Human versus machine: does artificial intelligence add value to identification of hypertrophic cardiomyopathy in pediatric patients?
European Heart Journal(2022)
摘要
Abstract Introduction An abnormal increase in left ventricular thickness is a hallmark of hypertrophic cardiomyopathy (HCM). Although standard measures of left ventricular voltage abnormalities on electrocardiograms (ECGs) have high false positive rates and poor correlation to left ventricular thickness, ECGs continue to be part of most screening programs. We developed a machine learning model for HCM classification from ECG in a pediatric cohort and compared its efficacy to that of clinician specialists. Purpose To compare clinician-based predictions of HCM to a machine learning model trained to classify pediatric patients with HCM utilizing 12-lead ECG. Methods ECGs from patients <19 years with HCM, including those with HCM gene mutations, were compared to those from age- and sex-matched controls with normal heart structure and function on echocardiogram. Patients with a known history of primary causes of left ventricular hypertrophy such as aortic stenosis or glycogen storage disease were excluded. A cascaded convolutional neural network was developed combining a residual neural network with a 2-layer dense neural network with 10-fold cross validation. The performance of the machine learning based HCM classification was compared to that of two independent clinicians. Results Analytic sample included data from 82 patients with clinical HCM compared to 91 healthy control subjects. The machine model AUC was 0.89 (0.84–0.94). Clinician inter-rater reliability was 0.8. The clinicians had a higher specificity (97% vs 82%) but lower sensitivity (50% vs 78%) than the machine learning model. Compared to clinician-classification, the positive predictive value was lower in the machine model (82% vs 93%). Conclusion In this preliminary study, machine learning classification of HCM utilizing 12-lead ECG had greater sensitivity than that of clinician interpretation. Machine learning may play a role in screening and diagnosis, including in subjects with normal-appearing ECG. Evaluation of its utility in a larger cohort is needed. Funding Acknowledgement Type of funding sources: Public hospital(s). Main funding source(s): Le Bonheur Children's Hospital
更多查看译文
关键词
artificial intelligence,pediatric patients,machine
AI 理解论文
溯源树
样例
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要