Early detection and intervention is the cornerstone for appropriate treatment of arrhythmia and prevention of complications and mortality. Although diverse deep learning models have been developed to detect arrhythmia, they have been criticized due to their unexplainable nature. In this study, we developed an explainable deep learning model (XDM) to classify arrhythmia, and validated its performance using diverse external validation data. In this retrospective study, the Sejong dataset comprising 86,802 electrocardiograms (ECGs) was used to develop and internally variate the XDM. The XDM based on a neural network-backed ensemble tree was developed with six feature modules that are able to explain the reasons for its decisions. The model was externally validated using data from 36,961 ECGs from four non-restricted datasets. During internal and external validation of the XDM, the average area under the receiver operating characteristic curves (AUCs) using a 12‑lead ECG for arrhythmia classification were 0.976 and 0.966, respectively. The XDM outperformed a previous simple multi-classification deep learning model that used the same method. During internal and external validation, the AUCs of explainability were 0.925-0.991. Our XDM successfully classified arrhythmia using diverse formats of ECGs and could effectively describe the reason for the decisions. Therefore, an explainable deep learning methodology could improve accuracy compared to conventional deep learning methods, and that the transparency of XDM can be enhanced for its application in clinical practice.

The electrocardiogram (ECG) records electrical signals in a non-invasive way to observe the condition of the heart, typically looking at the heart from 12 different directions. Several types of the cardiac disease are diagnosed by using 12-lead ECGs Recently, various wearable devices have enabled immediate access to the ECG without the use of wieldy equipment. However, they only provide ECGs with a couple of leads. This results in an inaccurate diagnosis of cardiac disease due to lacking of required leads. We propose a deep generative model for ECG synthesis from two asynchronous leads to ten leads. It first represents a heart condition referring to two leads, and then generates ten leads based on the represented heart condition. Both the rhythm and amplitude of leads generated resemble those of the original ones, while the technique removes noise and the baseline wander appearing in the original leads. As a data augmentation method, our model improves the classification performance of models compared with models using ECGs with only one or two leads.

Although left ventricular hypertrophy (LVH) has a high incidence and clinical importance, the conventional diagnosis criteria for detecting LVH using electrocardiography (ECG) has not been satisfied. We aimed to develop an artificial intelligence (AI) algorithm for detecting LVH. This retrospective cohort study involved the review of 21 286 patients who were admitted to two hospitals between October 2016 and July 2018 and underwent 12-lead ECG and echocardiography within 4 weeks. The patients in one hospital were divided into a derivation and internal validation dataset, while the patients in the other hospital were included in only an external validation dataset. An AI algorithm based on an ensemble neural network (ENN) combining convolutional and deep neural network was developed using the derivation dataset. And we visualized the ECG area that the AI algorithm used to make the decision. The area under the receiver operating characteristic curve of the AI algorithm based on ENN was 0.880 (95% confidence interval 0.877-0.883) and 0.868 (0.865-0.871) during the internal and external validations. These results significantly outperformed the cardiologist's clinical assessment with Romhilt-Estes point system and Cornell voltage criteria, Sokolov-Lyon criteria, and interpretation of ECG machine. At the same specificity, the AI algorithm based on ENN achieved 159.9%, 177.7%, and 143.8% higher sensitivities than those of the cardiologist's assessment, Sokolov-Lyon criteria, and interpretation of ECG machine. An AI algorithm based on ENN was highly able to detect LVH and outperformed cardiologists, conventional methods, and other machine learning techniques.

As the performance of a conventional track and trigger system in a rapid response system has been unsatisfactory, we developed and implemented an artificial intelligence for predicting in-hospital cardiac arrest, denoted the deep learning-based early warning system. The purpose of this study was to compare the performance of an artificial intelligence-based early warning system with that of conventional methods in a real hospital situation. This study was conducted at a hospital in which deep learning-based early warning system was implemented. We reviewed the records of adult patients who were admitted to the general ward of our hospital from April 2018 to March 2019. The study population included 8,039 adult patients. A total 83 events of deterioration occurred during the study period. The outcome was events of deterioration, defined as cardiac arrest and unexpected ICU admission. We defined a true alarm as an alarm occurring within 0.5-24 hours before a deteriorating event. We used the area under the receiver operating characteristic curve, area under the precision-recall curve, number needed to examine, and mean alarm count per day as comparative measures. The deep learning-based early warning system (area under the receiver operating characteristic curve, 0.865; area under the precision-recall curve, 0.066) outperformed the modified early warning score (area under the receiver operating characteristic curve, 0.682; area under the precision-recall curve, 0.010) and reduced the number needed to examine and mean alarm count per day by 69.2% and 59.6%, respectively. At the same specificity, deep learning-based early warning system had up to 257% higher sensitivity than conventional methods. The developed artificial intelligence based on deep-learning, deep learning-based early warning system, accurately predicted deterioration of patients in a general ward and outperformed conventional methods. This study showed the potential and effectiveness of artificial intelligence in an rapid response system, which can be applied together with electronic health records. This will be a useful method to identify patients with deterioration and help with precise decision-making in daily practice.

