RESUMO
The efficacy of convolutional neural network (CNN)-enhanced electrocardiography (ECG) in detecting hypertrophic cardiomyopathy (HCM) and dilated HCM (dHCM) remains uncertain in real-world applications. This retrospective study analyzed data from 19,170 patients (including 140 HCM or dHCM) in the Shinken Database (2010-2017). We evaluated the sensitivity, positive predictive rate (PPR), and F1 score of CNN-enhanced ECG in a ''basic diagnosis'' model (total disease label) and a ''comprehensive diagnosis'' model (including disease subtypes). Using all-lead ECG in the "basic diagnosis" model, we observed a sensitivity of 76%, PPR of 2.9%, and F1 score of 0.056. These metrics improved in cases with a diagnostic probability of ≥ 0.9 and left ventricular hypertrophy (LVH) on ECG: 100% sensitivity, 8.6% PPR, and 0.158 F1 score. The ''comprehensive diagnosis'' model further enhanced these figures to 100%, 13.0%, and 0.230, respectively. Performance was broadly consistent across CNN models using different lead configurations, particularly when including leads viewing the lateral walls. While the precision of CNN models in detecting HCM or dHCM in real-world settings is initially low, it improves by targeting specific patient groups and integrating disease subtype models. The use of ECGs with fewer leads, especially those involving the lateral walls, appears comparably effective.
Assuntos
Cardiomiopatia Hipertrófica , Eletrocardiografia , Redes Neurais de Computação , Humanos , Cardiomiopatia Hipertrófica/diagnóstico , Cardiomiopatia Hipertrófica/fisiopatologia , Cardiomiopatia Hipertrófica/complicações , Eletrocardiografia/métodos , Estudos Retrospectivos , Masculino , Feminino , Pessoa de Meia-Idade , Valor Preditivo dos Testes , Adulto , IdosoRESUMO
Background: We developed a convolutional neural network (CNN) model to detect atrial fibrillation (AF) using the sinus rhythm ECG (SR-ECG). However, the diagnostic performance of the CNN model based on different ECG leads remains unclear. MethodsâandâResults: In this retrospective analysis of a single-center, prospective cohort study, we identified 616 AF cases and 3,412 SR cases for the modeling dataset among new patients (n=19,170). The modeling dataset included SR-ECGs obtained within 31 days from AF-ECGs in AF cases and SR cases with follow-up ≥1,095 days. We evaluated the CNN model's performance for AF detection using 8-lead (I, II, and V1-6), single-lead, and double-lead ECGs through 5-fold cross-validation. The CNN model achieved an area under the curve (AUC) of 0.872 (95% confidence interval (CI): 0.856-0.888) and an odds ratio of 15.24 (95% CI: 12.42-18.72) for AF detection using the eight-lead ECG. Among the single-lead and double-lead ECGs, the double-lead ECG using leads I and V1 yielded an AUC of 0.871 (95% CI: 0.856-0.886) with an odds ratio of 14.34 (95% CI: 11.64-17.67). Conclusions: We assessed the performance of a CNN model for detecting AF using eight-lead, single-lead, and double-lead SR-ECGs. The model's performance with a double-lead (I, V1) ECG was comparable to that of the 8-lead ECG, suggesting its potential as an alternative for AF screening using SR-ECG.
RESUMO
Background: The potential of utilizing artificial intelligence with electrocardiography (ECG) for initial screening of aortic dissection (AD) is promising. However, achieving a high positive predictive rate (PPR) remains challenging. Methods and results: This retrospective analysis of a single-center, prospective cohort study (Shinken Database 2010-2017, N = 19,170) used digital 12-lead ECGs from initial patient visits. We assessed a convolutional neural network (CNN) model's performance for AD detection with eight-lead (I, II, and V1-6), single-lead, and double-lead (I, II) ECGs via five-fold cross-validation. The mean age was 63.5 ± 12.5 years for the AD group (n = 147) and 58.1 ± 15.7 years for the non-AD group (n = 19,023). The CNN model achieved an area under the curve (AUC) of 0.936 (standard deviation [SD]: 0.023) for AD detection with eight-lead ECGs. In the entire cohort, the PPR was 7 %, with 126 out of 147 AD cases correctly diagnosed (sensitivity 86 %). When applied to patients with D-dimer levels ≥1 µg/dL and a history of hypertension, the PPR increased to 35 %, with 113 AD cases correctly identified (sensitivity 86 %). The single V1 lead displayed the highest diagnostic performance (AUC: 0.933, SD: 0.03), with PPR improvement from 8 % to 38 % within the same population. Conclusions: Our CNN model using ECG data for AD detection achieved an over 30% PPR when applied to patients with elevated D-dimer levels and hypertension history while maintaining sensitivity. A similar level of performance was observed with a single-lead V1 ECG in the CNN model.
