Artificial intelligence–based ECG as a triage tool for acute myocardial infarction: a diagnostic systematic review and meta-analysis

Artificial intelligence–enabled electrocardiography for detection of left ventricular diastolic dysfunction: a systematic review and meta-analysis

Artificial intelligence-enhanced electrocardiography demonstrated good diagnostic performance for detecting LVDD and may support future rule-out or risk-enrichment strategies in selected populations, however, current evidence remains insufficient to support routine clinical implementation.

Wearable and Artificial-Intelligence-Enabled Multimodal Systems for Early Detection and Classification of Acute Myocardial Infarction: A Review of the Integration of Clinical, Electrocardiographic, and Echocardiographic Data

Acute myocardial infarction (AMI) — comprising ST-elevation myocardial infarction (STEMI) and non-ST-elevation myocardial infarction (NSTEMI) — remains the leading cause of cardiovascular death worldwide, and rapid, accurate differentiation between its subtypes is essential for time-critical treatment decisions. Conventional diagnosis relies on a combination of clinical assessment, 12-lead electrocardiography (ECG), cardiac biomarkers, and echocardiography, each of which carries recognized diagnostic limitations when used in isolation. Over the past decade, wearable electrocardiographic devices, point-of-care high-sensitivity troponin assays, handheld artificial-intelligence (AI)-assisted echocardiography, and multimodal deep-learning fusion architectures have matured sufficiently to be considered building blocks of an integrated, low-cost diagnostic pathway suitable for emergency, pre-hospital, and resource-constrained settings. This review synthesizes evidence from cardiology, biomedical engineering, and digital-health literature on (i) the epidemiological burden and diagnostic paradigm of AMI, (ii) the diagnostic performance of wearable ECG and point-of-care biomarker technologies, (iii) artificial-intelligence approaches to ECG- and echocardiography-based MI detection, (iv) multimodal data-fusion strategies, and (v) the specific opportunities and gaps relevant to low- and middle-income countries such as Uganda. Across the reviewed literature, deep-learning models applied to single- or multi-lead ECG achieve areas under the receiver-operating-characteristic curve generally exceeding 0.90 for MI detection, wearable smartwatch electrocardiograms achieve sensitivities of 83–100% and specificities of 79–100% for arrhythmia and ST-segment change detection under supervised conditions, and point-of-care high-sensitivity troponin assays achieve diagnostic accuracy comparable to central-laboratory assays with substantially shorter turnaround times. However, multimodal fusion of clinical, ECG, and echocardiographic streams into a single wearable STEMI/NSTEMI classifier remains largely unrealized in the peer-reviewed literature, and evidence from sub-Saharan African populations is markedly under-represented. We propose a conceptual framework for an integrated wearable multimodal diagnostic system and outline the technical, clinical, regulatory, and ethical considerations that must be addressed before such systems can be safely deployed in emergency and pre-hospital cardiac care in low-resource settings.

Improved Detection of Acute Coronary Occlusion Myocardial Infarction by an Artificial Intelligence Electrocardiogram Model in Swedish Emergency Departments

Objectives The Queen of Hearts (QoH) ECG artificial intelligence model has demonstrated improved sensitivity for detecting occlusion myocardial infarction (OMI) compared with STEMI criteria, but further validation is needed. We aimed to evaluate QoH's diagnostic performance in patients with chest pain at Swedish emergency departments (EDs). Methods This retrospective analysis included consecutive patients with chest pain at the ED from the ESC-TROP study (2017-2018). Patients transferred directly from the prehospital setting to the coronary care unit were not included. OMI classification was based on angiographic data and expert adjudication. QoH, conventional STEMI criteria, and the Glasgow ECG Analysis Algorithm were applied to all cases. In addition, extended STEMI criteria incorporating additional ECG leads (-V1, -V2, -V3, -aVL, -aVR, and -III) and OMI criteria for left bundle branch block (modified Sgarbossa criteria) and left ventricular hypertrophy (ST elevation V1-V3 ≥0.25 of R and S) were applied. Results Among 24,511 patients (mean age 59 ± 19 years, 52% male), 467 (1.9%) had OMI. QoH achieved higher sensitivity than STEMI criteria (52% [47 to 57] vs 23% [19 to 27]), similar specificity (99% [99 to 99] vs 98% [98 to 98]), higher positive predictive value (51% [47 to 54] vs 17% [15 to 20]), and similar negative predictive value (99% [99 to 99] vs 98 [98 to 99]). The Glasgow algorithm obtained 32% (28 to 37) sensitivity, 98% (98 to 98) specificity, 26% (23 to 30) PPV, and 99% (99 to 99) NPV, and corresponding number for the extended criteria were 41% (36 to 45), 95% (95 to 96), 14% (12 to 15), and 99% (99 to 99). Conclusion In ED patients with chest pain, QoH improved sensitivity in OMI detection compared with currently available ECG criteria, with similar specificity.

