Background/Objectives: Hip displacement is a common problem in children with cerebral palsy (CP). Typically, the recommended hip surveillance imaging for these children consists of an anteroposterior pelvic radiograph, from which we calculate the migration percentage (MP) to determine treatment plans (conservative/preventive therapy, femoral osteotomy, femoral and pelvic osteotomies, and salvage surgery). However, little is known about the accuracy of MP for treatment planning. We aim to compare treatment plans based on MP thresholds with plans determined by an orthopedic surgeon following review of the hip CTs. Methods: We retrospectively identified hip CTs performed in children who were ≤18 years old with CP (11/2018-07/2024). The inclusion criteria were: (1) a pelvic radiograph performed 6 months prior to the hip CT; and (2) no surgeries between the pelvic radiograph and the hip CT. These hip CTs were randomized and blindly reviewed by an orthopedic surgeon to determine each child's treatment plan (CT-treatment). Separately, a pediatric radiologist blindly reviewed the randomized pelvic radiographs and measured each hip's MP to determine each child's treatment plan (XR-treatment). We used kappa-agreement and Bland-Altman analyses to compare XR- and CT-treatments. Results: Our study cohort consisted of 139 children (mean age = 9.3 ± 3.8 years; male = 90) with 278 hips. The proportion of agreement and unweighted kappa between XR- and CT-treatment were both low: 0.532 (148/278) and 0.339, respectively. Bland-Altman analyses showed that XR-treatment and CT-treatment were exchangeable when MP ≤ 10% but were not exchangeable otherwise. Conclusions: We should be cautious about relying exclusively on pelvic radiographs and subsequent MP calculation in making treatment decisions for hip displacement in children with CP since many anatomic details become evident on 3D imaging.

Background:Developmental dysplasia of the hip (DDH) is one of the most common pediatric musculoskeletal conditions, with an incidence of 1 in 1,000 live births. The acetabular index (AI), measured on anteroposterior (AP) radiographs, is the primary radiographic tool for DDH diagnosis and monitoring. However, its accuracy is critically dependent on pelvic positioning; suboptimal studies frequently require repeat imaging, resulting in additional radiation exposure for the child. Objective:To determine whether frog-leg lateral (FL) radiographs are interchangeable with AP radiographs for AI measurement in children being evaluated for suspected DDH. Methods:In this retrospective, IRB-approved study, bilateral AP and FL hip radiographs from 100 pediatric patients (aged 6-24 months) were evaluated at a single tertiary pediatric center. Only AP studies meeting strict pelvic positioning criteria (rotation index 0.5-2.0; tilt index 0.9-1.4) were included. Four readers, including three pediatric radiologists and a pediatric orthopedic surgeon, independently measured bilateral AI on each view on two separate sessions, with a 4-week washout between sessions. Interchangeability was assessed using the Individual Equivalence Index (IEI) with a 3° margin. Reliability was assessed with intraclass correlation coefficients (ICC). Results:The √IEI was 2.07°, with a one-sided 95% upper confidence bound of 2.34°, both below the 3° interchangeability margin. Intra-reader ICCs ranged from 0.845 to 0.898 across both views. Three of four readers showed slightly higher repeatability on the FL view. Inter-reader ICC was 0.783 (95% CI: 0.720-0.833) for AP and 0.807 (95% CI: 0.744-0.854) for FL. Conclusions:AI measurements from FL radiographs are interchangeable with those from AP radiographs. When the AP view is suboptimally positioned, the FL view may be used for AI measurement, potentially eliminating repeat imaging and reducing radiation exposure in this pediatric population.

