Искусственный интеллект в педиатрической нейрорадиологии: текущий ландшафт, проблемы и будущие направления.

Introduction and background

Paediatric and fetal neuroradiology encompasses imaging of the developing brain and nervous system, presenting distinct challenges from adult neuroradiology due to rapid brain maturation, evolving myelination, and age-specific conditions. Differentiating normal developmental variations from pathology, such as delayed myelination vs. leukodystrophies, on magnetic resonance imaging (MRI), can be particularly challenging. Young children often require sedation for diagnostic quality MRI, while computed tomography (CT) necessitates strict adherence to As Low As Reasonably Achievable (ALARA) principles to minimise radiation exposure. Additionally, congenital anomalies (including cortical malformations) and rare genetic syndromes (including leukodystrophies) require specialised expertise.

These unique aspects underpin both the rationale and challenges for artificial intelligence (AI) in paediatric neuroradiology. Modern multi-parametric imaging techniques (e.g. diffusion and perfusion MRI) provide extensive data but risk overwhelming radiologists [ 1 – 4 ]. AI algorithms, including machine learning (ML) and deep learning, excel at detecting complex patterns in high dimensional datasets, for example, models trained on multi-parametric paediatric brain MRI can integrate features across several sequences to segment brain tumours or highlight subtle focal cortical dysplasia that may be difficult to appreciate on routine visual review. AI could synthesise large imaging and clinical datasets, supporting diagnosis and prognostication where traditional interpretation is challenged [ 1 , 5 , 6 ]. Importantly, AI may address longstanding paediatric issues by facilitating shorter MRI protocols or correcting motion artefacts to avoid sedation and optimising CT radiation doses intelligently [ 7 – 11 ]. AI can further support radiologists by identifying subtle, age dependent imaging features in childhood tumours or epilepsy, correlating imaging with clinical and genomic data for personalised medicine [ 1 , 12 ]. Due to the rarity of many paediatric neurological disorders and limited datasets, a comprehensive evaluation of AI research is essential [ 13 ]. Herein, we present a narrative summary of a 10-year scoping review of the AI in paediatric neuroradiology literature. This review highlights AI’s transformative potential in paediatric neuroradiology, leveraging its strengths in data integration, pattern recognition, and automation to enhance diagnostic accuracy and patient safety while also considering challenges for discovery research and barriers to implementation adoption.

Information sources and search strategy

We conducted a structured narrative review to map the landscape of AI applications in paediatric and fetal neuroradiology. The search covered the period 1 January 2015 to 31 December 2025, with a final search performed on 27 January 2026 to ensure completeness.

Searches were performed across PubMed, Google Scholar, medRxiv, arXiv, and major imaging journals and society platforms. Preprints were considered eligible when explicitly labelled. The search combined seeded queries, iterative query expansion, and snowballing (both backward and forward citation chasing).

An example Google Scholar query was:

(“paediatric” OR “pediatric”) AND (neuroradiology OR “brain MRI” OR “head CT”) AND (“artificial intelligence” OR “machine learning” OR “deep learning”)

For PubMed, we focused on titles and abstracts, incorporating MeSH terms where available:

((“paediatric”[tiab] OR “pediatric”[tiab]) AND (neuroradiology[tiab] OR brain[tiab]) AND (“artificial intelligence”[tiab] OR “deep learning”[tiab] OR “machine learning”[tiab]))

Citation snowballing was repeated for all included studies to identify additional relevant work. Studies were included if they met the following criteria: Population: paediatric cohorts (0–17 years) or mixed populations with extractable paediatric data Domain: neuroradiology imaging tasks Methods: applications of AI, including deep learning and machine learning approaches Study types: original studies, reviews, technical guidance, or consensus documents relevant to clinical practice We did not explicitly exclude studies reporting non-imaging AI applications or opinion pieces but prioritised original research in our synthesis.

