Assessing the Impact of Downsampled ECGs and Alternative Loss Functions in Multi-Label Classification of 12-Lead ECGs.

Safe automatic one-lead electrocardiogram analysis in screening for atrial fibrillation.

The differentiation of wide complex tachycardias (WCTs) into ventricular tachycardia (VT) and supraventricular wide tachycardia (SWCT) via 12-lead ECG (electrocardiogram) interpretation is a crucial yet demanding clinical task. Decades of research have been dedicated to simplifying and improving this differentiation via manual algorithms. Despite such research, the effectiveness of such algorithms still remains limited, primarily due to reliance on user expertise. To combat this limitation, automated algorithms have been created that show promise as alternatives to manual ECG interpretation. However, direct comparison of the methods’ diagnostic performances has not been undertaken. A recent publication (LoCoco et al., 2024) compared the diagnostic performance between traditional manual ECG interpretation approaches (i.e. Brugada, Vereckei aVR, and VT Score) to novel automated wide QRS complex tachycardia differentiation algorithms (i.e. WCT Formula I, WCT Formula II, VT Prediction Model, Solo Model, and Paired Model). Two electrophysiologists independently applied the 3 manual WCT differentiation approaches to 213 ECGs. Simultaneously, computerized data from the same paired WCT with baseline ECGs were processed by the 5 automated WCT differentiation algorithms. Following these analyses, the diagnostic performance of automated algorithms was compared with manual ECG interpretation approaches. In this article, a summary of data components relating to diagnostic performance of the methods tested is presented.

The 12-lead electrocardiogram (ECG) is a fast, non-invasive, powerful tool to diagnose or to evaluate the risk of various cardiac diseases. The vast majority of arrhythmias are diagnosed solely on 12-lead ECG. Initial detection of myocardial ischemia such as myocardial infarction (MI), acute coronary syndrome (ACS) and effort angina is also dependent upon 12-lead ECG. ECG reflects the electrophysiological state of the heart through body mass, and thus contains important information on the electricity-dependent function of the human heart. Indeed, 12-lead ECG data are complex. Therefore, the clinical interpretation of 12-lead ECG requires intense training, but still is prone to interobserver variability. Even with rich clinically relevant data, non-trained physicians cannot efficiently use this powerful tool. Furthermore, recent studies have shown that 12-lead ECG may contain information that is not recognized even by well-trained experts but which can be extracted by computer. Artificial intelligence (AI) based on neural networks (NN) has emerged as a strong tool to extract valuable information from ECG for clinical decision making. This article reviews the current status of the application of NN-based AI to the interpretation of 12-lead ECG and also discusses the current problems and future directions.

Electrocardiogram (ECG) Interpretation is one of the most important and critical skills for various medical specialists. It is essential for diagnosing and managing many cardiac diseases, including life-threatening conditions. To evaluate competency in ECG interpretation skills among medical students in the Makkah region and the associated factors. Medical students from all Makkah region medical schools were asked to complete a web-based survey containing questions on 17 ECG strips. There were questions on primary ECG parameters (rate, rhythm, and axis), emergencies, and common ECG abnormalities. Multiple logistic regression analysis was used to determine factors associated with their performance. A P value of < 0.05 was considered as significant. We enrolled 1239 medical students from medical schools in the Makkah region completed the questionnaire. The participants' ECG interpretation competency scores were generally low, with a median of 35.29% (IQR 17.65%-52.94%). Fifth and sixth-year students demonstrated significantly higher ECG competency scores compared students in the earlier years (P < 0.001). Logistic regression analysis showed that being in clinical years (OR: 1.65, 95% CI: 1.18-2.30, P = 0.003), self-study as primary source of ECG knowledge (OR: 0.58, 95% CI: 0.43-0.79, P = 0.001), and using both online and face-to-face learning methods (OR: 1.44, 95% CI: 1.00-2.07, P = 0.049) were significantly associated with higher odds of adequate ECG interpretation competency. The study showed a significant need for improvement in ECG interpretation skills among medical students in the Makkah region. The findings underscore the importance of integrating practical ECG interpretation training throughout the medical curriculum, with emphasis on clinical exposure, self-study, and blended learning approaches.