Pediatric imaging presents distinct and urgent sustainability challenges, in part driven by its unique subspecialty demands: safeguarding the lifetime radiation risks of children, providing accurate diagnoses during their dynamic periods of growth, and ensuring family-centered care. These unique challenges impose additional strains on our ecosystem. To help alleviate this added burden, we propose a three-pillar model of sustainability specific to pediatric imaging, encompassing environmental, economic, and social factors. In particular, we address the sustainability challenges central to pediatric radiology by introducing AI not only as a tool for diagnostic accuracy, but also as an engine for sustainable practice. In this review, we move beyond the generic discussions of “green” radiology by illustrating how AI can be deployed to confront specific challenges across all three pillars of sustainability. Our review is centered around nine concrete, clinically grounded AI solutions, with three examples dedicated to each pillar. When strategically applied, these AI solutions have the potential to optimize energy efficiency, decrease consumables, extend equipment lifecycles, streamline operations, increase revenue, enhance transparency, improve pediatric care, promote equity, and empower patients and families. We also address other critical considerations in this sustainability domain, including AI’s own carbon footprint and the need for pediatric-specific validation. Collectively, AI’s extensive capabilities can drive our pediatric imaging towards diagnostic excellence, while optimizing environmental health, operational efficiency, and social equity.

Artificial intelligence (AI) applications have transformed radiology, yet pediatric medical imaging remains substantially underrepresented in AI development, validation, regulation, and implementation. Unlike adults, children go through continuous physiologic and anatomic changes that require age-specific models trained on representative developmental data. However, pediatric AI is limited by scarce publicly available datasets, fragmented institutional data, rare diseases, heterogeneous reporting practices, and insufficient external validation. Ethical and regulatory challenges are also a concern in children, including consent for secondary data use, off-label use of adult-trained AI models, and the need for postdeployment surveillance. Additionally, reimbursement is misaligned and must be optimized to allow innovation. This AJR Expert Panel Narrative Review examines the current landscape of pediatric AI in radiology and proposes practical priorities to support its safe and equitable adoption. The panel gives key recommendations, emphasizing the importance of an implementation roadmap to establish a dedicated pediatric AI infrastructure and standards that are essential to ensure diagnostic accuracy, workflow efficiency, and optimal clinical outcomes for children while minimizing bias and protecting patient safety.

Introduction Fractures are common injuries in abused children, second only to cutaneous bruising (1). Although fundamental to the documentation of abuse, the fractures are rarely life-threatening and few result in long-term deformity. Specific types of fractures are known to be associated with abuse, and their recognition is important for their accurate identification and in understanding their significance. Many reports of unexplained subdural hematomas (SDHs) in infants had appeared before Caffey’s historic 1946 article, but it was only after he associated these lesions with certain patterns of skeletal injury that the modern medical entity of child abuse was formulated (2). In a sense, recognition of the role of skeletal injuries in child abuse became the catalyst for the surge of interest in child maltreatment after Caffey’s original description. In the years that followed, most reports of child abuse focused mainly on the radiologic alterations associated with the skeletal trauma (3–14). The confident documentation of skeletal injury, facilitated by characteristic radiologic alterations, provided investigators the opportunity to study the multiple facets of child abuse. Eventually, the blend of the clinical and the radiologic findings led Kempe, Silverman, and others to bring these associations to the status of the “battered child syndrome” (15).

Most artificial intelligence (AI) research in radiology has focused on adults. Understanding macro-level trends in pediatric radiology AI can help guide, streamline, and bolster future research. To detail the current landscape of published AI research in pediatric radiology, filling a key research gap, as most radiology AI research has focused on adults. We conducted a scoping review, with a comprehensive literature search of Medline, Embase, Web of Science, and Cochrane Library from 2005 to 2024. Literature included for review were (1) original articles, (2) investigations that focused on pediatric populations (<18 years of age), and (3) articles with direct applications to clinical radiology and AI. We extracted each article’s study information, clinical application of focus, imaging modality, and the use of AI. We used descriptive frequencies to analyze summary statistics, and Chi-square testing to determine differences between categories. In total, we found 4,376 articles and included 789 articles in the review. The top three countries most active in scholarship related to AI in pediatric radiology were China (220, 27.9