Screening and early diagnosis of mitral regurgitation (MR) are crucial for preventing irreversible progression of MR. In this study, we developed and validated an artificial intelligence (AI) algorithm for detecting MR using electrocardiography (ECG). This retrospective cohort study included data from two hospital. An AI algorithm was trained using 56,670 ECGs from 24,202 patients. Internal validation of the algorithm was performed with 3174 ECGs of 3174 patients from one hospital, while external validation was performed with 10,865 ECGs of 10,865 patients from another hospital. The endpoint was the diagnosis of significant MR, moderate to severe, confirmed by echocardiography. We used 500 Hz ECG raw data as predictive variables. Additionally, we showed regions of ECG that have the most significant impact on the decision-making of the AI algorithm using a sensitivity map. During the internal and external validation, the area under the receiver operating characteristic curve of the AI algorithm using a 12-lead ECG for detecting MR was 0.816 and 0.877, respectively, while that using a single-lead ECG was 0.758 and 0.850, respectively. In the 3157 non-MR individuals, those patients that the AI defined as high risk had a significantly higher chance of development of MR than the low risk group (13.9% vs. 2.6%, p < 0.001) during the follow-up period. The sensitivity map showed the AI algorithm focused on the P-wave and T-wave for MR patients and QRS complex for non-MR patients. The proposed AI algorithm demonstrated promising results for MR detecting using 12-lead and single-lead ECGs.

In emergency medical services (EMSs), accurately predicting the severity of a patient's medical condition is important for the early identification of those who are vulnerable and at high-risk. In this study, we developed and validated an artificial intelligence (AI) algorithm based on deep learning to predict the need for critical care during EMS. We conducted a retrospective observation cohort study. The algorithm was established using development data from the Korean national emergency department information system, which were collected during visits in real time from 151 emergency departments (EDs). We validated the algorithm using EMS run sheets from two EDs. The study subjects comprised adult patients who visited EDs. The endpoint was critical care, and we used age, sex, chief complaint, symptom onset to arrival time, trauma, and initial vital signs as the predicted variables. The number of patients in the development data was 8,981,181, and the validation data comprised 2604 EMS run sheets from two hospitals. The area under the receiver operating characteristic curve of the algorithm to predict the critical care was 0.867 (95% confidence interval, [0.864-0.871]). This result outperformed the Emergency Severity Index (0.839 [0.831-0.846]), Korean Triage and Acuity System (0.824 [0.815-0.832]), National Early Warning Score (0.741 [0.734-0.748]), and Modified Early Warning Score (0.696 [0.691-0.699]). The AI algorithm accurately predicted the need for the critical care of patients using information during EMS and outperformed the conventional triage tools and early warning scores.

Severe, symptomatic aortic stenosis (AS) is associated with poor prognoses. However, early detection of AS is difficult because of the long asymptomatic period experienced by many patients, during which screening tools are ineffective. The aim of this study was to develop and validate a deep learning-based algorithm, combining a multilayer perceptron and convolutional neural network, for detecting significant AS using ECGs. This retrospective cohort study included adult patients who had undergone both ECG and echocardiography. A deep learning-based algorithm was developed using 39 371 ECGs. Internal validation of the algorithm was performed with 6453 ECGs from one hospital, and external validation was performed with 10 865 ECGs from another hospital. The end point was significant AS (beyond moderate). We used demographic information, features, and 500-Hz, 12-lead ECG raw data as predictive variables. In addition, we identified which region had the most significant effect on the decision-making of the algorithm using a sensitivity map. During internal and external validation, the areas under the receiver operating characteristic curve of the deep learning-based algorithm using 12-lead ECG for detecting significant AS were 0.884 (95% CI, 0.880-0.887) and 0.861 (95% CI, 0.858-0.863), respectively; those using a single-lead ECG signal were 0.845 (95% CI, 0.841-0.848) and 0.821 (95% CI, 0.816-0.825), respectively. The sensitivity map showed the algorithm focused on the T wave of the precordial lead to determine the presence of significant AS. The deep learning-based algorithm demonstrated high accuracy for significant AS detection using both 12-lead and single-lead ECGs.