BackgroundAn electrocardiogram (ECG) is a standard tool used to detect various cardiovascular abnormalities. Detection sensitivity for atrial fibrillation (AF) was recently shown to be greatly increased by using short, intermittent ECG recordings. Modern mobile ECG recording devices that can monitor patients' heart activities around the clock have made this a reality. The Apple Watch is one of these portable ECG devices that can detect heart rhythms and is approved by the American FDA for screening and detecting AF.ObjectivesWe compared the results of the Apple Watch I lead ECG with conventional ECG results to assess the sensitivity and specificity of the Apple Watch I lead ECG. We then determined the abnormalities that can be detected by the Apple Watch I lead ECG.MethodsThis study was conducted on outpatient cardiac clinics at King Abdullah bin Abdulaziz University Hospital (KAAUH), and Prince Sultan Cardiac Center (PSCC), from May to October 2021. A standard 12-lead ECG was recorded and compared with the Apple Watch I lead ECG. A total of 469 ECG comparisons were included in this study and evaluated by two investigators. The data on patient demographics, medical and medication history, and ECG data were reviewed and analyzed using IBM SPSS Statistics for Windows, Version 23 (Released 2015; IBM Corp., Armonk, New York, United States).ResultsNo significant differences were seen between the Apple Watch and the 12-lead ECG in terms of the studied ECG characteristics. A significant and strong positive correlation between the heart rate measurements in the 12-lead ECG and Apple Watch ECG was documented. The most commonly found abnormalities in the Apple Watch ECG were AF in 37 (7.9%), followed by first-degree atrioventricular (AV) block in 32 (6.8%). The sensitivity of Apple Watch's automated interpretation to detect an AF was 99.54%, while the manual interpretation yielded a sensitivity of 100%.ConclusionThe results of this study demonstrated a robust relationship between the 12-lead ECG and Apple Watch ECG in the diagnosis of arrhythmias. Consequently, cardiac patients may consider the Apple Watch ECG a trustworthy remote monitoring technique.

Does ChatGPT-4 succeed in the ECG interpretation: friend or foe to cardiologists?

Clinical competence in electrocardiography. A statement for physicians from the ACP/ACC/AHA Task Force on Clinical Privileges in Cardiology.

SPICED-ACS: Study of the potential impact of a computer-generated ECG diagnostic algorithmic certainty index in STEMI diagnosis: Towards transparent AI

Background/Introduction The electrocardiogram (ECG) is an ubiquitously used non-invasive tool for diagnosis and risk prediction in cardiology, granting deep extensive insights into the heart. Artificial intelligence (AI) is a modern resource allowing the processing of vast complex datasets in a way that is comparable to humans. Risk stratification in cardiovascular patients is mainly based on scoring systems, such as the ESC-SCORE, relying on traditional risk variables like cholesterol levels or arterial hypertension, rather than actual cardiac structure and function. Goal of this project was to predict mortality using AI in patients with cardiovascular risk based on the current cardiac situation represented by a standard 12-lead ECG recording. Methods The study population is based on an ongoing registry that started in 2010 and enrolled patients scheduled for an invasive coronary angiography due to suspected chronic coronary syndrome. Data of the following study patients were analysed: enrolment within the first two study years with available long-term follow-up data on the outcome measure overall mortality, availability of an ECG at admission without pacemaker stimulation and availability of all variables needed to calculate the ESC-SCORE (in the version weighed for a German population) as comparison. This led to a cohort of 720 patients, of whom 70 died within the follow-up period. Information on presence of a relevant coronary artery disease (CAD) was available for all patients, to differentiate between primary and secondary prevention. A deep learning architecture that was previously developed to detect myocardial scar in raw ECG time-series data was used. This model was trained with 1400 ECG recordings, from the publicly available PTB-XL dataset with 700 of those ECGs labelled for acute, recent or old myocardial infarction while 700 were labelled as healthy. This pre-trained model was then applied to our study cohort to predict long-term mortality based on a single 12-lead ECG obtained at admission. Results For mortality prediction in patients without CAD (primary prevention) the AI model compares to the ESC-SCORE with an AUROC of 0.606 vs 0.584. For CAD patients (secondary prevention) the AI model compares with an AUROC of 0.612 vs 0.658. Detailed results are presented in Table 1. Conclusion(s) Our data underlines the potential of an AI based approach, predicting mortality in cardiovascular patients using only single 12-lead ECG recordings. Additionally, our model achieved similar predictive information to established risk classification systems, such as the ESC-SCORE. Since data acquisition is still ongoing, we will continue to improve our model. In future work training AI to specifically predict mortality while also exploring explainable AI could lead to breakthrough findings in ECG interpretation. Funding Acknowledgement Type of funding sources: Public Institution(s). Main funding source(s): FlexiFunds by Forschungscampus Mittelhessen