Radiography is routinely used to evaluate the pediatric and young adult pelvis and hips. Unfortunately, the 2D nature of this imaging modality is insufficient in accurately depicting and evaluating complex 3D anatomical structures. In contrast, computed tomography (CT) provides exquisite details of 3D anatomy, typically at the expense of a higher radiation dose. Recent studies have suggested that ultra-low-dose CT (ULDCT) with tin filtration may overcome this diagnostic imaging dilemma by offering high-quality CT images with reduced radiation exposure. To compare patient-specific radiation exposure of diagnostic-quality hip ULDCTs and pelvic radiographs and thereby validate an optimized clinical protocol for hip ULDCT imaging in pediatric and young adult patients. We retrospectively searched the image archive at our large tertiary children’s hospital for hip CTs and anteroposterior (AP) pelvic radiographs performed within 6 months of each other (Dec 2023 - May 2024). The inclusion criteria were (1) hip CTs performed in accordance with our established ULDCT imaging protocol and (2) AP pelvic radiographs acquired in accordance with the American College of Radiology (ACR) guidelines. To calculate the effective doses of the pelvic radiographs and hip CTs, we used the National Cancer Institute dosimetry system for Radiography and Fluoroscopy (NCIRF) and Computed Tomography (NCICT), respectively. A paired two-tailed t-test was used to compare the effective doses of the hip CTs and AP pelvic radiographs. The study cohort included 29 patients (9 males, 20 females), stratified into the pediatric group (<18 years, n=17), young adult group (18-30 years, n=12), and entire cohort, with mean ages of 10.7 (SD, 6.0), 22.3 (SD, 3.7), and 15.5 (SD, 6.9) years, respectively. The average effective doses from ULDCT were 0.33 mSv (pediatric), 0.23 mSv (young adult), and 0.29 mSv (entire cohort), not significantly different from AP pelvic radiograph doses of 0.26, 0.29, and 0.27 mSv, respectively. In contrast, cumulative radiographic doses were significantly higher at 0.73 mSv, 0.76 mSv, and 0.74 mSv. ULDCT is a clinically feasible approach for pediatric and young adult hip imaging, offering diagnostic-quality CT images with substantially reduced radiation exposure (at a radiation dose level comparable to that of a single AP pelvic radiograph).

BackgroundRadiographic skeletal survey plays an important role in the diagnosis of infant abuse. Some practitioners have expressed concerns about the radiation exposure from this examination.ObjectiveTo utilize state-of-the-art hybrid computational phantoms to more accurately estimate radiation doses of skeletal surveys performed for suspected infant abuse.Materials and methodsWe searched our imaging database to identify skeletal surveys performed for suspected infant abuse (5/2020-5/2022). Initial skeletal surveys consisted of 25 standardized radiographs while follow-up skeletal surveys consisted of 16 standardized radiographs (no frontal or lateral views of the skull; or lateral views of the spine, knees, and ankles). To estimate the patient-specific organ and effective dose, we applied the National Cancer Institute dosimetry system for Radiography and Fluoroscopy (with on-the-fly 3D Monte Carlo simulation) to the male and female infant hybrid computational phantoms.ResultsThe mean total effective radiation dose was 0.627 mSv (initial survey) and 0.495 mSv (follow-up survey). For both surveys, the anteroposterior chest radiograph was the largest contributor to effective dose (contributing 0.101 mSv and 0.108 mSv, respectively). In the initial skeletal survey, the lens and the eyeballs received the highest organ absorbed doses (with the skull radiographs as the major contributors); and in the follow-up skeletal survey, the breasts received the highest organ absorbed dose (with the chest radiographs as the major contributors).ConclusionsWe employed hybrid computational phantoms to better estimate the radiation profile of skeletal surveys performed for suspected infant abuse, thus enabling us to update and optimize this life-saving imaging protocol.

Pediatric rotator cuff (RTC) injuries are uncommon, yet supraspinatus tendon (SST) signal alterations on T2-weighted imaging are frequently observed. To compare rates of SST signal alterations on shoulder MRIs of adolescents who are considered low- and high-risk for RTC injury. We retrospectively reviewed non-arthrogram shoulder MRI reports in 12-17-year-old patients at a large tertiary children’s hospital (01/2010—09/2024). We identified a low-risk patient cohort who lacked (a) clinical concern for RTC pathology, (b) athletic history associated with RTC injuries, (c) recent trauma, or (d) prior shoulder intervention. We also identified an age- and sex-matched high-risk patient cohort who had clinical concern for RTC pathology. Two experienced pediatric radiologists independently and blindly reviewed the shoulder MRIs in a random order from these cohorts. SST was evaluated using coronal oblique fat-suppressed T2-weighted sequences. Logistic regression models were developed to investigate differences between cohorts. Both low- and high-risk cohorts included 26 patients (14 males). Their median (inter-quartile range) ages were 14.0 (2.0) years and 14.5 (3.0) years, respectively. In the low-risk cohort, SST signal alterations were identified in 23 (88.5

