Background When using a smartwatch to obtain electrocardiogram (ECG) signals from multiple leads, the device has to be placed on different parts of the body sequentially. The ECG signals measured from different leads are asynchronous. Artificial intelligence (AI) models for asynchronous ECG signals have barely been explored. Objective We aimed to develop an AI model for detecting acute myocardial infarction using asynchronous ECGs and compare its performance with that of the automatic ECG interpretations provided by a commercial ECG analysis software. We sought to evaluate the feasibility of implementing multiple lead–based AI-enabled ECG algorithms on smartwatches. Moreover, we aimed to determine the optimal number of leads for sufficient diagnostic power. Methods We extracted ECGs recorded within 24 hours from each visit to the emergency room of Ajou University Medical Center between June 1994 and January 2018 from patients aged 20 years or older. The ECGs were labeled on the basis of whether a diagnostic code corresponding to acute myocardial infarction was entered. We derived asynchronous ECG lead sets from standard 12-lead ECG reports and simulated a situation similar to the sequential recording of ECG leads via smartwatches. We constructed an AI model based on residual networks and self-attention mechanisms by randomly masking each lead channel during the training phase and then testing the model using various targeting lead sets with the remaining lead channels masked. Results The performance of lead sets with 3 or more leads compared favorably with that of the automatic ECG interpretations provided by a commercial ECG analysis software, with 8.1%-13.9% gain in sensitivity when the specificity was matched. Our results indicate that multiple lead-based AI-enabled ECG algorithms can be implemented on smartwatches. Model performance generally increased as the number of leads increased (12-lead sets: area under the receiver operating characteristic curve [AUROC] 0.880; 4-lead sets: AUROC 0.858, SD 0.008; 3-lead sets: AUROC 0.845, SD 0.011; 2-lead sets: AUROC 0.813, SD 0.018; single-lead sets: AUROC 0.768, SD 0.001). Considering the short amount of time needed to measure additional leads, measuring at least 3 leads—ideally more than 4 leads—is necessary for minimizing the risk of failing to detect acute myocardial infarction occurring in a certain spatial location or direction. Conclusions By developing an AI model for detecting acute myocardial infarction with asynchronous ECG lead sets, we demonstrated the feasibility of multiple lead-based AI-enabled ECG algorithms on smartwatches for automated diagnosis of cardiac disorders. We also demonstrated the necessity of measuring at least 3 leads for accurate detection. Our results can be used as reference for the development of other AI models using sequentially measured asynchronous ECG leads via smartwatches for detecting various cardiac disorders.

Biosignals such as electrocardiogram or photoplethysmogram are widely used for determining and monitoring the medical condition of patients. It was recently discovered that more information could be gathered from biosignals by applying artificial intelligence (AI). At present, one of the most impactful advancements in AI is deep learning. Deep learning-based models can extract important features from raw data without feature engineering by humans, provided the amount of data is sufficient. This AI-enabled feature presents opportunities to obtain latent information that may be used as a digital biomarker for detecting or predicting a clinical outcome or event without further invasive evaluation. However, the black box model of deep learning is difficult to understand for clinicians familiar with a conventional method of analysis of biosignals. A basic knowledge of AI and machine learning is required for the clinicians to properly interpret the extracted information and to adopt it in clinical practice. This review covers the basics of AI and machine learning, and the feasibility of their application to real-life situations by clinicians in the near future.

Coronary artery calcium (CAC), which can be measured in various types of computed tomography (CT) examinations, is a hallmark of coronary artery atherosclerosis. However, despite the clinical value of CAC scores in predicting cardiovascular events, routine measurement of CAC scores is limited due to high cost, radiation exposure, and lack of widespread availability. It would be of great clinical significance if CAC could be predicted by electrocardiograms (ECGs), which are cost-effective and routinely performed during various medical checkups. We aimed to develop binary classification artificial intelligence (AI) models that predict CAC using only ECGs as input. Moreover, we aimed to address the generalizability of our model in different environments by externally validating our model on a dataset from a different institution. Among adult patients, standard 12-lead ECGs were extracted if measured within 60 days before or after the CAC scores, and labeled with the corresponding CAC scores. We constructed deep convolutional neural network models based on residual networks using only the raw waveforms of the ECGs as input, predicting CAC at different levels, namely CAC score ≥100, ≥400 and ≥1,000. Our AI models performed well in predicting CAC in the training and internal validation dataset [area under the receiver operating characteristics curve (AUROC) 0.753 ± 0.009, 0.802 ± 0.027, and 0.835 ± 0.024 for the CAC score ≥100, ≥400, and ≥1,000 model, respectively]. Our models also performed well in the external validation dataset (AUROC 0.718, 0.777 and 0.803 for the CAC score ≥100, ≥400, and ≥1,000 model, respectively), indicating that our model can generalize well to different but plausibly related populations. Model performance in terms of AUROC increased in the order of CAC score ≥100, ≥400, and ≥1,000 model, indicating that higher CAC scores might be associated with more prominent structural changes of the heart detected by the model. With our AI models, a substantial proportion of previously unrecognized CAC can be afforded with a risk stratification of CAC, enabling initiation of prophylactic therapy, and reducing the adverse consequences related to ischemic heart disease.