Title and abstract screening was conducted by a single reviewer, followed by full-text assessment. Findings were synthesised narratively, structured by modality and task. We placed particular emphasis on dataset provenance, external validation, clinical integration, and readiness for implementation. While preprints were included to capture emerging developments, conclusions relied preferentially on peer-reviewed evidence.

We aligned the review process with best-practice recommendations for narrative reviews, following the scale for the assessment of narrative review articles (SANRA) framework to ensure clarity, transparency, and scientific rigour. Specifically, we addressed SANRA’s [ 14 ] six items: importance, aims, literature search, referencing, scientific reasoning, and presentation of endpoint data prospectively from the outset of our search by design.

Current state of artificial intelligence in paediatric neuroradiology

AI research in paediatric neuroradiology has expanded, although it remains less developed than adult imaging. Studies typically focus on tasks such as image segmentation, lesion detection, classification, and outcome prediction, centred around specific paediatric neurological conditions. Table 1 summarises representative AI applications and outcomes in paediatric neuroradiology. Table 1 Representative artificial intelligence applications in paediatric neuroradiology Study (year) Focus (disease/task) AI method Key finding/outcome Limitations Jeong et al. (2022) [ 15 ] Paediatric epilepsy (MRI negative cases) Multi-modal deep CNN Detected previously occult epileptogenic foci in ~15% of drug-resistant epilepsy patients with initially normal MRI Retrospective and observational study. Small sample size ( n =41). Selection bias (specialised population) Quon et al. (2020) [ 16 ] Hydrocephalus (ventricle segmentation) 3-Dimensional (D) CNN segmentation Automated segmentation of cerebral ventricles for volume calculation, aiding paediatric hydrocephalus evaluation Retrospective and observational study. Volume only – does not classify hydrocephalus. Selection bias (normal only) Yeom et al. (2022) [ 17 ] Medulloblastoma (tumour subtyping) MRI radiomics+machine learning Noninvasively classified 4 molecular subgroups of medulloblastoma from MRI with high accuracy (radiogenomic profiling) Retrospective. Heterogeneous MRI scans were normalised. Only presurgical imaging without diffusion, fingerprinting, or perfusion. No spatial context Vecchiato et al. (2021) [ 18 ] Non-sedated paediatric MRI (motion artefact) Retrospective motion correction AI-driven k-space reordering (DISORDER) significantly improved MRI image quality and diagnostic readability in 75% of scans without sedation n =32. Longer scan times. Their methods increase motion artefact and before subsequent correction. Impact of reconstruction delay (up to 1 h) and a quantification of maximum tolerable degradation not assessed Chang et al. (2021) [ 19 ] Paediatric brain tumours (prognostication) Radiomic feature analysis Identified MRI texture features correlating with tumour molecular subtypes; certain features predicted poorer survival ( P <0.05) Retrospective. n =27. Limited accuracy. Limited images only used 3-D three-dimensional, AI artificial intelligence, CNN convolutional neural network, MRI magnetic resonance imaging, n sample size number

Early AI successes in paediatric neuroradiology have primarily involved segmentation and quantitative analysis [ 5 , 6 ]. Deep learning models have been used effectively to segment paediatric brain structures and lesions on MRI. For example, automatic ventricular segmentation in hydrocephalus enables precise volume measurement [ 9 , 16 ]. Quon et al. developed an AI tool segmenting cerebral ventricles on MRI/CT to track hydrocephalus progression objectively [ 16 ]. These techniques have also been combined with 3-dimensional printing (Fig. 1 ) [ 20 ]. Tumour segmentation, particularly using U-Net-based models, is another significant area, generating reproducible tumour volumes in paediatric gliomas and medulloblastomas [ 17 ]. The recent BraTS-PEDs challenge highlights international efforts to benchmark paediatric tumour segmentation using multi-centre datasets [ 21 ]. Fig. 1 A segmentation and 3-dimensional printing pipeline for ventriculomegaly via deep learning. From medical images, the lateral ventricles’ masks are obtained. Processed masks are transformed into physical models. Reproduced with permission under the open-access terms of the Nature Portfolio journals [ 20 ]