The spatiotemporal changes of a developing anatomical structure is a dynamic process, and quantifying this process within a population and between populations is a fundamental yet challenging task in medical image analysis. Central to this task is the availability of longitudinal imaging data for 4D statistical shape analysis. Unfortunately, this type of longitudinal data is expensive, time-consuming, and difficult to collect. Practically, the majority of imaging data are 3D cross-sectional data, which are inadequate in describing the dynamic shape changes of anatomical structures. In this paper, we introduce a novel temporal atlas-guided deep learning model for longitudinal data generation. Unlike existing methods that directly generate longitudinal data from input images or sequences, we characterize distinctive geometric shape representations in both cross-sectional and longitudinal latent spaces of diffeomorphisms, while optimizing the quality of both atlas and longitudinal data generation. To the best of our knowledge, this is the first deep learning approach that leverages temporal atlas-based representation for longitudinal data generation. The innovative nature of our framework lies in its ability to jointly perform within-age and cross-age shape registration, thus maximizing registration performance while maintaining desirable deformation qualities. Our work's ability to model spatiotemporal dynamics makes it highly versatile and applicable to a wide range of domains, including modeling the normal and abnormal development of anatomical structures for improved clinical diagnosis and treatment planning. The code of this work is available at https:// github.com/wushaoju/TAG-GLE.

To deploy an AI model to measure limb-length discrepancy (LLD) and prospectively validate its performance. We encoded the inference of an LLD AI model into a docker container, incorporated it into a computational platform for clinical deployment, and conducted two prospective validation studies: a shadow trial (07/2024–9/2024) and a clinical trial (11/2024–01/2025). During each trial period, we queried for LLD EOS scanograms to serve as inputs to our model. For the shadow trial, we hid the AI-annotated outputs from the radiologists, and for the clinical trial, we displayed the AI-annotated output to the radiologists at the time of study interpretation. Afterward, we collected the bilateral femoral and tibial lengths from the radiology reports and compared them against those generated by the AI model. We used median absolute difference (MAD) and interquartile range (IQR) as summary statistics to assess the performance of our model. Our shadow trial consisted of 84 EOS scanograms from 84 children, with 168 femoral and tibial lengths. The MAD (IQR) of the femoral and tibial lengths were 0.2 cm (0.3 cm) and 0.2 cm (0.3 cm), respectively. Our clinical trial consisted of 114 EOS scanograms from 114 children, with 228 femoral and tibial lengths. The MAD (IQR) of the femoral and tibial lengths were 0.3 cm (0.4 cm) and 0.2 cm (0.3 cm), respectively. We successfully employed a computational platform for seamless integration and deployment of an LLD AI model into our clinical workflow, and prospectively validated its performance. Question No AI models have been clinically deployed for limb-length discrepancy (LLD) assessment in children, and the prospective validation of these models is unknown. Findings We deployed an LLD AI model using a homegrown platform, with prospective trials showing a median absolute difference of 0.2–0.3 cm in estimating bone lengths. Clinical relevance An LLD AI model with performance comparable to that of radiologists can serve as a secondary reader in increasing the confidence and accuracy of LLD measurements.