Background Heart rate variability (HRV) extracted from electrocardiogram measured for a short period during a resting state is clinically used as a bio-signal reflecting the emotional state. However, as interest in wearable devices increases, greater attention is being paid to HRV extracted from long-term electrocardiogram, which may contain additional clinical information. The purpose of this study was to examine the characteristics of HRV parameters extracted through long-term electrocardiogram and explore the differences between participants with and without depression and anxiety symptoms. Methods Long-term electrocardiogram was acquired from 354 adults with no psychiatric history who underwent Holter monitoring. Evening and nighttime HRV and the ratio of nighttime-to-evening HRV were compared between 127 participants with depressive symptoms and 227 participants without depressive symptoms. Comparisons were also made between participants with and without anxiety symptoms. Results Absolute values of HRV parameters did not differ between groups based on the presence of depressive or anxiety symptoms. Overall, HRV parameters increased at nighttime compared to evening. Participants with depressive symptoms showed a significantly higher nighttime-to-evening ratio of high-frequency HRV than participants without depressive symptoms. The nighttime-to-evening ratio of HRV parameters did not show a significant difference depending on the presence of anxiety symptoms. Conclusion HRV extracted through long-term electrocardiogram showed circadian rhythm. Depression may be associated with changes in the circadian rhythm of parasympathetic tone.

Background: Conventional cardiovascular risk prediction models provide insights into population-level risk factors and have been widely adopted in clinical practice. However, these models have limited generalizability and flexibility. Large language models (LLMs) have demonstrated remarkable proficiency for use in various industries. Methods: In this study, we have investigated the feasibility of Large Language Models (LLMs) such as ChatGPT-3.5, ChatGPT-4, and Bard for predicting 10-year cardiovascular risk of a patient. We used data from the UK Biobank Cohort, a major biomedical database in the UK, and the Korean Genome and Epidemiology Study (KoGES), a large-scale prospective study in Korea, for additional validation and multi-institutional research. These databases provided a wide array of information including age, sex, medical history, lipid profile, blood pressure, and physical measurement. Based on these data, we generated language sentences for individual analysis and input these into the LLM to derive results. The performance of the LLMs was then compared with the Framingham Risk Score (FRS), a conventional risk prediction model, using this real-world data. We confirmed the model performance of both the LLMs and FRS, evaluating their accuracy, sensitivity, specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), and F1 score. Their performance in predicting 10-year cardiovascular risk was compared through Kaplan–Meier survival analysis and Cox-hazard ratio analysis. Findings: GPT-4 achieved performance comparable to the FRS in cardiovascular risk prediction in both the UK Biobank {accuracy (0.834 vs 0.773) and F1 score (0.138 vs 0.132)} and KoGES {accuracy (0.902 vs 0.874)}. The Kaplan–Meier survival analysis of GPT-4 demonstrated distinct survival patterns among groups, which revealed a strong association between the GPT risk prediction output and survival outcomes. The additional analysis of limited variables using GPT-3.5 indicated that ChatGPT′s prediction performance was preserved despite the omission of a few variables in the prompt, especially without physical measurement data Interpretation: This study proposed that ChatGPT can achieve performance comparable to conventional models in predicting cardiovascular risk. Furthermore, ChatGPT exhibits enhanced accessibility, flexibility, and the ability to provide user-friendly outputs. With the evolution of LLMs, such as ChatGPT, studies should focus on applying LLMs to various medical scenarios and subsequently optimizing their performance.