AI-driven lesion detection and classification methods are also emerging, especially in epilepsy and tumour characterisation. In paediatric epilepsy, subtle cortical malformations, often undetectable on standard MRI, are a significant challenge (Fig. 2 ) [ 22 ]. Jeong et al. reported a deep learning model using multi-modal MRI that localised epileptogenic zones in drug-resistant epilepsy, identifying lesions missed on initial MRIs in 15% of cases [ 15 ]. The Multi-Centre Epilepsy Lesion Detection (MELD) group and their recent extension using graph neural networks also continue to make significant contributions to the field [ 22 – 24 ]. Additionally, prototype tools have been developed for paediatric head CT, targeting automated detection of intracranial haemorrhage or skull fractures, though robust paediatric validation remains limited compared to adult studies [ 25 , 26 ]. Fig. 2 Example classifier predictions with saliency scores for “Patient 1” (an example with fluid attenuation inversion recovery (FLAIR) data) and “Patient 2” (without FLAIR data). a Classifier predictions ( dark red ) and manual lesion mask ( black line ) visualised on brain surfaces (only lesional hemisphere is shown). Classifier predictions ( dark red ) visualised on T1 volume. b Z -scored mean feature values within predicted lesions coloured with integrated gradients saliency scores. Positive saliency scores indicate feature values driving the classifier’s “lesion” prediction. Negative scores indicate feature values that are inconsistent with the prediction. c Lesional cortex highlighted on the patients’ scans exhibits salient features automatically identified by the classifier. Reproduced from [ 22 ] under the terms of the Creative Commons CC BY license

Radiomics (quantitative imaging feature extraction) methods have shown promise for classification and outcome prediction, especially when combined with machine learning. This has been used notably in paediatric neurooncology [ 1 , 8 , 12 , 17 ]. Radiomic features can be correlated with clinical or genomic data, enabling non-invasive tumour characterisation. Yeom et al. demonstrated that MRI-based ML models accurately differentiate molecular subgroups of medulloblastoma (Wingless/WNT, Sonic Hedgehog/SHH, Group 3, Group 4), with area under the curve values between 0.80–0.90 [ 17 ]. Chang et al. similarly reported MRI texture features differentiating medulloblastoma subgroups and predicting survival outcome [ 19 ]. Additionally, ML has been utilised to estimate gestational age or brain maturity from neonatal MRI and to predict neurodevelopmental outcomes [ 27 , 28 ]. These studies underscore AI’s potential to reveal subtle prognostic imaging biomarkers beyond visual assessment, as illustrated in Fig. 3 . Fig. 3 a Artificial intelligence (AI) augments the capabilities of neuro-oncologists/radiation-oncologists by enabling integrated diagnosis, offering deeper insights into the disease, facilitating precise prognosis by predicting outcomes, and assisting in patient stratification to tailor treatment plans to individual needs. b AI supports neuroradiologists by leveraging images for automated detection and tumour segmentation, identifying molecular subtypes of tumours, providing quantitative measurements, and delivering diagnostic assistance to distinguish tumours from necrotic regions, all while ensuring automated quality checks. c AI aids neurosurgeons during surgery, contributing to surgical margin assessment and offering real-time diagnosis information and guidance, enhancing surgical precision and patient outcomes. d AI assists neuropathologists in the analysis of fresh/frozen samples, providing automated measurement of features, aiding in tumour classification and grading, improving tumour detection, and delivering comprehensive analysis of cellular and tissue structures through histo-molecular classification. e Handling mutation data, single-cell information, methylation patterns, RNA sequencing, and more, AI empowers molecular pathologists by supporting biomarker identification, pathway identification, treatment response prediction, variant identification, and serving as a diagnosis assistant, streamlining the complex molecular analysis process (created with BioRender.com). Reproduced from [ 29 ] with permission under the open-access terms of the Nature Portfolio journals

Overall, existing literature demonstrates proof-of-concept successes in paediatric neuroradiology AI. However, studies typically involve small retrospective cohorts (often tens to low hundreds) [ 5 , 6 ] due to the rarity of paediatric conditions and barriers to multi-centre data sharing. Compared to adult neuroradiology, paediatric AI applications remain primarily in the research domain, with limited clinical deployment and widespread calls for further validation [ 1 , 11 , 13 ].