To compare the accuracy and reliability of 2D and 3D methods for measuring femoral version against an anatomic reference standard using 3D-printed femoral phantoms. CT data from three skeletally mature pediatric patients (2 females, 1 male; age 17.8 ± 0.1 years) were used as digital templates for 3D-printed haptic femur models. The anatomic reference standard for femoral version was determined by an established drilling technique. Models underwent CT scanning in neutral position and variations of flexion, internal/external rotation, and abduction/adduction (in increments of 15°, 30°, 45°). Two radiologists measured femoral version on axial datasets using 2D techniques (Murphy, Lee, Reikerås) and a 3D-reconstructed model in a blinded fashion. Mean absolute error (MAE) between measurements and the reference standard and mean absolute differences (MAD) between readers were calculated. Across all positions in all femurs, MAE was 22.9° ± 23.7° for the Murphy technique, 12.1° ± 8.5° for the Lee technique, 12.2° ± 9.6° for the Reikerås technique, and 2.3° ± 1.3° for the 3D method. For all 2D methods, MAE was greatest with adduction, flexion, and combinations of both. For the 3D method, femur position had no impact on MAE. MAD between readers was 9.7° for the Murphy technique, 4.7° for the Lee technique, 4.4° for the Reikerås technique, and 2.9° for the 3D method. Femoral version measurements based on 3D reconstructions are more accurate than traditional 2D techniques, more robust to femur positioning, and more consistent between readers. Accurate measurement of femoral version is important for understanding and treating lower extremity deformities in the skeletally mature pediatric population.

Urinary tract dilation (UTD) is a frequent problem in infants. Automated and objective classification of UTD from renal ultrasounds would streamline their interpretations. To develop and evaluate the performance of different deep learning models in predicting UTD classifications from renal ultrasound images. We searched our image archive to identify renal ultrasounds performed in infants ≤ 3-months-old for the clinical indications of prenatal UTD and urinary tract infection (9/2023—8/2024). An expert pediatric uroradiologist provided the ground truth UTD labels for representative sagittal sonographic renal images. Three different deep learning models trained with cross-entropy loss were adapted with four-fold cross-validation experiments to determine the overall performance. Our curated database included 492 right and 487 left renal ultrasounds (mean age ± standard deviation = 1.2 ± 0.1 months for both cohorts, with 341 boys/151 girls and 339 boys/148 girls, respectively). The model prediction accuracies for the right and left kidneys were 88.7

PurposeTo improve the quality of abdominal diffusion-weighted MR images (DW-MRI) when acquired using single-repetition (NEX = 1) protocols, and thereby increase apparent diffusion coefficient (ADC) map accuracy and lesion conspicuity at high b-values. We aim to reduce the effect of blurring due to motion that obscures small lesions when averaging multiple repetition images at each b-value, which is the current clinical standard.MethodsWe propose a self-supervised denoising diffusion probabilistic model (ssDDPM) to improve DW-MRI quality given noisy single-repetition acquisitions in pediatric abdominal scans. The ssDDPM is designed for multi-b-value DW-MRI and incorporates diffusion signal decay model (i.e., ADC model) constraints into its loss term. The model is trained to denoise single-repetition images from multiple b-values while ensuring that the output adheres to the signal decay model. Training was performed on a dataset of 120 pediatric subjects with liver tumors. The performance of ssDDPM was compared with non-local means (NLM) filtering and deep image prior (DIP) denoising techniques. These techniques have the capability to denoise single repetition images unlike the other techniques in literature that requires multiple direction or repetition images. Evaluation included qualitative radiologist's image quality assessment, receiver operating characteristic (ROC) analysis for lesion detection, and ADC fitting accuracy compared with motion-free, breath-hold reference data.ResultsThe ssDDPM demonstrated superior performance over comparison methods in terms of image quality, lesion conspicuity, and ADC map accuracy in NEX = 1 images. It received higher scores in radiologist assessments and showed better lesion discrimination in ROC analysis. Additionally, ssDDPM provided more precise and accurate ADC estimates when compared with the motion-free, breath-hold reference data.ConclusionThe ssDDPM effectively reduces motion related deblurring and enhances the quality of DW-MRI images by directly denoising single-repetition (NEX = 1) images while respecting signal decay model constraints. This method improves the assessment of pediatric liver lesions, offering a more accurate and efficient diagnostic tool with reduced scan times, when compared with current clinical practice and other denoising techniques.

