Pediatric Radiology最新文献

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.

This narrative review maps the current landscape of artificial intelligence (AI) in paediatric and fetal neuroradiology, critically evaluating current practice, barriers to clinical adoption, and future potential. We searched for peer-reviewed studies from the last decade, focusing on image segmentation, lesion detection, classification, prognostication, and clinical decision support in paediatric brain imaging. Particular consideration was given to unique paediatric factors such as brain development and data scarcity. AI techniques, notably deep learning, have demonstrated success in automated brain tumour segmentation, detection of epileptogenic lesions, and radiomics-based classifiers predicting tumour histology and molecular subtypes. Despite these advancements, clinical adoption remains limited. Key barriers identified include high implementation costs, limited large-scale diverse paediatric datasets, and concerns regarding safety, bias, and regulatory approval. Addressing these issues through data-sharing initiatives, federated learning, paediatric-specific validation, and revised ethical and regulatory frameworks is crucial. Ongoing multi-institutional collaborations can facilitate AI's integration into paediatric neuroradiology, complementing radiologists and improving paediatric care.

Background: The integration of deep learning in medical imaging has reached proficiency levels akin to expert clinicians, particularly in tasks requiring precise image categorization.

Objective: This study developed KD-SqueezeNet, a lightweight deep learning model, to classify neonatal lung diseases via chest radiographs, aiming to enhance diagnostic accuracy and efficiency.

Materials and methods: Retrospective analysis included 2,089 neonates with clinical and imaging records. Chest radiographs were categorized into five groups: bronchopulmonary dysplasia (group 1, n=205), pneumonia (group 2, n=505), pneumothorax (group 3, n=201), respiratory distress syndrome (group 4, n=629), and normal (group 5, n=549). Data were divided into training, testing, and validation sets with an 8:1:1 ratio. Performance metrics included accuracy (Ac), precision (Pr), recall (Rc), F1 score (F1), parameter count, and area under the receiver operating characteristic curve (AUROC).

Results: KD-SqueezeNet, an interpretable model integrating knowledge distillation, outperformed EfficientNet, GhostNet, InceptionNet, RegNet, and Vision Transformer. For binary classification (healthy/diseased), it achieved Ac=0.93, Pr=0.93, Rc=0.93, F1=0.93, and AUROC=0.97 with only 723,522 parameters. In four-class classification (group 1/group 2/group 3/group 4), it attained Ac=0.86, Pr=0.86, Rc=0.86, and F1=0.86 (724,548 parameters), with class-specific AUROCs: group 1 (0.97), group 2 (0.94), group 3 (0.96), group 4 (0.97).

Conclusion: KD-SqueezeNet not only excels in accuracy and stability but also demonstrates efficient utilization of computational resources, making it suitable for rapid diagnosis and deployment in practical applications. It holds significant clinical value in terms of saving time and server space, providing auxiliary diagnostics, supporting clinical decision-making, and improving patient outcomes in large-scale screening contexts.

Background: Pediatric tuberculosis diagnosis relies heavily on imaging, yet access, equipment standards, and dose monitoring differ widely across health systems. Evidence describing how these contextual factors influence imaging use and radiation exposure in children remains scarce.

Objective: To describe pediatric tuberculosis imaging practices and estimated radiation doses across two distinct resource settings, Spain (hospital-based, high-resource) and Mozambique (primary care-based, low-resource), to inform strategies for safe, equitable, and context-appropriate imaging.

Methods and materials: A descriptive mixed-methods study combined retrospective data of children (<16 years) diagnosed with tuberculosis (Spain 2015-2021; Mozambique 2018-2021) with complementary surveys of imaging providers. In Spain, chest X-ray and computed tomography parameters were extracted from digital imaging and communications in medicine files to estimate organ-specific doses using the National Cancer Institute dosimetry systems for radiography and computed tomography. In Mozambique, dose estimates were based on standardized pediatric protocols and site survey parameters due to limited digital data. Surveys captured information on imaging access, guideline use, and professional training.

Results: Imaging data were available for 84 Spanish and 83 Mozambican children. In Spain, children underwent multiple chest X-rays (mean four per child) and computed tomographies (mean three per child), resulting in cumulative lung doses up to ~20 mGy cm², remaining below diagnostic reference levels. In Mozambique, most children had one or two chest X-rays, with cumulative lung doses <0.05 mGy cm². Survey findings indicated structured dose optimization and quality assurance practices in Spain, versus limited equipment and predominantly non-physician interpretation in Mozambique.

Conclusion: Context-appropriate improvements in pediatric imaging such as strengthened infrastructure, training, dose monitoring, and quality assurance are essential to ensure safe exposure and equitable, reliable tuberculosis diagnosis for children.

Background: White matter injury is a leading cause of neurodevelopmental impairment in premature infants. Timely initiation of novel therapies in clinical development will require early identification of clinically significant white matter injury. Head ultrasound is commonly obtained at 7 days and 30 days of life (DOL) to screen for brain injuries in premature infants.

Objective: To determine if white matter injury severity on head ultrasound at 7 DOL and 30 DOL is associated with white matter injury severity on term-equivalent age MRI and with neurological outcomes through age 2 years.

Materials and methods: We identified subjects via a search of the electronic health record for preterm infants born at ≤32 weeks gestational age (GA) with evidence of white matter injury in neuroimaging reports. Head ultrasounds at 7 days and 30 days and term-equivalent age MRIs were scored using established scoring systems by three expert readers, with final scoring established by consensus. We used ordinal logistic regression to determine the association between white matter severity on ultrasound and MRI. Multivariable models were adjusted for GA at birth and severity of intraventricular hemorrhage. Neurological outcomes (cerebral palsy, epilepsy, and neurosensory impairment) were determined by medical records review with a median corrected age at follow-up of 23.0 months.