Current state of artificial intelligence in fetal neuroimaging

Fetal brain imaging provides crucial prenatal insights but is technically demanding due to fetal motion, evolving neuroanatomy, and variability in image quality. AI methods, particularly deep learning, are addressing these longstanding obstacles with the potential to transform prenatal diagnosis and management. Clinically relevant AI applications include detailed anatomical segmentation, robust motion correction, early anomaly detection, accurate gestational age estimation, and potential prediction of neurodevelopmental outcomes. Wu et al. [ 30 ] provide an example of an AI used for multi-planar reconstruction. Potential application of reconstruction techniques and their synergy with AI is shown in Fig. 4 [ 31 ]. These developments offer promising avenues for improved diagnostic precision and informed parental counselling in clinical fetal neuroradiology practice. Fig. 4 Appreciation of the quality of reconstruction depending on the weight ( W ) that controls the strength of the regularisation. The framework for optimising the reconstruction quality through parameter fine-tuning in the presence of motion is illustrated at two gestational ages: 26 weeks and 30 weeks. Two representative clinical cases are provided for comparison. The results for three values of W are presented. For W =0.1, the reconstruction looks blurry with poor tissue contrast. Using W=3 improves the contrast but the images look noisy. For W =0.75, the reconstruction is sharp, with a contrast between different brain tissues similar to that observed in the isotropic ground truth. Clinical cases from which the simulated images are derived highlight the accuracy of a reconstruction for this intermediate value of W , especially with regard to the definition of the corpus callosum and the delineation of the cortex. These reconstructions have been shown to aid artificial intelligence segmentation, a technologic synergy. Reproduced from Wu et al. with permission under the open-access terms of the Nature Portfolio journals [ 30 ]

Automated segmentation via convolutional neural networks (CNNs) has substantially enhanced fetal brain MRI quantification, providing volumetric metrics and morphometric analysis even in complex clinical scenarios. Models such as multi-scale CNNs and 3-dimensional U-Net architectures now achieve consistent segmentation accuracy (Dice≥0.90) in both normative and pathological conditions, including fetuses with congenital brain anomalies or ventriculomegaly [ 32 ]. The clinical impact of this is notable, allowing neuroradiologists to objectively monitor ventricular dilation or cortical abnormalities longitudinally, facilitating early intervention planning. However, substantial technical challenges persist, notably dataset scarcity, cross-institutional heterogeneity, and limited external validation, underscoring the importance of robust multi-centre collaborations for future clinical translation [ 33 ]. Complementary AI-driven motion correction techniques, using unsupervised learning frameworks and generative adversarial networks (GANs), now offer reliable retrospective mitigation of motion-induced image degradation. Such advancements can reduce repeat examinations and sedation requirements, potentially improving patient safety and clinical workflow [ 30 , 34 , 35 ]. Nevertheless, these methods currently remain primarily within research contexts and demand rigorous clinical validation and integration into diagnostic imaging protocols to establish real-world utility [ 30 , 35 ].

