Faculty Bibliography

Background Previous studies suggest that use of artificial intelligence (AI) algorithms as diagnostic aids may improve the quality of skeletal age assessment, though these studies lack evidence from clinical practice. Purpose To compare the accuracy and interpretation time of skeletal age assessment on hand radiograph examinations with and without the use of an AI algorithm as a diagnostic aid. Materials and Methods In this prospective randomized controlled trial, the accuracy of skeletal age assessment on hand radiograph examinations was performed with (n = 792) and without (n = 739) the AI algorithm as a diagnostic aid. For examinations with the AI algorithm, the radiologist was shown the AI interpretation as part of their routine clinical work and was permitted to accept or modify it. Hand radiographs were interpreted by 93 radiologists from six centers. The primary efficacy outcome was the mean absolute difference between the skeletal age dictated into the radiologists' signed report and the average interpretation of a panel of four radiologists not using a diagnostic aid. The secondary outcome was the interpretation time. A linear mixed-effects regression model with random center- and radiologist-level effects was used to compare the two experimental groups. Results Overall mean absolute difference was lower when radiologists used the AI algorithm compared with when they did not (5.36 months vs 5.95 months; P = .04). The proportions at which the absolute difference exceeded 12 months (9.3% vs 13.0%, P = .02) and 24 months (0.5% vs 1.8%, P = .02) were lower with the AI algorithm than without it. Median radiologist interpretation time was lower with the AI algorithm than without it (102 seconds vs 142 seconds, P = .001). Conclusion Use of an artificial intelligence algorithm improved skeletal age assessment accuracy and reduced interpretation times for radiologists, although differences were observed between centers. Clinical trial registration no. NCT03530098 © RSNA, 2021 Online supplemental material is available for this article. See also the editorial by Rubin in this issue.

Background: Recently there has been a shift in radiology away from a peer review model toward a peer learning model, focusing more on collaborative learning, creating an environment more accepting of medical errors and embracing learning opportunities. As stated in the 2015 Institute of Medicine report, organizations that embrace error as learning opportunities outperform those that do not.
Purpose(s): To create an e-Peer Learning group to increase collaborative sharing of learning opportunities across institutions and assess the utility of the program among participants.
Material(s) and Method(s): The e-Peer Learning group consists of radiologists from 6 different pediatric radiology institutions. The representative members have exchanged short presentations of 1-3 learning cases monthly since 11/2017, including missed, difficult, classic, or unusual diagnoses. The format is of the case and imaging, followed by a few important learning points. Cases are then shared more widely amongst all the radiologists at the participating institutions. We recently distributed a survey to participants for feedback about the program.
Result(s): 60 radiologists participated in the survey, representing each participating institution. Participants were asked a few questions on a scale of 1-5 (1 highest; 5 lowest). Regarding the educational value of the cases, 40 participants (67.8%) answered the highest educational value of 1, and another 13 (22%) gave a value of 2. Regarding howmuch new information was learned, 34 participants (56.67%) gave a rating of 1 (learned a lot) while another 18 (30%) gave a value of 2. 29 participants (48.33%) said the cases have changed their practice. Overall, 58/60 (96.67%) stated that they wish to continue receiving cases.
Conclusion(s): Our e-Peer Learning group has successfully created a non-punitive, collaborative learning environment across multiple institutions. Our survey has shown that participants value the program and have learned new information that may potentially change clinical practice. We believe this model can be expanded or adapted to other groups

PURPOSE/OBJECTIVE:Familial dysautonomia (FD) is a rare autosomal recessive disease that affects the development of sensory and autonomic neurons, including those in the cranial nerves. We aimed to determine whether conventional brain magnetic resonance imaging (MRI) could detect morphologic changes in the trigeminal nerves of these patients. METHODS:Cross-sectional analysis of brain MRI of patients with genetically confirmed FD and age- and sex-matched controls. High-resolution 3D gradient-echo T1-weighted sequences were used to obtain measurements of the cisternal segment of the trigeminal nerves. Measurements were obtained using a two-reader consensus. RESULTS:in controls (P < 0.001). No association between trigeminal nerve area and age was found in patients or controls. CONCLUSIONS:Using conventional MRI, the caliber of the trigeminal nerves was significantly reduced bilaterally in patients with FD compared to controls, a finding that appears to be highly characteristic of this disorder. The lack of correlation between age and trigeminal nerve size supports arrested neuronal development rather than progressive atrophy.

Purpose or Case Report: Skeletal bone age assessment is a common clinical practice to investigate endocrinology, genetic and growth disorders of children. Clinical interpretation and bone age analyses are time-consuming, labor intensive and often subject to inter-observer variability. Bone age prediction models developed with deep learning methodologies can be leveraged to automate bone age interpretation and reporting. The bone age model developed at our institution was offered to interested health systems and institutions to implement and validate the model. This study discusses the logistical, technical, and clinical issues encountered with this model implementation. Methods & Materials: After IRB approval, multiple U.S. based radiology departments were solicited to adopt and validate the Stanford University bone age model. A total of 8 institutions (4 standalone pediatric hospitals and 4 academic radiology departments) agreed to partner with the primary investigators. IRBs at each institution were required in addition to registration with ClinicalTrials.gov registry. Standardization of the data use agreements was performed. Patient data and protected health information data was retained at each institution. Technical requirements included model hosting at each institution and integration to send images to the model server and results to the interpreting radiologists.
Result(s): Multiple logistical, technical, and clinical issues were encountered. IRBs at the various institutions had different requirements including waiving patient consent. Technical differences between institutions included model hosting, PACS integrations, interfaces with the reporting system, and image preprocessing. Clinical differences included report templates, calculation of bone age standard deviation, use of Brush foundation, and ability to directly send bone predictions to the reporting system (versus displaying the results as a separate interface). The bone age model was successfully implemented at 7 institutions and approximately 190 studies have been evaluated.
Conclusion(s): There are myriad challenges to implementing and validating models developed with deep learning methodologies. As models are developed for various clinical use cases including bone age assessment, it will be incumbent on radiology practices and health information systems to integrate these models into clinical practice