Results: Fifty infants with a median GA at birth of 27.1 weeks were included in our retrospective cohort. White matter injury severity on 7-DOL (odds ratio 1.8, 95% CI 1.3-2.6) and 30-DOL (odds ratio 1.5, 95% CI 1.2-2.0) ultrasound was independently associated with severity on MRI. Higher injury severity on 7-DOL ultrasound was associated with cerebral palsy (odds ratio 2.4, 95% CI 1.3-4.3), while higher injury severity on 30-DOL ultrasound was associated with both cerebral palsy (odds ratio 1.7, 95% CI 1.2-2.5) and neurosensory impairment (odds ratio 1.7, 95% CI 1.2-2.4).

Conclusion: Preterm infants with white matter injury on 7-DOL or 30-DOL head ultrasound are at elevated risk for white matter injury on term-equivalent age brain MRI and for future neurological impairment.

Background: Ovarian-Adnexal Reporting & Data System Ultrasound (O-RADS US) is a validated scoring system in adult women with adnexal lesions to help assess the risk of potential malignancy. Limited data exists for children in whom malignancy is rare.

Objective: To evaluate inter-radiologist agreement and diagnostic performance when using the O-RADS US in pediatric patients with ovarian lesions.

Materials and methods: Retrospective IRB-approved study included pelvic ultrasounds (US) from 2015 to 2020 in pediatric patients (<18 years). Pelvic US with ovarian lesions measuring >3 cm in premenarchal patients and >5 cm in menarchal patients were included. Three pediatric radiologists reviewed each US and recorded imaging characteristics and O-RADS classification. Diagnostic performance was assessed, and agreement among radiologists was calculated.

Results: In total, 160 pelvic US exams were included in 160 patients, with a mean patient age of 12.1 years (SD=4.9). Most lesions were classified as O-RADS 2 (almost certainly benign), and fewer cases as O-RADS 4 (intermediate risk) or O-RADS 5 (high risk). Inter-radiologist agreement for O-RADS category was moderate (κ=0.42). Diagnostic performance of US O-RADS demonstrated high sensitivity and NPV (100% for all three reviewers). Specificities were 74-82%, and PPV was low at 5-7% for distinguishing malignant/borderline lesions from benign lesions.

Conclusions: Application of the O-RADS US system in pediatric patients may be challenging due to the low overall malignancy rate. Nevertheless, an O-RADS 2 classification provides meaningful reassurance, reflecting minimal malignancy risk in children. Larger studies are needed to determine the clinical utility of O-RADS US and whether pediatric-specific modifications are required.

Artificial intelligence (AI) has the potential to disrupt many fields, and radiology is no exception. The applications of AI in this field go beyond automated diagnosis since they can be used in any stage of the radiological pipeline, from patient referral to image interpretation and recommended course of action. However, it is important to distinguish between clinical usefulness and overpromises. This distinction is especially important for pediatrics, which presents additional challenges like the ethical considerations of working with children, the smaller dataset available for training, and a general lack of explicit labeling that indicates if a tool is suitable for pediatric populations. Here, we give pediatric radiologists a non-technical overview of AI and its subfields, and the potential benefits that it brings to radiology, so they are better equipped to critically evaluate AI and its clinical value. Far from replacing radiologists, AI should be viewed as a companion tool aimed at reducing inefficiencies, enhancing accuracy, and improving patient-centered care.

Background: Accurate measurement of cerebellar volume is crucial for diagnosing cerebellar atrophy or hypoplasia in infants. Although deep learning has achieved some success in cerebellar segmentation, existing studies primarily focus on adults. Challenges remain in applying these techniques to infants, especially those under 2 years old, due to the small cerebellar size and low tissue contrast in magnetic resonance imaging (MRI). As a result, developing segmentation models to identify abnormal cerebellar volumes in infants remains an unmet need.

Objective: To develop and validate a deep learning model for cerebellar volume measurement in infant brain MRI across different ages, with a focus on detecting cerebellar hypoplasia or atrophy.

Materials and methods: A deep learning segmentation model was developed using a publicly available dataset of 558 neonatal MRI scans. The model was validated on two independent datasets: a normal set (492 scans of typical infant brains) and an abnormal set (40 scans of cerebellar hypoplasia or atrophy). Two radiologists manually refined the segmented cerebellar regions to ensure consistency, followed by quantification of cerebellar volumes and diameters. The model's segmentation accuracy was assessed using the Dice similarity coefficient (DSC). A linear regression model was developed using the normal set to predict subjects' ages based on actual age, sex, and cerebellar volume. The difference between the predicted and actual ages (Δage) was calculated and used to classify subjects as normal or abnormal. The diagnostic performance of Δage in distinguishing between the normal and abnormal sets was assessed using the area under the receiver operating characteristic (ROC) curve (AUC).

Results: The mean DSC for cerebellar segmentation was 0.962 in the normal set and 0.882 in the abnormal set. In the normal set, cerebellar diameters were as follows: the left-right diameter ranged from 44.688 mm to 102.266 mm, the anterior-posterior diameter from 32.656 mm to 69.180 mm, and the superior-inferior diameter from 24.000 mm to 61.000 mm. The average cerebellar volume in the normal set was 79.305 cm³, showing rapid growth from birth to 10.0 months of age, with a slower growth rate from 10.0 months to 24.0 months. The AUC for Δage in identifying subjects with cerebellar hypoplasia or atrophy was 0.851.

Conclusion: The proposed deep learning model can accurately segment the cerebellum in infant brain MRI, quantify cerebellar volumes, and assist in identifying cerebellar hypoplasia or atrophy.