RESUMO
INTRODUCTION: In the ongoing COVID-19 pandemic, the development of a system that would prevent the infection of healthcare providers is in urgent demand. We sought to investigate the feasibility and validity of a telemedicine-based system in which healthcare providers remotely check the vital signs measured by patients with COVID-19. METHODS: Patients hospitalized with confirmed or suspected COVID-19 measured and uploaded their vital signs to secure cloud storage. Additionally, the respiratory rates were monitored using a mat-type sensor placed under the bed. We assessed the time until the values became available on the Cloud and the agreements between the patient-measured vital signs and simultaneous healthcare provider measurements. RESULTS: Between 26 May-23 September 2020, 3835 vital signs were measured and uploaded to the cloud storage by the patients (n=16, median 72 years old, 31% women). All patients successfully learned how to use these devices with a 10-minute lecture. The median time until the measurements were available on the cloud system was only 0.35 min, and 95.2% of the vital signs were available within 5 min of the measurement. The agreement between the patients' and healthcare providers' measurements was excellent for all parameters. Interclass coefficient correlations were as follows: systolic (0.92, p<0.001), diastolic blood pressure (0.86, p<0.001), heart rate (0.89, p<0.001), peripheral oxygen saturation (0.92, p<0.001), body temperature (0.83, p<0.001), and respiratory rates (0.90, p<0.001). CONCLUSIONS: Telemedicine-based self-assessment of vital signs in patients with COVID-19 was feasible and reliable. The system will be a useful alternative to traditional vital sign measurements by healthcare providers during the COVID-19 pandemic.
Assuntos
COVID-19 , Telemedicina , Humanos , Feminino , Idoso , Masculino , Projetos Piloto , COVID-19/diagnóstico , Pandemias , Autoavaliação (Psicologia) , Monitorização Fisiológica , Sinais VitaisRESUMO
Background: There is increasing evidence that 12-lead electrocardiograms (ECG) can be used to predict biological age, which is associated with cardiovascular events. However, the utility of artificial intelligence (AI)-predicted age using ECGs remains unclear. Methods: Using a single-center database, we developed an AI-enabled ECG using 17 042 sinus rhythm ECGs (SR-ECG) to predict chronological age (CA) with a convolutional neural network that yields AI-predicted age. Using the 5-fold cross validation method, AI-predicted age deriving from the test dataset was yielded for all ECGs. The incidence by AgeDiff and the areas under the curve by receiver operating characteristic curve with AI-predicted age for cardiovascular events were analyzed. Results: During the mean follow-up period of 460.1 days, there were 543 cardiovascular events. The annualized incidence of cardiovascular events was 2.24 %, 2.44 %, and 3.01 %/year for patients with AgeDiff < -6, -6 to ≤6, and >6 years, respectively. The areas under the curve for cardiovascular events with CA and AI-predicted age, respectively, were 0.673 and 0.679 (Delong's test, P = 0.388) for all patients; 0.642 and 0.700 (P = 0.003) for younger patients (CA < 60 years); and 0.584 and 0.570 (P = 0.268) for older patients (CA ≥ 60 years). Conclusions: AI-predicted age using 12-lead ECGs showed superiority in predicting cardiovascular events compared with CA in younger patients, but not in older patients.