Congenital limb length discrepancy disorders are frequently associated with a variety of vascular anomalies and have unique genetic and phenotypic features. Many of these syndromes have been linked to sporadic somatic mosaicism involving mutations of the phosphoinositide 3-kinase (PI3K)/protein kinase B (AKT)/mammalian target of rapamycin (mTOR) pathway, which has an important role in tissue growth and angiogenesis. Radiologists who are aware of congenital limb length discrepancies can make specific diagnoses based on imaging findings. Although genetic confirmation is necessary for a definitive diagnosis, the radiologist serves as a central figure in the identification and treatment of these disorders. The clinical presentations, diagnostic and imaging workups, and treatment options available for patients with Klippel-Trenaunay syndrome, CLOVES (congenital lipomatous overgrowth, vascular anomalies, epidermal nevi, and scoliosis/spinal deformities) syndrome, fibroadipose vascular anomaly, phosphatase and tensin homolog mutation spectrum, Parkes-Weber syndrome, and Proteus syndrome are reviewed. ^©RSNA, 2019.

Tuberous sclerosis (TS) is an autosomal dominant condition associated with mutations in the TSC1 and/or TSC2 genes. Clinical manifestations are multisystemic, and they often include lesions in the brain, skin, heart, kidneys, and bones. TSC2 gene mutations can be seen concomitantly with autosomal dominant polycystic kidney disease gene mutations. We present a case of a fetus with prenatal diagnosis of TS that had unique asymmetrical distribution of renal cystic disease. We describe the extensive work up with both fetal and neonatal magnetic resonance imaging with correlating images of the unilateral polycystic renal disease in addition to typical TS brain findings.

BACKGROUND: Rapid and accurate diagnosis of appendicitis, particularly with respect to the presence or absence of perforation, is essential in guiding appropriate management. Although many studies have explored sonographic findings associated with acute appendicitis, few investigations discuss specific signs that can reliably differentiate perforated appendicitis from acute appendicitis prior to abscess formation. OBJECTIVE: The purpose of our study was to identify sonographic findings that improve the specificity of US in the diagnosis of perforated appendicitis. Our assessment of hepatic periportal echogenicity, detailed analysis of intraperitoneal fluid, and formulation of select constellations of sonographic findings expands upon the literature addressing this important diagnostic challenge. MATERIALS AND METHODS: We retrospectively reviewed 116 abdominal US examinations for evaluation of abdominal pain in children ages 2 to 18 years from January 2008 to September 2011 at a university hospital pediatric radiology department. The study group consisted of surgical and pathology proven acute appendicitis (n = 51) and perforated appendicitis (n = 22) US exams. US exams without a sonographic diagnosis of appendicitis (n = 43) confirmed by follow-up verbal communication were included in the study population as the control group. After de-identification, the US exams were independently reviewed on a PACS workstation by four pediatric radiologists blinded to diagnosis and all clinical information. We recorded the presence of normal or abnormal appendix, appendicolith, appendiceal wall vascularity, thick-walled bowel, dilated bowel, right lower quadrant (RLQ) echogenic fat, increased hepatic periportal echogenicity, bladder debris and abscess or loculated fluid. We also recorded the characteristics of intraperitoneal fluid, indicating the relative quantity (number of abdominal regions) and quality of the fluid (simple fluid or complex fluid). We used logistic regression for correlated data to evaluate the association of diagnosis with the presence versus absence of each US finding. We conducted multivariable analysis to identify constellations of sonographic findings that were predictive of perforated appendicitis. RESULTS: The individual US findings of abscess/loculated fluid, appendicolith, dilated bowel and increased hepatic periportal echogenicity were significantly associated with perforated appendicitis when compared with acute appendicitis (P < 0.01). The sonographic observation of increased hepatic periportal echogenicity demonstrated a statistically significant association with perforated appendicitis compared with acute appendicitis (P < 0.01). The presence of complex fluid yielded a specificity of 87.7% for perforated appendicitis compared with the acute appendicitis group. The US findings of >/=2 regions or >/=3 regions with fluid had specificity of 87.3% and 99.0%, respectively, for perforated appendicitis compared with the acute appendicitis group. Select combinations of sonographic findings yielded high specificity in the diagnosis of perforated appendicitis compared with acute appendicitis. These constellations yielded higher specificity than that of each individual finding in isolation. The constellation of dilated bowel, RLQ echogenic fat, and complex fluid had the highest specificity (99.5%) for perforated appendicitis (P < 0.01). CONCLUSION: Our study demonstrates that identification of select constellations of findings using abdominal sonography, in addition to focused US examination of the right lower quadrant, can improve sonographic diagnosis of perforated appendicitis in the pediatric population.