Purpose To compare the performance of a deep-learning bone age assessment model based on hand radiographs with that of expert radiologists and that of existing automated models. Materials and Methods The institutional review board approved the study. A total of 14 036 clinical hand radiographs and corresponding reports were obtained from two children's hospitals to train and validate the model. For the first test set, composed of 200 examinations, the mean of bone age estimates from the clinical report and three additional human reviewers was used as the reference standard. Overall model performance was assessed by comparing the root mean square (RMS) and mean absolute difference (MAD) between the model estimates and the reference standard bone ages. Ninety-five percent limits of agreement were calculated in a pairwise fashion for all reviewers and the model. The RMS of a second test set composed of 913 examinations from the publicly available Digital Hand Atlas was compared with published reports of an existing automated model. Results The mean difference between bone age estimates of the model and of the reviewers was 0 years, with a mean RMS and MAD of 0.63 and 0.50 years, respectively. The estimates of the model, the clinical report, and the three reviewers were within the 95% limits of agreement. RMS for the Digital Hand Atlas data set was 0.73 years, compared with 0.61 years of a previously reported model. Conclusion A deep-learning convolutional neural network model can estimate skeletal maturity with accuracy similar to that of an expert radiologist and to that of existing automated models. RSNA, 2017 An earlier incorrect version of this article appeared online. This article was corrected on January 19, 2018.

BACKGROUND AND PURPOSE: Enterovirus D68 was responsible for widespread outbreaks of respiratory illness throughout the United States in August and September 2014. During this time, several patients presented to our institution with acute flaccid paralysis and cranial nerve dysfunction. The purpose of this report is to describe the unique imaging findings of this neurologic syndrome occurring during an enterovirus D68 outbreak.

Purpose To determine whether repeated exposure of the pediatric brain to a linear gadolinium-based contrast agent (GBCA) is associated with an increase in signal intensity (SI) relative to that in GBCA-naive control subjects at unenhanced T1-weighted magnetic resonance (MR) imaging. Materials and Methods This single-center, retrospective study was approved by the institutional review board and compliant with HIPAA. The authors evaluated 46 pediatric patients who had undergone at least three GBCA-enhanced MR examinations (30 patients for two-group analysis and 16 for pre- and post-GBCA exposure comparisons) and 57 age-matched GBCA-naive control subjects. The SI in the globus pallidus, thalamus, dentate nucleus, and pons was measured at unenhanced T1-weighted MR imaging. Globus pallidus-thalamus and dentate nucleus-pons SI ratios were calculated and compared between groups and relative to total cumulative gadolinium dose, age, sex, and number of and mean time between GBCA-enhanced examinations. Analysis included the Wilcoxon signed rank test, Wilcoxon rank sum test, and Spearman correlation coefficient. Results Patients who underwent multiple GBCA-enhanced examinations had increased SI ratios within the dentate nucleus (mean SI ratio ± standard error of the mean for two-group comparison: 1.007 ± 0.0058 for GBCA-naive group and 1.046 ± 0.0060 for GBCA-exposed group [P < .001]; mean SI ratio for pre- and post-GBCA comparison: 0.995 ± 0.0062 for pre-GBCA group and 1.035 ± 0.0063 for post-GBCA group [P < .001]) but not the globus pallidus (mean SI ratio for two-group comparison: 1.131 ± 0.0070 for GBCA-naive group and 1.014 ± 0.0091 for GBCA-exposed group [P = .21]; mean SI ratio for pre- and post-GBCA comparison: 1.068 ± 0.0094 for pre-GBCA group and 1.093 ± 0.0134 for post-GBCA group [P = .12]). There was a significant correlation between dentate nucleus SI and total cumulative gadolinium dose (r = 0.4; 95% confidence interval [CI]: 0.03, 0.67; P = .03), but not between dentate nucleus SI and patient age (r = 0.23; 95% CI: -0.15, 0.56; P = .22), sex (mean SI ratio: 1.046 ± 0.0072 for boys and 1.045 ± 0.0110 for girls; P = .88), number of contrast-enhanced examinations (r = 0.13; 95% CI: -0.25, 0.48; P = .49), or time between contrast-enhanced examinations (r = -0.06; 95% CI: -0.42, 0.32; P = .75). Conclusion SI in the pediatric brain increases on unenhanced T1-weighted MR images with repeated exposure to a linear GBCA. RSNA, 2016.