In fetal neurosonography, CNNs enhance screening by automatically verifying critical standard planes (transventricular, transcerebellar), promoting more consistent anatomical assessment. Furthermore, AI classifiers effectively distinguish normal from abnormal brain ultrasounds (accuracy ~96%), generating lesion-localising heatmaps that guide clinicians towards regions warranting detailed examination [ 36 , 37 ]. The clinical utility of such approaches is evident, potentially reducing inter-operator variability, minimising missed diagnoses of subtle abnormalities, and improving referral decisions. Similarly, AI-driven gestational age prediction from fetal MRI images, leveraging sophisticated attention-guided deep-learning networks, demonstrates accuracy within approximately 1 gestational week, outperforming traditional biometric methods particularly during advanced gestation [ 38 , 39 ]. In clinical practice, this precise developmental staging could identify subtle deviations from expected brain maturation, signalling neurodevelopmental vulnerability, particularly in fetuses exhibiting ventriculomegaly or other structural anomalies [ 40 ]. Preliminary investigations into using prenatal cortical expansion metrics from MRI for postnatal neurodevelopmental prognosis indicate a promising clinical horizon, although large-scale longitudinal validation is still required, especially in the context of the required follow-up duration (as neurodevelopmental outcomes often evolve beyond early childhood) [ 41 , 42 ]. Ultimately, however, these remain primarily research applications and transitioning such AI tools from research to routine clinical care demands systematic evaluation through prospective multi-centre studies, rigorous external validation, and explainable AI methodologies that ensure clinical acceptability, transparency, and trustworthiness.