BACKGROUND. Fracture dating is important in suspected infant abuse. Birth-related clavicle fractures are common and may provide surrogates to aid long bone fracture dating in infants. OBJECTIVE. The purpose of this study was to assess the impact of a template-matching clavicle fracture timeline atlas on radiologists' performance in dating birth-related fractures of the clavicle, humerus, and femur in young infants. METHODS. This retrospective study included infants 90 days old or younger who underwent radiography of a birth-related clavicle fracture from April 1, 2021, to July 31, 2024 or a birth-related fracture of the humerus or femur from December 1, 2011, to July 31, 2024. All eligible radiographs of each fracture were identified, representing distinct observations for the purposes of analysis. Patient age (expressed as days) at the time of radiograph acquisition served as the reference standard for fracture ages. A nonrigid image registration technique was applied to a nonoverlapping preassembled database of radiographs of birth-related clavicle fracture, to create a fracture dating atlas. Six readers (three trainees and three pediatric radiologists) independently reviewed the radiographs in separate sessions without and with use of the atlas to estimate fracture ages. Interreader agreement was assessed using intraclass correlation coefficients (ICCs). Fracture aging performance was assessed using mean absolute errors (MAEs). RESULTS. The analysis included 145 infants (87 male and 58 female infants) with 269 fracture radiographs (104 of the clavicle, 128 of the humerus, and 37 of the femur). The mean fracture age was 26 ± 19 [SD], 22 ± 14, and 21 ± 13 days for clavicle, humerus, and femur fractures, respectively. Interreader agreement for estimating fracture ages improved from moderate (ICC = 0.69) without use of the atlas to excellent (ICC = 0.91) with use of the atlas. The MAE in fracture dating was significantly lower (p < .05) with than without use of the atlas for all six readers for clavicle fractures (range, 4.8-5.5 vs 5.8-10.1 days), for all six readers for humeral fractures (range, 6.0-12.1 vs 3.0-3.8 days), and for five of six readers for femur fractures (range, 7.4-17.2 vs 3.3-4.8 days). MAE without and with use of the atlas was 8.8 versus 4.3 days, respectively, across trainee readers and 8.4 versus 4.0 days, respectively, across attending physician readers. CONCLUSION. The fracture dating atlas yielded significant improvements in radiologists' performance for dating infant clavicle, humerus, and femur fractures. CLINICAL IMPACT. Clavicle fracture healing patterns can serve as surrogates for dating long bone fractures commonly encountered in infant abuse.

Osteoid osteoma (OO) is the third most prevalent benign bone neoplasm in children. Although it predominantly affects the diaphysis of long bones, OO can assume an intra-articular location in the epiphysis or the intracapsular portions of bones. The most common location of intra-articular OO is the hip joint. The presentation of intra-articular OOs often poses a diagnostic enigma, both from clinical and radiologic perspectives. Initial symptoms are often vague and nonspecific, characterized by joint pain, stiffness, and limited range of motion, which frequently contributes to a delayed diagnosis. Radiographic findings range from normal to a subtle sclerotic focus, which may or may not have a lucent nidus. In contrast to their extra-articular counterparts, intra-articular lesions have distinct features at MRI, including synovitis, joint effusion, and bone marrow edema-like signal intensity. While CT remains the standard for identifying the nidus, even CT may be inadequate in visualizing it in some cases, necessitating the use of bone scintigraphy or fluorine 18-labeled sodium fluoride PET/CT for definitive diagnosis. Radiologists frequently play a pivotal role in suggesting this diagnosis. However, familiarity with the unique imaging attributes of intra-articular OO is key to this endeavor. Awareness of these distinctive imaging findings of intra-articular OO is crucial for avoiding diagnostic delay, ensuring timely intervention, and preventing unnecessary procedures or surgeries resulting from a misdiagnosis. The authors highlight and illustrate the different manifestations of intra-articular OO as compared with the more common extra-articular lesions with respect to clinical presentation and imaging findings. ©RSNA, 2024 Supplemental material is available for this article.