BACKGROUND: Computed tomography (CT) is commonly used for children when there is concern for traumatic brain injury (TBI) and is a significant source of ionizing radiation. Our objective was to determine the feasibility and accuracy of fast MRI (motion-tolerant MRI sequences performed without sedation) in young children. METHODS: In this prospective cohort study, we attempted fast MRI in children ,6 years old who had head CT performed and were seen in the emergency department of a single, level 1 pediatric trauma center. Fast MRI sequences included 3T axial and sagittal T2 single-shot turbo spin echo, axial T1 turbo field echo, axial fluid-attenuated inversion recovery, axial gradient echo, and axial diffusion-weighted single-shot turbo spin echo planar imaging. Feasibility was assessed by completion rate and imaging time. Fast MRI accuracy was measured against CT findings of TBI, including skull fracture, intracranial hemorrhage, or parenchymal injury. RESULTS: Among 299 participants, fast MRI was available and attempted in 225 (75%) and completed in 223 (99%). Median imaging time was 59 seconds (interquartile range 52-78) for CT and 365 seconds (interquartile range 340-392) for fast MRI. TBI was identified by CT in 111 (50%) participants, including 81 skull fractures, 27 subdural hematomas, 24 subarachnoid hemorrhages, and 35 other injuries. Fast MRI identified TBI in 103 of these (sensitivity 92.8%; 95% confidence interval 86.3-96.8), missing 6 participants with isolated skull fractures and 2 with subarachnoid hemorrhage. CONCLUSIONS: Fast MRI is feasible and accurate relative to CT in clinically stable children with concern for TBI.

Deep learning (DL) is a widely applied mathematical modeling technique. Classically, DL models utilize large volumes of training data, which are not available in many healthcare contexts. For patients with brain tumors, non-invasive diagnosis would represent a substantial clinical advance, potentially sparing patients from the risks associated with surgical intervention on the brain. Such an approach will depend upon highly accurate models built using the limited datasets that are available. Herein, we present a novel genetic algorithm (GA) that identifies optimal architecture parameters using feature embeddings from state-of-the-art image classification networks to identify the pediatric brain tumor, adamantinomatous craniopharyngioma (ACP). We optimized classification models for preoperative Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and combined CT and MRI datasets with demonstrated test accuracies of 85.3%, 83.3%, and 87.8%, respectively. Notably, our GA improved baseline model performance by up to 38%. This work advances DL and its applications within healthcare by identifying optimized networks in small-scale data contexts. The proposed system is easily implementable and scalable for non-invasive computer-aided diagnosis, even for uncommon diseases.

BACKGROUND Randomized controlled trials (RCTs) in pediatric venous thromboembolism (VTE) treatment have been challenged by unsubstantiated design assumptions and/or poor accrual. Pilot/feasibility (P/F) studies are critical to future RCT success. METHODS Kids-DOTT is a multicenter RCT investigating non-inferiority of a 6-week (shortened) vs. 3-month (conventional) duration of anticoagulation in patients <21 years old with provoked venous thrombosis. Primary efficacy and safety endpoints are symptomatic recurrent VTE at 1 year and anticoagulant-related, clinically-relevant bleeding. In the P/F phase, 100 participants were enrolled in an open, blinded endpoint, parallel-cohort RCT design. RESULTS No eligibility violations or randomization errors occurred. Of enrolled patients, 69% were randomized, 3% missed the randomization window, and 28% were followed in pre-specified observational cohorts for completely occlusive thrombosis or persistent antiphospholipid antibodies. Retention at 1 year was 82%. Inter-observer agreement between local vs. blinded central determination of venous occlusion by imaging at 6 weeks post-diagnosis was strong (κ-statistic=0.75; 95% confidence interval [CI] 0.48–1.0). Primary efficacy and safety event rates were 3.3% (95% CI 0.3–11.5%) and 1.4% (0.03–7.4%). CONCLUSIONS The P/F phase of Kids-DOTT has demonstrated validity of vascular imaging findings of occlusion as a randomization criterion, and defined randomization, retention, and endpoint rates to inform the fully-powered RCT.

No abstract

Our case series demonstrates the potential benefit of VWI in children with stroke because it may identify patients who will have progressive arterial disease.