Current clinical application

Many emerging AI applications in paediatric neuroradiology are already being prototyped or tested in clinical research settings. Table 2 in this review lists available AI devices relevant to paediatric neuroimaging, highlighting the limited number explicitly designed or validated for this population. Note that these devices are advertised as applicable to children or across all ages, even though some of the use cases do not appear to be directly relevant to children, e.g. dementia and coronary artery disease. Available data was taken in good faith, as it was presented on radiology.healthairegister.com. A recent federated learning model, FL-PedBrain, is an emerging example. It successfully classified and segmented paediatric brain tumours across 19 institutions, demonstrating performance comparable to traditional centralised training [ 44 ]. This real-world study illustrates how federated learning circumvents data-sharing barriers by “bringing AI to the data”. Similarly, an AI model analysing neonatal MRI accurately predicted developmental maturity and identified infants at risk for developmental delays, facilitating earlier interventions [ 28 ]. Table 2 Radiology AI Devices whose Intended use incudes both neuroradiology and either all ages or children. Reproduced from Kelly et al 2025 [ 43 ] Product name Company Modality Disease targeted Key-features Suggested use (before/during/after interpretation) AI-Rad Companion Brain MR Siemens Healthineers MR Neurodegenerative disease, Multiple Sclerosis Brain volume quantification, segmentation, normative comparison, white matter hyperintensities, report generation, longitudinal comparison During: perception aid (prompting all abnormalities/results/heatmaps) IB Delta Suite Imaging Biometrics MR Neuro lesions Image standardisation, Delta T1 maps JAD-02K JLK Inc. MR Dementia Cortical area segmentation, cortical thickness quantification, brain aging status, normal comparison During: perception aid (prompting all abnormalities/results/heatmaps), report suggestion JBA-01K JLK Inc. MR Cerebral aneurysm Cerebral aneurysm detection During: perception aid (prompting all abnormalities/results/heatmaps), report suggestion NeuroQuant Cortechs.ai MR Neurodegenerative brain disorders Brain region quantification (75 structures), brain change data, normative data comparisons, dynamic atlas, custom report options Before: flagging acute findingsDuring: perception aid (prompting all abnormalities/results/heatmaps), interactive decision support (shows abnormalities/results only on demand) NeuroreaderÆ Brainreader MR Neurodegenerative disorders Brain region quantification, comparison to a normative database Diagnostic aid Pixyl.Neuro.FL Pixyl MR Leukoencephalopathy, vascular dementia, small-vessel disease White matter hyperintensities detection, white matter hyperintensities volume quantification, longitudinal analysis During: perception aid (prompting all abnormalities/results/heatmaps) Pixyl.Neuro.MS Pixyl MR Multiple Sclerosis Lesion detection, lesion quantification, lesion classification, lesional volume quantification, disease activity, longitudinal analysis, image annotation, McDonald MRI criteria During: perception aid (prompting all abnormalities/results/heatmaps), report suggestion After: diagnosis verification CoLumbo Smart Soft Healthcare MR Disk herniation (Protrusion/Extrusion modifiers), Central Spinal Stenosis (CSS), Shizas and Lee morphological grading, decreased vertebral body heights, decreased disk height, hypo/hyperlordosis, nerve root impingement, listhesis Lumbar spine annotation, smart navigation, editable report, editable separate pathology report, measurement calculation based on segmentation During: perception aid (prompting all abnormalities/results/heatmaps), interactive decision support (shows abnormalities/results only on demand), report suggestion AIRAscore structure AIRAmed MR Neurodegenerative diseases, e.g. dementia, Alzheimer’s disease, multiple sclerosis, Parkinson’s Disease Brain tissue and anatomy segmentation, volume quantification, normative comparison, report generation, WMH detection and quantification During: perception aid (prompting all abnormalities/results/heatmaps), interactive decision support (shows abnormalities/results only on demand) Annalise Enterprise CTB annalise.ai CT 130 findings present in the emergent, urgent, and non-urgent care settings including: Intracranial haemorrhage, haematomas (acute subdural/extradural), fractures, pneumocephalus, infarct, mass effect & midline shift, obstructive hydrocephalus, Intraaxial mass, transependymal oedema, colloid cyst, vasogenic oedema, and many others. Detection of up to 130 findings (acute and chronic) on brain CT, worklist triage, notification, visualisation, confidence bar, normal/abnormal differentiation Before: adapting worklist order, flagging acute findings During: perception aid (prompting all abnormalities/results/heatmaps), interactive decision support (shows abnormalities/results only on demand) Brain aneurysm (BA) Aidoc CT Brain aneurysm Triage tool, workflow improvement Triage/prioritisation of medical images CINA-ASPECTS Avicenna.AI CT Stroke Automated ASPECT score, visual localisation information During: perception aid (prompting all abnormalities/results/heatmaps) e-ASPECTS Brainomix CT Stroke ASPECTS score, ischemic volume quantification, HDVS detection, hyperdensity detection and quantification, report generation icobrain ms icometrix MR Multiple sclerosis White matter abnormality quantification, brain volume change, report generation During: perception aid (prompting all abnormalities/results/heatmaps), report