RESUMO
Background: This study sought to develop an artificial intelligence-derived model to detect the dilated phase of hypertrophic cardiomyopathy (dHCM) on digital electrocardiography (ECG) and to evaluate the performance of the model applied to multiple-lead or single-lead ECG. Methods: This is a retrospective analysis using a single-center prospective cohort study (Shinken Database 2010-2017, n = 19,170). After excluding those without a normal P wave on index ECG (n = 1,831) and adding dHCM patients registered before 2009 (n = 39), 17,378 digital ECGs were used. Totally 54 dHCM patients were identified of which 11 diagnosed at baseline, 4 developed during the time course, and 39 registered before 2009. The performance of the convolutional neural network (CNN) model for detecting dHCM was evaluated using eight-lead (I, II, and V1-6), single-lead, and double-lead (I, II) ECGs with the five-fold cross validation method. Results: The area under the curve (AUC) of the CNN model to detect dHCM (n = 54) with eight-lead ECG was 0.929 (standard deviation [SD]: 0.025) and the odds ratio was 38.64 (SD 9.10). Among the single-lead and double-lead ECGs, the AUC was highest with the single lead of V5 (0.953 [SD: 0.038]), with an odds ratio of 58.89 (SD:68.56). Conclusion: Compared with the performance of eight-lead ECG, the most similar performance was achieved with the model with a single V5 lead, suggesting that this single-lead ECG can be an alternative to eight-lead ECG for the screening of dHCM.
RESUMO
BACKGROUND: This study aimed to increase the knowledge on how to enhance the performance of artificial intelligence (AI)-enabled electrocardiography (ECG) to detect atrial fibrillation (AF) on sinus rhythm ECG (SR-ECG). METHODS: It is a retrospective analysis of a single-center, prospective cohort study (Shinken Database). We developed AI-enabled ECG using SR-ECG to predict AF with a convolutional neural network (CNN). Among new patients in our hospital (n = 19,170), 276 AF label (having ECG on AF [AF-ECG] in the ECG database) and 1896 SR label with following three conditions were identified in the derivation dataset: (1) without structural heart disease, (2) in AF label, SR-ECG was taken within 31 days from AF-ECG, and (3) in SR label, follow-up ≥ 1,095 days. Three patterns of AF label were analyzed by timing of SR-ECG to AF-ECG (before/after/before-or-after, CNN algorithm 1 to 3). The outcome measurement was area under the curve (AUC), sensitivity, specificity, accuracy, and F1 score. As an extra-testing dataset, the performance of AI-enabled ECG was tested in patients with structural heart disease. RESULTS: The AUC of AI-enabled ECG with CNN algorithm 1, 2, and 3 in the derivation dataset was 0.83, 0.88, and 0.86, respectively; when tested in patients with structural heart disease, 0.75, 0.81, and 0.78, respectively. CONCLUSION: We confirmed high performance of AI-enabled ECG to detect AF on SR-ECG in patients without structural heart disease. The performance enhanced especially when SR-ECG after index AF-ECG was included in the algorithm, which was consistent in patients with structural heart disease.
RESUMO
Background: The coronavirus disease 2019 (COVID-19) pandemic impacts not only patients but also healthcare providers. This study seeks to investigate whether a telemedicine system reduces physical contact in addressing the COVID-19 pandemic and mitigates nurses' distress and depression. Methods: Patients hospitalized with COVID-19 in 4 hospitals and 1 designated accommodation measured and uploaded their vital signs to secure cloud storage for remote monitoring. Additionally, a mat-type sensor placed under the bed monitored the patients' respiratory rates. Using the pre-post prospective design, visit counts and health care providers' mental health were assessed before and after the system was introduced. Results: A total of 100 nurses participated in the study. We counted the daily visits for 48 and 69 patients with and without using the telemedicine system. The average patient visits were significantly less with the system (16.3 [5.5-20.3] vs 7.5 [4.5-17.5] times/day, P = .009). Specifically, the visit count for each vital sign assessment was about half with the telemedicine system (all P < .0001). Most nurses responded that the system was easy to use (87.1%), reduced work burden (75.2%), made them feel relieved (74.3%), and was effective in reducing the infection risk in hospitals (79.1%) and nursing accommodations (95.0%). Distress assessed by Impact of Event Scale-Revised and depression by Patient Health Questionnaire-9 were at their minimum even without the system and did not show any significant difference with the system (P = .72 and .57, respectively). Conclusions: Telemedicine-based self-assessment of vital signs reduces nurses' physical contact with COVID-19 patients. Most nurses responded that the system is easy and effective in reducing healthcare providers' infection risk.