Objectives The T1-weighted GRE (gradient recalled echo) sequence with the Dixon technique for water/fat separation is an essential component of abdominal MRI (magnetic resonance imaging), useful in detecting tumors and characterizing hemorrhage/fat content. Unfortunately, the current implementation of this sequence suffers from several problems: (1) low resolution to maintain high pixel bandwidth and minimize chemical shift; (2) image blurring due to respiratory motion; (3) water/fat swapping due to the natural ambiguity between fat and water peaks; and (4) off-resonance fat blurring due to the multipeak nature of the fat spectrum. The goal of this study was to evaluate the image quality of water/fat separation using a high-resolution 3-point Dixon golden angle radial acquisition with retrospective motion compensation and multipeak fat modeling in children undergoing abdominal MRI. Materials and Methods Twenty-two pediatric patients (4.2 ± 2.3 years) underwent abdominal MRI on a 3 T scanner with routine abdominal protocol and with a 3-point Dixon radial-VIBE (volumetric interpolated breath-hold examination) sequence. Field maps were calculated using 3D graph-cut optimization followed by fat and water calculation from k-space data by iteratively solving an optimization problem. A 6-peak fat model was used to model chemical shifts in k-space. Residual respiratory motion was corrected through soft-gating by weighting each projection based on the estimated respiratory motion from the center of the k-space. Reconstructed images were reviewed by 3 pediatric radiologists on a PACS (picture archiving and communication systems) workstation. Subjective image quality and water/fat swapping artifact were scored by each pediatric radiologist using a 5-point Likert scale. The VoL (variance of Laplacian) of the reconstructed images was used to objectively quantify image sharpness. Results Based on the overall Likert scores, the images generated using the described method were significantly superior to those reconstructed by the conventional 2-point Dixon technique (P < 0.05). Water/fat swapping artifact was observed in 14 of 22 patients using 2-point Dixon, and this artifact was not present when using the proposed method. Image sharpness was significantly improved using the proposed framework. Conclusions In smaller patients, a high-quality water/fat separation with sharp visualization of fine details is critical for diagnostic accuracy. High-resolution golden angle radial-VIBE 3-point Dixon acquisition with 6-peak fat model and soft-gated motion correction offers improved image quality at the expense of an additional ~1-minute acquisition time. Thus, this technique offers the potential to replace the conventional 2-point Dixon technique.

Purpose:The limited volume of medical training data remains one of the leading challenges for machine learning for diagnostic applications. Object detectors that identify and localize pathologies require training with a large volume of labeled images, which are often expensive and time-consuming to curate. To reduce this challenge, we present a method to support distant supervision of object detectors through generation of synthetic pathology-present labeled images. Approach:Our method employs the previously proposed cyclic generative adversarial network (cycleGAN) with two key innovations: (1) use of "near-pair" pathology-present regions and pathology-absent regions from similar locations in the same subject for training and (2) the addition of a realism metric (Fréchet inception distance) to the generator loss term. We trained and tested this method with 2800 fracture-present and 2800 fracture-absent image patches from 704 unique pediatric chest radiographs. The trained model was then used to generate synthetic pathology-present images with exact knowledge of location (labels) of the pathology. These synthetic images provided an augmented training set for an object detector. Results:In an observer study, four pediatric radiologists used a five-point Likert scale indicating the likelihood of a real fracture (1 = definitely not a fracture and 5 = definitely a fracture) to grade a set of real fracture-absent, real fracture-present, and synthetic fracture-present images. The real fracture-absent images scored 1.7±1.0, real fracture-present images 4.1±1.2, and synthetic fracture-present images 2.5±1.2. An object detector model (YOLOv5) trained on a mix of 500 real and 500 synthetic radiographs performed with a recall of 0.57±0.05 and an F2 score of 0.59±0.05. In comparison, when trained on only 500 real radiographs, the recall and F2 score were 0.49±0.06 and 0.53±0.06, respectively. Conclusions:Our proposed method generates visually realistic pathology and that provided improved object detector performance for the task of rib fracture detection.

To evaluate the diagnostic performance and image quality of accelerated Turbo Spin Echo sequences using deep-learning (DL) reconstructions compared to conventional sequences in knee and ankle MRIs of children and young adults. IRB-approved prospective study consisting of 49 MRIs from 48 subjects (10 males, mean age 16.4 years, range 7–29 years), with each MRI consisting of both conventional and DL sequences. Sequences were evaluated blindly to determine predictive values, sensitivity, and specificity of DL sequences using conventional sequences and knee arthroscopy (if available) as references. Physeal patency and appearance were evaluated. Qualitative parameters were compared. Presence of undesired image alterations was assessed. The prevalence of abnormal findings in the knees and ankles were 11.7