous Coronary Intervention in ST-Segment Elevation and Non–ST-Segment Elevation Acute Myocardial Infarction (from the CADILLAC Trial).” The American Journal of Cardiology, vol. 98, no. 3, Aug. 2006, pp. 331–37. DOI.org (Crossref), https://doi.org/10.1016/j.amjcard.2006.01.102. 15. Demolder, Anthony, et al. “Prediction of Certainty in Artificial Intelligence-Enabled Electrocardiography.” Journal of Electrocardiology, vol. 83, Mar. 2024, pp. 71–79. ScienceDirect, https://doi.org/10.1016/j.jelectrocard.2024.01.008. 16. Department of Cardiology, Athens Navy and Veterans Hospital, 70 Dinokratous, 11521, Athens, Greece, and Grigorios Avdikos. “Expertise in ECG Interpretation and Artificial Intelligence ECG Models for Occlusion Myocardial Infarction Diagnosis.” Japan Journal of Research, vol. 5, no. 10, Oct. 2024. DOI.org (Crossref), https://doi.org/10.33425/2690-8077.1146. 47 17. Dhingra, Lovedeep S., et al. “Heart Failure Risk Stratification Using Artificial Intelligence Applied to Electrocardiogram Images: A Multinational Study.” European Heart Journal, Jan. 2025, p. ehae914. DOI.org (Crossref), https://doi.org/10.1093/eurheartj/ehae914. 18. Fang, Ying, et al. “Machine Learning-Based Myocardial Infarction Bibliometric Analysis.” Frontiers in Medicine, vol. 12, Feb. 2025. Frontiers, https://doi.org/10.3389/fmed.2025.1477351. 19. Faramand, Ziad, et al. “Performance and Limitations of Automated ECG Interpretation Statements in Patients with Suspected Acute Coronary Syndrome.” Journal of Electrocardiology, vol. 69, Nov. 2021, pp. 45–50. DOI.org (Crossref), https://doi.org/10.1016/j.jelectrocard.2021.08.014. 20. G, Ranganathan. “A Study to Find Facts Behind Preprocessing on Deep Learning Algorithms.” Journal of Innovative Image Processing, vol. 3, no. 1, Apr. 2021, pp. 66– 74. DOI.org (Crossref), https://doi.org/10.36548/jiip.2021.1.006. 21. Gliner, Vadim, et al. “Clinically Meaningful Interpretability of an AI Model for ECG Classification.” Npj Digital Medicine, vol. 8, no. 1, Feb. 2025, p. 109. DOI.org (Crossref), https://doi.org/10.1038/s41746-025-01467-8. 22. Hamayun, Shah, et al. “Innovations in Interventional Cardiology: Pioneering Techniques for a New Era.” Current Problems in Cardiology, vol. 49, no. 12, Dec. 2024, p. 102836. ScienceDirect, https://doi.org/10.1016/j.cpcardiol.2024.102836. 23. Haq, Kazi T., et al. “Applying Artificial Intelligence to ECG Analysis: Promise of a Better Future.” Circulation: Arrhythmia and Electrophysiology, vol. 13, no. 8, Aug. 2020, p. e009111. DOI.org (Crossref), https://doi.org/10.1161/CIRCEP.120.009111. 24. “Heart Disease Remains Leading Cause of Death as Key Health Risk Factors Continue to Rise.” ScienceDaily, https://www.sciencedaily.com/releases/2025/01/250127124438.htm. Accessed 4 Apr. 2025. 25. Hilbel, Thomas, and Norbert Frey. “Review of Current ECG Consumer Electronics (Pros and Cons).” Journal of Electrocardiology, vol. 77, Mar. 2023, pp. 23–28. ScienceDirect, https://doi.org/10.1016/j.jelectrocard.2022.11.010. 48 26. Ibanez, Borja, et al. “2017 ESC Guidelines for the Management of Acute Myocardial Infarction in Patients Presenting with ST-Segment Elevation.” European Heart Journal, vol. 39, no. 2, Jan. 2018, pp. 119–77. DOI.org (Crossref), https://doi.org/10.1093/eurheartj/ehx393. 27. Jin, Benjamin E., et al. “A Simple Device to Illustrate the Einthoven Triangle.” Advances in Physiology Education, vol. 36, no. 4, Dec. 2012, pp. 319–24. DOI.org (Crossref), https://doi.org/10.1152/advan.00029.2012. 28. Joynt Maddox, Karen E., et al. “Forecasting the Burden of Cardiovascular Disease and Stroke in the United States Through 2050—Prevalence of Risk Factors and Disease: A Presidential Advisory From the American Heart Association.” Circulation, vol. 150, no. 4, July 2024. DOI.org (Crossref), https://doi.org/10.1161/CIR.0000000000001256. 