Screening and early diagnosis of pulmonary hypertension (PH) are critical for managing progression and preventing associated mortality; however, there are no tools for this purpose. We developed and validated an artificial intelligence (AI) algorithm for predicting PH using electrocardiography (ECG). This historical cohort study included data from consecutive patients from 2 hospitals. The patients in one hospital were divided into derivation (56,670 ECGs from 24,202 patients) and internal validation (3,174 ECGs from 3,174 patients) datasets, whereas the patients in the other hospital were included in only an external validation (10,865 ECGs from 10,865 patients) dataset. An AI algorithm based on an ensemble neural network was developed using 12-lead ECG signal and demographic information from the derivation dataset. The end-point was the diagnosis of PH. In addition, the interpretable AI algorithm identified which region had the most significant effect on decision making using a sensitivity map. During the internal and external validation, the area under the receiver operating characteristic curve of the AI algorithm for detecting PH was 0.859 and 0.902, respectively. In the 2,939 individuals without PH at initial echocardiography, those patients that the AI defined as having a higher risk had a significantly higher chance of developing PH than those in the low-risk group (31.5% vs 5.9%, p < 0.001) during the follow-up period. The sensitivity map showed that the AI algorithm focused on the S-wave, P-wave, and T-wave for each patient by QRS complex characteristics. The AI algorithm demonstrated high accuracy for PH prediction using 12-lead and single-lead ECGs.

Anaemia is an important health-care burden globally, and screening for anaemia is crucial to prevent multi-organ injury, irreversible complications, and life-threatening adverse events. We aimed to establish whether a deep learning algorithm (DLA) that enables non-invasive anaemia screening from electrocardiograms (ECGs) might improve the detection of anaemia. We did a retrospective, multicentre, diagnostic study in which a DLA was developed using ECGs and then internally and externally validated. We used data from two hospitals, Sejong General Hospital (hospital A) and Mediplex Sejong Hospital (hospital B), in South Korea. Data from hospital A was for DLA development and internal validation, and data from hospital B was for external validation. We included individuals who had at least one ECG with a haemoglobin measurement within 1 h of the index ECG and excluded individuals with missing demographic, electrocardiographic, or haemoglobin information. Three types of DLA were developed with 12-lead, 6-lead (limb lead), and single-lead (lead I) ECGs to detect haemoglobin concentrations of 10 g/dL or less. The DLA was built by a convolutional neural network and used 500-Hz raw ECG, age, and sex as input data. The study period ran from Oct 1, 2016, to Sept 30, 2019, in hospital A and March 1, 2017, to Sept 30, 2019, in hospital B. 40 513 patients at hospital A and 4737 patients at hospital B were eligible for inclusion. We excluded 281 patients at hospital A and 72 patients at hospital B because of missing values for clinical information and ECG data. The development dataset comprised 57 435 ECGs from 31 898 patients, and the algorithm was internally validated with 7974 ECGs from 7974 patients. The external validation dataset included 4665 ECGs from 4665 patients. 586 (internal) and 194 (external) patients within the combined dataset were found to be anaemic. During internal and external validation, the area under the receiver operating characteristics curve (AUROC) of the DLA using a 12-lead ECG for detecting anaemia was 0·923 for internal validation and 0·901 for external validation. Using a 90% sensitivity operating point for the development data, the sensitivity, specificity, negative predictive value, and positive predictive value of internal validation were 89·8%, 81·5%, 99·4%, and 20·0%, respectively, and those of external validation were 86·1%, 76·2%, 99·2%, and 13·5%, respectively. The DLA focused on the QRS complex for deciding the presence of anaemia in a sensitivity map. The AUROCs of DLAs using 6 leads and a single lead were in the range of 0·841–0·890. Interpretation In this study, using raw ECG data, a DLA accurately

In-hospital cardiac arrest is a major burden in health care. Although several track-and-trigger systems are used to predict cardiac arrest, they often have unsatisfactory performances. We hypothesized that a deeplearning-based artificial intelligence algorithm (DLA) could effectively predict cardiac arrest using electrocardiography (ECG). We developed and validated a DLA for predicting cardiac arrest using ECG. We conducted a retrospective study that included 47,505 ECGs of 25,672 adult patients admitted to two hospitals, who underwent at least one ECG from October 2016 to September 2019. The endpoint was occurrence of cardiac arrest within 24 h from ECG. Using subgroup analyses in patients who were initially classified as non-event, we confirmed the delayed occurrence of cardiac arrest and unexpected intensive care unit transfer over 14 days. We used 32,294 ECGs of 10,461 patients and 4483 ECGs of 4483 patients from a hospital were used as development and internal validation data, respectively. Additionally, 10,728 ECGs of 10,728 patients from another hospital were used as external validation data, which confirmed the robustness of the developed DLA. During internal and external validation, the areas under the receiver operating characteristic curves of the DLA in predicting cardiac arrest within 24 h were 0.913 and 0.948, respectively. The high risk group of the DLA showed a significantly higher hazard for delayed cardiac arrest (5.74% vs. 0.33%, P < 0.001) and unexpected intensive care unit transfer (4.23% vs. 0.82%, P < 0.001). A sensitivity map of the DLA displayed the ECG regions used to predict cardiac arrest, with the DLA focused most on the QRS complex. Our DLA successfully predicted cardiac arrest using diverse formats of ECG. The results indicate that cardiac arrest could be screened and predicted not only with a conventional 12-lead ECG, but also with a single-lead ECG using a wearable device that employs our DLA.