COMMUNITY-TO-INSTITUTION ATHLETIC CARDIOVASCULAR SCREENING: VALIDATION OF AN ELECTROCARDIOGRAM WORKFLOW MODEL

A comprehensive artificial intelligence–enabled electrocardiogram interpretation program

ObjectiveTo implement an all-day online artificial intelligence (AI)-assisted detection of ST-elevation myocardial infarction (STEMI) by prehospital 12-lead electrocardiograms (ECGs) to facilitate patient triage for timely reperfusion therapy.MethodsThe proposed AI model combines a convolutional neural network and long short-term memory (CNN-LSTM) to predict STEMI on prehospital 12-lead ECGs obtained from mini-12-lead ECG devices equipped in ambulance vehicles in Central Taiwan. Emergency medical technicians (EMTs) from the 14 AI-implemented fire stations performed the on-site 12-lead ECG examinations using the mini portable device. The 12-lead ECG signals were transmitted to the AI center of China Medical University Hospital to classify the recordings as “STEMI” or “Not STEMI”. In 11 non-AI fire stations, the ECG data were transmitted to a secure network and read by available on-line emergency physicians. The response time was defined as the time interval between the ECG transmission and ECG interpretation feedback.ResultsBetween July 17, 2021, and March 26, 2022, the AI model classified 362 prehospital 12-lead ECGs obtained from 275 consecutive patients who had called the 119 dispatch centers of fire stations in Central Taiwan for symptoms of chest pain or shortness of breath. The AI's response time to the EMTs in ambulance vehicles was 37.2 ± 11.3 s, which was shorter than the online physicians' response time from 11 other fire stations with no AI implementation (113.2 ± 369.4 s, P < 0.001) after analyzing another set of 335 prehospital 12-lead ECGs. The evaluation metrics including accuracy, precision, specificity, recall, area under the receiver operating characteristic curve, and F1 score to assess the overall AI performance in the remote detection of STEMI were 0.992, 0.889, 0.994, 0.941, 0.997, and 0.914, respectively. During the study period, the AI model promptly identified 10 STEMI patients who underwent primary percutaneous coronary intervention (PPCI) with a median contact-to-door time of 18.5 (IQR: 16–20.8) minutes.ConclusionImplementation of an all-day real-time AI-assisted remote detection of STEMI on prehospital 12-lead ECGs in the field is feasible with a high diagnostic accuracy rate. This approach may help minimize preventable delays in contact-to-treatment times for STEMI patients who require PPCI.

Where do derived precordial leads fail?

Background Previous research has demonstrated acceptable diagnostic accuracy of artificial intelligence (AI)-enabled sinus rhythm (SR) electrocardiogram (ECG) interpretation for identifying paroxysmal atrial fibrillation (AF). However, interethnic validations of the AI algorithms have not been widely conducted. Purpose We aimed to develop our own AI model for the identification of paroxysmal AF based on SR ECGs in the Korean population and to validate its diagnostic performance in Brazilian citizens. Methods We trained a Transformer-based vision network on 90% of a dataset comprising 808,194 ECGs from 121,282 patients at Seoul National University Bundang Hospital (2003-2020). The remaining 10% of the dataset was used for internal validation. The model was trained to compute a risk score for paroxysmal AF or new-onset AF within 2 years. External validation was conducted using non-AF ECGs from the CODE 15% dataset provided by the Telehealth Network of Minas Gerais, Brazil. Results In internal validation, our AI model achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.911 (95% CI: 0.902 - 0.921), with a sensitivity of 78.1% and a specificity of 89.0%. Subgroup analyses showed an AUROC of 0.893 (95% CI: 0.876 - 0.909) for patients in routine health checkups or outpatient settings, and 0.854 for patients with "Normal ECG" interpretations. In external validation with the CODE 15% dataset, the AI model exhibited an AUROC of 0.884 (95% CI: 0.869-0.900), which increased to 0.906 (95% CI: 0.893-0.919) when adjusted for age and sex (Figure 1). In the subset of patients with "Normal ECG" interpretations, the AUROC was 0.826 (95% CI: 0.769-0.883), increasing to 0.861 (95% CI: 0.814-0.908) after applying the same adjustments. Conclusions Our AI-powered SR ECG interpretation model demonstrated excellent diagnostic performance in predicting paroxysmal AF, with valid performance in the Brazilian population as well. This suggests that the model has potential for broad application across different ethnic groups.Figure 1