29. Kanaji, Yoshihisa, et al. “Predictive Value of Artificial Intelligence‐Enabled Electrocardiography in Patients With Takotsubo Cardiomyopathy.” Journal of the American Heart Association, vol. 13, no. 5, Mar. 2024, p. e031859. DOI.org (Crossref), https://doi.org/10.1161/JAHA.123.031859. 30. Kavak, Serkan, et al. “Application of CNN for Detection and Localization of STEMI Using 12-Lead ECG Images.” IEEE Access, vol. 10, 2022, pp. 38923–30. DOI.org (Crossref), https://doi.org/10.1109/ACCESS.2022.3165966. 31. Kazi, Dhruv S., et al. “Forecasting the Economic Burden of Cardiovascular Disease and Stroke in the United States Through 2050: A Presidential Advisory From the American Heart Association.” Circulation, vol. 150, no. 4, July 2024. DOI.org (Crossref), https://doi.org/10.1161/CIR.0000000000001258. 32. Khan, Salman, et al. “Efficacy of AI Models in Detecting Heart Failure Using ECG Data: A Systematic Review and Meta-Analysis.” Cureus, Feb. 2025. DOI.org (Crossref), https://doi.org/10.7759/cureus.78683. 33. Khera, Rohan, et al. “Transforming Cardiovascular Care With Artificial Intelligence: From Discovery to Practice.” Journal of the American College of Cardiology, vol. 84, no. 1, July 2024, pp. 97–114. DOI.org (Crossref), https://doi.org/10.1016/j.jacc.2024.05.003. 34. Knoery, C. R., et al. “SPICED-ACS: Study of the Potential Impact of a ComputerGenerated ECG Diagnostic Algorithmic Certainty Index in STEMI Diagnosis: Towards 49 Transparent AI.” Journal of Electrocardiology, vol. 57, Nov. 2019, pp. S86–91. ScienceDirect, https://doi.org/10.1016/j.jelectrocard.2019.08.006. 35. Lekadir, Karim, et al. “FUTURE-AI: International Consensus Guideline for Trustworthy and Deployable Artificial Intelligence in Healthcare.” BMJ, Feb. 2025, p. e081554. DOI.org (Crossref), https://doi.org/10.1136/bmj-2024-081554. 36. Lin, Chin, et al. “Artificial Intelligence–Powered Rapid Identification of STElevation Myocardial Infarction via Electrocardiogram (ARISE) — A Pragmatic Randomized Controlled Trial.” NEJM AI, vol. 1, no. 7, June 2024. DOI.org (Crossref), https://doi.org/10.1056/AIoa2400190 . 37. Martínez-Sellés, Manuel, and Manuel Marina-Breysse. “Current and Future Use of Artificial Intelligence in Electrocardiography.” Journal of Cardiovascular Development and Disease, vol. 10, no. 4, Apr. 2023, p. 175. www.mdpi.com, https://doi.org/10.3390/jcdd10040175. 38. Murphy, Joseph G., et al. “From Test Taking to Lifesaving: Will AI Be Your Next Cardiologist?” Mayo Clinic Proceedings, vol. 100, no. 1, Jan. 2025, pp. 14–16. DOI.org (Crossref), https://doi.org/10.1016/j.mayocp.2024.11.013. 39. Muzammil, Muhammad Ali, et al. “Artificial Intelligence-Echanced Electrocardiography for Accurate Diagnosis and Management of Cardiovascular Diseases.” Journal of Electrocardiology, vol.83, Mar. 2024, pp. 30-40. ScienceDirect, http://doi.org/10.1016/j.jelectrocard.2024.01.006. 40. Narang, Akhil, et al. “Utility of a Deep-Learning Algorithm to Guide Novices to Acquire Echocardiograms for Limited Diagnostic Use.” JAMA Cardiology, vol. 6, no. 6, June 2021, p. 624. DOI.org (Crossref), https://doi.org/10.1001/jamacardio.2021.0185. 41. Nguyen, Giang, et al. “Machine Learning and Deep Learning Frameworks and Libraries for Large-Scale Data Mining: A Survey.” Artificial Intelligence Review, vol. 52, no. 1, June 2019, pp. 77–124. Springer Link, https://doi.org/10.1007/s10462-01809679-z. 42. Olawade, David B., et al. “Advancements and Applications of Artificial Intelligence in Cardiology: Current Trends and Future Prospects.” Journal of Medicine, Surgery, and Public Health, vol. 3, Aug. 2024, p. 100109. ScienceDirect, https://doi.org/10.1016/j.glmedi.2024.100109. 50 43. Ose, Benjamin, et al. “Artificial Intelligence Interpretation of the Electrocardiogram: A State-of-the-Art Review.” Current Cardiology Reports, vol. 26, no. 6, June 2024, pp. 561–80. Springer Link, https://doi.org/10.1007/s11886-024-02062-1. 