In the midst of the coronavirus disease 2019 (COVID-19) pandemic, we are seeing widespread disease burden affecting patients of all ages across the globe. However, much remains to be understood as clinicians, epidemiologists, and researchers alike are working to describe and characterize the disease process while caring for patients at the frontlines. We describe the case of a 6-month-old infant admitted and diagnosed with classic Kawasaki disease, who also screened positive for COVID-19 in the setting of fever and minimal respiratory symptoms. The patient was treated per treatment guidelines, with intravenous immunoglobulin and high-dose aspirin, and subsequently defervesced with resolution of her clinical symptoms. The patient’s initial echocardiogram was normal, and she was discharged within 48 hours of completion of her intravenous immunoglobulin infusion, with instruction to quarantine at home for 14 days from the date of her positive test results for COVID-19. Further study of the clinical presentation of pediatric COVID-19 and the potential association with Kawasaki disease is warranted, as are the indications for COVID-19 testing in the febrile infant.

This is a prepublication version of an article that has undergone peer review and been accepted for publication but is not the final version of record. This paper may be cited using the DOI and date of access. This paper may contain information that has errors in facts, figures, and statements, and will be corrected in the final published version. The journal is providing an early version of this article to expedite access to this information. The American Academy of Pediatrics, the editors, and authors are not responsible for inaccurate information and data described in this version.

BackgroundCine balanced steady-state free precession (SSFP), the preferred sequence for ventricular function, demands uninterrupted radio frequency (RF) excitation to maintain the steady-state during suspended respiration. This is difficult to accomplish in sedated children. In this work, we validate a respiratory triggered (RT) SSFP sequence that drives the magnetization to steady-state before commencing retrospectively cardiac gated cine acquisition in a sedated pediatric population.MethodsThis prospective study was performed on 20 sedated children with congenital heart disease (8.6 ± 4 yrs). Identical imaging parameters were used for multiple number of signal averages (MN) and RT cine SSFP sequences covering both the ventricles in short-axis (SA) orientation. Image quality assessment and quantitative volumetric analysis was performed on the datasets by two blinded observers. One-sided Wilcoxon signed rank test and Box plot analysis were performed to compare the clinical scores. Bland-Altman (BA) analysis was performed on LV and RV volumes.ResultsScan duration for SA stack using RT-SSFP (3.9 ± 0.8 min) was slightly shorter than MN-SSFP (4.6 ± 0.9 min) acquisitions. The endocardial edge definition was significantly better for RT than MN, blood to myocardial contrast was better for RT than MN without reaching statistical significance, and inter slice alignment was comparable. BA analysis indicates that the variability of volumetric indices between RT and MN is comparable to inter and intra-observer variability reported in the literature.ConclusionsThe free breathing RT-SSFP sequence allows diagnostic images in sedated children with significantly better edge definition when compared to MN-SSFP, without any penalty for total scan time.

Biomedical repositories such as the UK Biobank provide increasing access to prospectively collected cardiac imaging, however these data are unlabeled, which creates barriers to their use in supervised machine learning. We develop a weakly supervised deep learning model for classification of aortic valve malformations using up to 4,000 unlabeled cardiac MRI sequences. Instead of requiring highly curated training data, weak supervision relies on noisy heuristics defined by domain experts to programmatically generate large-scale, imperfect training labels. For aortic valve classification, models trained with imperfect labels substantially outperform a supervised model trained on hand-labeled MRIs. In an orthogonal validation experiment using health outcomes data, our model identifies individuals with a 1.8-fold increase in risk of a major adverse cardiac event. This work formalizes a deep learning baseline for aortic valve classification and outlines a general strategy for using weak supervision to train machine learning models using unlabeled medical images at scale.