Cardiovascular disease remains a leading cause of death and a considerable disease burden, especially in low to middle income countries. With increasing urbanization and epidemiologic transition, there is a significant increase in the prevalence of life-style related risk factors which contribute to the global burden of cardiovascular disease and in particular heart disease. With growing evidence highlighting the catastrophic physical and financial effects of heart disease and its associated risk factors, there is a need to detect heart disease via preventive or screening programs in a cost effective manner. There are several diagnostic tests available, among these the 12-lead electrocardiogram (ECG). Previous studies have examined the ECG in epidemiological studies in high income countries and reported moderate sensitivity and high specificity. However, there is a paucity of studies from low to middle income countries and in resource poor locations. The aim of this thesis was to systematically examine the 12-lead ECG in two diverse resource poor communities- Soweto in South Africa (from the Heart of Soweto study) and Alice Springs (from the Heart of the Heart study) in the Northern Territories, Australia. ECG analysis was undertaken in those deemed free of heart disease and in those with common forms of heart disease using the Minnesota code and the Sokolow Lyon voltage criteria (to measure left ventricular hypertrophy). As well as obtaining a 12-lead ECG, comprehensive clinical assessments were performed including anthropometric measurements, blood pressure measurements, heart and lung sounds, recording of any clinical symptoms such as angina, breathlessness, oedema, dizziness and palpitations. Echocardiography was obtained on the majority of participants and this allowed analysis with electrocardiographic data to be made. Where applicable, blood tests and other diagnostic tests were performed as well as documentation of previous medical history. The results in this thesis present data from two locations where there has previously been a paucity of contemporary data available. This study evaluated more than 2,700 ECGs for the purpose of describing ECG characteristics and data in those with established disease and confirmed heart disease-free participants. Using echocardiography to confirm individuals free of heart disease in Sowetans, a total of 387 ECGs were analysed. Of these, 27% demonstrated major ECG abnormalities (defined as Q waves, ST segment changes and conduction abnormalities) and 42% demonstrated minor ECG abnormalities (namely tall R waves, inverted T waves and ST elevation). These findings suggest in this population, further delineation is required between these normal physiological variants (that is minor abnormalities) and abnormal characteristics associated with underlying cardiac pathologies. Specifically examining those with heart failure (n=756), major ECGs abnormalities were evident in 91% of ECGs and 97% of ECGs demonstrated minor abnormalities. Further analyses across a broad spectrum of conditions including heart failure and cardiomyopathies (n=1927) revealed a high proportion of ECGs with major ECG abnormalities (54%) including LVH (as measured by Sokolow Lyon > 38mm), bundle branch blocks and major Q waves and ST/T wave changes. All of these ECG characteristics have previously been identified as important prognostic markers in those with diagnosed heart disease. Although no distinctive ECG characteristics were associated with specific conditions, but rather an increased prevalence of major abnormalities across the spectrum from coronary artery disease to dilated cardiomyopathy with valve disease was observed. A corresponding decline in cardiac dysfunction was also evident in echocardiographic parameters. Analysis of ECGs from Australian Aboriginals (n=340) validated the commonly applied Caucasian parameters in this cohort with no deviation from the currently used parameters. In those with heart disease (n=70), rheumatic heart disease and valve disease were common and major ECG abnormalities evident in nearly half of these ECGs. To determine if the ECG was a valid method of detecting underlying heart disease, sensitivity and specificity was calculated using the echocardiogram as the gold standard. Calculating the sensitivity in the African cohort revealed a sensitivity of 88% and specificity of 48% while the Heart of the Heart study revealed a sensitivity of 54% and specificity of 71%. Based on these findings the 12-lead ECG has a limited ability to detect underlying heart disease with a high rate of false negatives. Clinically, these data demonstrate the complexity of disease in this cohort and suggests that the presence of major ECG abnormalities strongly supports underlying heart disease and should be further investigated by echocardiography but given the high number of ECG abnormalities in those deemed heart disease free, the 12-lead ECG has a limited ability to detect underlying heart disease in these resource poor settings.