44. Paixão, Gabriela Miana, et al. “ECG-AGE FROM ARTIFICIAL INTELLIGENCE: A NEW PREDICTOR FOR MORTALITY? THE CODE (CLINICAL OUTCOMES IN DIGITAL ELECTROCARDIOGRAPHY) STUDY.” Journal of the American College of Cardiology, vol. 75, no. 11, Mar. 2020, p. 3672. DOI.org (Crossref), https://doi.org/10.1016/S0735-1097(20)34299-6. 45. Parvat, Aniruddha, et al. “A Survey of Deep-Learning Frameworks.” 2017 International Conference on Inventive Systems and Control (ICISC), 2017, pp. 1–7. IEEE Xplore, https://doi.org/10.1109/ICISC.2017.8068684. 46. Patel, Syed J., et al. “Advancements in Artificial Intelligence for Precision Diagnosis and Treatment of Myocardial Infarction: A Comprehensive Review of Clinical Trials and Randomized Controlled Trials.” Cureus, vol. 16, no. 5, p. e60119. PubMed Central, https://doi.org/10.7759/cureus.60119. Accessed 4 Apr. 2025. 47. Qayyum, Sardar N., et al. “Revolutionizing Electrocardiography: The Role of Artificial Intelligence in Modern Cardiac Diagnostics.” Annals of Medicine & Surgery, vol. 87, no. 1, Jan. 2025, pp. 161–70. DOI.org (Crossref), https://doi.org/10.1097/MS9.0000000000002778. 48. Rao, Sunil V., et al. “2025 ACC/AHA/ACEP/NAEMSP/SCAI Guideline for the Management of Patients With Acute Coronary Syndromes: A Report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines.” Circulation, vol. 151, no. 13, Apr. 2025. DOI.org (Crossref), https://doi.org/10.1161/CIR.0000000000001309. 49. Rokos, Ivan C. “Artificial Intelligence for STEMI Detection: The ‘Shanghai Algorithm’ Provides a Step Forward.” International Journal of Cardiology, vol. 317, Oct. 2020, pp. 231–32. ScienceDirect, https://doi.org/10.1016/j.ijcard.2020.07.002. 50. Salam, Amar M. “The Invention of Electrocardiography Machine.” Heart Views : The Official Journal of the Gulf Heart Association, vol. 20, no. 4, Nov. 2019, p. 181. pmc.ncbi.nlm.nih.gov, https://doi.org/10.4103/HEARTVIEWS.HEARTVIEWS_102_19. 51 51. Sandeep, Bhushan, et al. “Feasibility of Artificial Intelligence Its Current Status, Clinical Applications, and Future Direction in Cardiovascular Disease.” Current Problems in Cardiology, vol. 49, no. 2, Feb. 2024, p. 102349. ScienceDirect, https://doi.org/10.1016/j.cpcardiol.2023.102349. 52. Sau, Arunashis, et al. “Artificial Intelligence-Enabled Electrocardiogram for Mortality and Cardiovascular Risk Estimation: A Model Development and Validation Study.” The Lancet Digital Health, vol. 6, no. 11, Nov. 2024, pp. e791–802. DOI.org (Crossref), https://doi.org/10.1016/S2589-7500(24)00172-9. 53. Shrestha, Ajay, and Ausif Mahmood. “Review of Deep Learning Algorithms and Architectures.” IEEE Access, vol. 7, 2019, pp. 53040–65. IEEE Xplore, https://doi.org/10.1109/ACCESS.2019.2912200. 54. Silverman, M. E. “Willem Einthoven--the Father of Electrocardiography.” Clinical Cardiology, vol. 15, no. 10, Oct. 1992, pp. 785–87. PubMed, https://doi.org/10.1002/clc.4960151020. 55. Spatz, Erica S., et al. “Wearable Digital Health Technologies for Monitoring in Cardiovascular Medicine.” New England Journal of Medicine, edited by Stephen H. Friend et al., vol. 390, no. 4, Jan. 2024, pp. 346–56. DOI.org (Crossref), https://doi.org/10.1056/NEJMra2301903. 56. Squire, Benjamin T., et al. “Effect of Prehospital Cardiac Catheterization Lab Activation on Door-to-Balloon Time, Mortality, and False-Positive Activation.” Prehospital Emergency Care, vol. 18, no. 1, Jan. 2014, pp. 1–8. DOI.org (Crossref), https://doi.org/10.3109/10903127.2013.836263. 57. Tamura, Akira, et al. “A New Electrocardiographic Criterion to Differentiate Between Takotsubo Cardiomyopathy and Anterior Wall ST-Segment Elevation Acute Myocardial Infarction.” The American Journal of Cardiology, vol. 108, no. 5, Sept. 2011, pp. 630–33. DOI.org (Crossref), https://doi.org/10.1016/j.amjcard.2011.04.006. 58. Wardah, Wafaa, et al. “Protein Secondary Structure Prediction Using Neural Networks and Deep Learning: A Review.” Computational Biology and Chemistry, vol. 81, Aug. 2019, pp. 1–8. DOI.org (Crossref), https://doi.org/10.1016/j.compbiolchem.2019.107093. 