ECG-based detection of occlusion myocardial infarction: a dedicated deep neural network versus multimodal large language models and physicians — a retrospective diagnostic accuracy study

Accurate electrocardiogram (ECG) interpretation for acute coronary occlusion is a critical, time-sensitive task in emergency care. The field is increasingly shifting from the ST-segment elevation myocardial infarction paradigm to the broader concept of occlusion myocardial infarction (OMI). This study compared the diagnostic accuracy of a dedicated occlusion-detection deep neural network, Queen of Hearts (QoH), two general-purpose multimodal large language models (LLMs), and emergency physicians for ECG-based OMI detection. Response consistency and confidence calibration of the LLMs were also evaluated. This retrospective diagnostic accuracy study used 36 twelve-lead ECGs from patients referred for emergent coronary angiography, including 24 with angiographically confirmed acute coronary occlusion and 12 without. Five emergency medicine specialists, five emergency medicine residents, and two LLMs, ChatGPT 5.2 and Gemini 3 Pro, interpreted all ECGs under a standardized moderate-risk acute coronary syndrome scenario. The same ECGs were submitted to QoH. Each LLM evaluated the set five times in separate sessions, whereas QoH was queried once because it provides deterministic output. In the primary head-to-head analysis, each interpreter or interpreter group contributed a single decision per ECG. Areas under the curve (AUCs) were compared using the DeLong test, binary metrics using the exact McNemar test, and reliability using Fleiss’ kappa. QoH showed the highest discrimination for OMI (AUC 0.96), followed by ChatGPT (0.81) and Gemini (0.62). QoH also achieved the highest sensitivity (95.8%, missing one of 24 occlusions) and accuracy (86.1%), with significantly higher sensitivity than emergency specialists (p = 0.016). Among physicians, specialists performed best (accuracy 75.0%; sensitivity 66.7%; specificity 91.7%). ChatGPT showed high specificity (91.7%) but low sensitivity (54.2%), whereas Gemini performed least well overall (accuracy 55.6%). The LLMs produced variable interpretations across repeated queries (Fleiss’ kappa 0.24–0.49), and Gemini showed marked overconfidence (Brier score 0.40). In this OMI-focused study, QoH showed the highest discrimination and sensitivity for acute coronary occlusion. General-purpose LLMs showed clinically important limitations and do not currently support autonomous OMI diagnosis. These findings support further evaluation of task-specific models for ECG-based occlusion detection and suggest, at most, an adjunctive, human-supervised role for current general-purpose LLMs.

AI-Assisted Electrocardiogram Interpretation Improves ST-Elevation Myocardial Infarction Diagnostic Accuracy Among Advanced Practice Providers: A Prospective Randomized Crossover Study.

A Conceptual Multimodal Wearable Platform for Acute Myocardial Infarction Risk Estimation Using Clinical, Electrocardiographic and Echocardiographic Data

Acute myocardial infarction (AMI) is a time-critical cardiovascular emergency in which early recognition of high-risk ischemic patterns can influence treatment decisions and access to timely reperfusion. Current assessment relies mainly on clinical evaluation, 12-lead electrocardiography (ECG), serial high-sensitivity cardiac troponin (hs-cTn) measurements, and, where indicated, cardiac imaging. While these approaches provide complementary diagnostic information, delays in obtaining and interpreting investigations, limited laboratory and specialist capacity, and difficulty recognizing subtle or initially non-diagnostic ischemic changes can delay appropriate care. These challenges are particularly relevant in pre-hospital and resource-constrained settings. This review examines clinical, technological, and engineering evidence relevant to the development of a conceptual multimodal wearable platform for early recognition and risk estimation of AMI, with emphasis on STEMI-pattern ischemia, suspected NSTEMI, and occlusion myocardial infarction (OMI). The review also develops a proposed system architecture, preliminary engineering requirements, a quality-aware approach to multimodal data fusion, and a staged pathway for development and validation. Evidence was synthesized from the supplied reference set, clinical guidelines, systematic reviews, studies of wearable ECG systems, AI-assisted ECG analysis, artificial-intelligence applications in echocardiography, and emerging transdermal biomarker technologies. Established clinical practices were considered separately from technologies that remain at an experimental or early-development stage. The reviewed evidence suggests that the proposed modalities could provide complementary information rather than replace existing diagnostic methods. Clinical variables can help establish pre-test probability, while ECG provides immediate information on cardiac electrical activity. Echocardiography can add mechanical evidence, including regional wall-motion abnormalities, whereas emerging optical and transdermal technologies may eventually provide biochemical information. Wearable ECG systems offer potential for continuous or on-demand monitoring, although differences in lead configuration, motion artifacts, signal-quality variation, and electrode placement limit their direct equivalence to standard 12-lead ECG. Portable echocardiography is more realistically suited to episodic, connected assessment than to integration into a conventional wrist-worn device. Similarly, although artificial intelligence may improve recognition of clinically relevant patterns, issues relating to generalizability, calibration, missing data, false alerts, uncertainty, and prospective clinical validation remain unresolved. Based on the current evidence, the most realistic near-term application is a modular multimodal decision-support platform that produces calibrated risk estimates and reports signal quality alongside the available physiological measurements. Such a system should complement, rather than replace, clinical assessment, standard 12-lead ECG, serial hs-cTn testing, and indicated cardiac imaging. The proposed framework therefore provides a practical foundation for prototype development, analytical performance testing, silent-mode prospective evaluation, external validation, and subsequent clinical outcome studies. It also avoids assuming a level of technological and clinical maturity that has not yet been demonstrated.