52 59. Wehbe, Ramsey M., et al. “Deep Learning for Cardiovascular Imaging: A Review.” JAMA Cardiology, vol. 8, no. 11, Nov. 2023, pp. 1089–98. Silverchair, https://doi.org/10.1001/jamacardio.2023.3142. 60. Yeh, Chi-Hsiao, et al. “Artificial Intelligence-Enhanced Electrocardiography Improves the Detection of Coronary Artery Disease.” Computational and Structural Biotechnology Journal, vol. 27, 2025, pp. 278–86. DOI.org (Crossref), https://doi.org/10.1016/j.csbj.2024.12.032. 61. Youngquist, Scott T., et al. “A Comparison of Door‐to‐balloon Times and False‐ positive Activations between Emergency Department and Out‐of‐hospital Activation of the Coronary Catheterization Team.” Academic Emergency Medicine, vol. 15, no. 8, Aug. 2008, pp. 784–87. DOI.org (Crossref), https://doi.org/10.1111/j.15532712.2008.00186.x. 62. Zhao, Yifan, et al. “Early Detection of ST-Segment Elevated Myocardial Infarction by Artificial Intelligence with 12-Lead Electrocardiogram.” International Journal of Cardiology, vol. 317, Oct. 2020, pp. 223–30. DOI.org (Crossref), https://doi.org/10.1016/j.ijcard.2020.04.089. 53 Appendices Appendix A: Key Study Insights - Healthcare Settings: Academic hospital were the most common setting Other settings included specialized centers, community hospital, and public repositories. Diversity in setting is crucial for understanding real-world performance, but may limit direct comparisons between studies - Geographic region: Studies spanned multiple countries, with the United States and China being the most represented. Other locations included the UK, Japan and Brazil The geographic diversity is valuable for assessing algorithm generalizability, but may present additional confounding factors due to operational difference between hospital systems and practices. - Implementation Barriers: The lack of data on implemenatation challenges demonstrates a significant gap in the knowledge for the practical aspect in deploying AI-ECG algorithms in real-work clinical settings. These findings bring specific attention to the complexity and challenges in implementing AI-ECG algorithms across diverse healthcare setting and georgraphic regions. As some algorithms demonstrate high rates in consistency, other algorithms highlight variation within data collections that could impact their effectiveness in different clinical settings. Limited information on the implementation brings specific attention for additional research that addresses the practical challenges of integrating these algorithms into current healthcare models. 54 Appendix B: Inclusion and Exclusion Criteria Used in Literature Review Inclusion Criteria: - Peer-reviewed journal articles - Published between 2004 and 2024 (historical value) - Published between 2014 and 2024 (statistical value and current relevance) - Focused on AI application in ECG interpretation - Studies involving adult human subjects - Articles written in English Exclusion Criteria: - Animal or pediatric studies - Non-peer-reviewed or non-academic sources - Studies that used simulated or computer-generated ECG data without clinical validation. 55 Appendix C: Abbreviations AMI: Acute Myocardial Infarction AUC: Area Under the Curve CAD: Coronary Artery Disease CI: Confidence Interval CNN: Convoluntional Neuronal Netword CT: Computed Tomographic CVD: Coronary Vascular Disease D2B: Door-to-balloon DL: Deep Learning DNN: Deep Neuronal Network ECG: Electrocardiogram ED: Emergency Department EMS: Emergency Medical Service FDA: Food and Drug Administration IHD: Ischemic Heart Disease LSTM: Long Short Term Memory ML: Machine Learning MRI: Magnetic Resonace Imaging NSTEMI: Non-ST-Elevation Myocardial Infarction RNN: Recurrent Neuronal Network STEMI: ST-Elevation Myocardial Infarction TIMI: Thrombolysis In Myocardial Infarction