suggestion icobrain tbi icometrix CT, MR Traumatic brain injury Volume quantification, brain volume changes, report generation, cisternal compression, midline shift, mass lesions, traumatic axonal injuries During: perception aid (prompting all abnormalities/results/heatmaps), report suggestion Intracranial Hemorrhage (ICH) Aidoc CT Intracranial Hemorrhage Triage tool, workflow improvement Triage/prioritisation of medical images. Ischemic Stroke Aidoc CT Large vessel occlusion (M1, ICA), medium vessel occlusions (M2), posterior occlusions Triage tool, workflow improvement Triage/prioritisation of medical images. Neurocloud SISCOM Qubiotech MR, SPECT Epilepsy Automatic co-registration SPECT-MRI, interictal SPECT subtraction ictal image, identification of epileptogenic focus, percentage overlap of the focus During: perception aid (prompting all abnormalities/results/heatmaps), interactive decision support (shows abnormalities/results only on demand), report suggestion Neurophet AQUA Neurophet MR Mild cognitive impairment, Alzheimer's disease, dementia Brain region segmentation, volume quantification, normative comparison, report generation, white matter hyperintensity quantification, multi-time-point analysis. During: perception aid (prompting all abnormalities/results/heatmaps) qER Qure.ai CT Intracranial hemorrhage, infarct, cranial fracture, pneumocephalus, hydrocephalus, midline shift, cerebral atrophy Brain pathology detection, mobile notifications, worklist prioritisation, report generation, traumatic brain injury tracking Before: adapting worklist order, flagging acute findings During: perception aid (prompting all abnormalities/results/heatmaps), interactive decision support (shows abnormalities/results only on demand), report suggestion QyScore Qynapse MR, PET Dementia, Multiple sclerosis, Alzheimer's disease, neuro-degenerative diseases Brain region quantification, comparison to reference population, white matter hyperintensities segmentation and evolution in time and space, report generation During: perception aid (prompting all abnormalities/results/heatmaps), report suggestion After: diagnosis verification Rapid MRI RapidAI MR Stroke SyMRI NEURO SyntheticMR MR Multiple sclerosis, atrophy, dementia, neurogenerative diseases Contrast image synthesis, brain region segmentation, degree of myelination, TRACE4AD DeepTrace Technologies Srl MR Alzheimer’s disease dementia Mild cognitive impairment (MCI) subject risk score (low risk vs high risk) of being affected by or progressing to Alzheimer's disease dementia within 24 months, cognitive, functional and behavioral deficits of the subject, subject gray matter volume and images with normative comparison, best voxel-based map predictors overlayed on the subject gray matter, best neuropsychological predictors During: interactive decision support (shows abnormalities/results only on demand), report suggestion After: diagnosis verification Viz ANEURYSM Viz.ai CT Cerebral aneurysm Aneurysm detection, notification Viz ICH Viz.ai CT ICH SwiftMR AIRS Medical MR Not applicable MRI scan acceleration, MRI image enhancement Jazz AI Medical MR Multiple sclerosis (MS), brain metastasis Brain lesion detection, volume quantification During: interactive decision support (shows abnormalities/results only on demand), report suggestion VUNO Med-DeepBrain‚ VUNO MR Atrophy, Dementia Brain parcellation, Atrophy quantification During: perception aid (prompting all abnormalities/results/heatmaps), interactive decision support (shows abnormalities/results only on demand), report suggestion AI Artificial intelligence, ASPECTS Alberta Stroke Program Early CT Score, CSS central spinal stenosis, CT computed tomography, HDVS hyperdense vessel sign, ICA internal carotid artery, ICH intracranial haemorrhage, M1 first segment of the middle cerebral artery, M2 second segment of the middle cerebral artery, MCI mild cognitive impairment, MR magnetic resonance, MRI magnetic resonance imaging, MS multiple sclerosis, PET positron emission tomography, SISCOM subtraction ictal single-photon emission computed tomography co-registered to magnetic resonance imaging, SPECT single-photon emission computed tomography, TBI traumatic brain injury, WMH white matter hyperintensities

Acquisition-embedded AI is already shipped on scanners. Siemens Deep Resolve (including Boost/Sharp), General Electric AIR Recon DL, and Philips SmartSpeed are marketed to accelerate MRI and improve signal-to-noise ratio, with paediatric protocol packs and claims of faster examinations. Independent paediatric evidence is emerging but remains limited to small, mostly single-centre studies focused on image quality rather than patient outcomes such as sedation rates. Early paediatric neuroimaging evaluations report feasibility of deep-learning reconstruction for synthetic MRI and accelerated 3-D T1-weighted protocols, with improved image quality and reduced acquisition times; however, designs are largely observational and underpowered for clinical endpoints [ 43 , 45 , 46 ].

Clinically, AI-driven MRI reconstruction and motion correction tools are undergoing trials to assess their potential to reduce sedation requirements. Anecdotal evidence further supports AI’s clinical value; a recent report highlighted an AI algorithm detecting subtle cortical dysplasia previously missed by human interpretation, enabling successful epilepsy surgery [ 24 ]. As such examples move increasingly into clinical practice, supported by those in the adult neuroradiology and general paediatric populations, confidence in AI will continue to grow. Over the next decade, broader participation in AI registries and clinical trials will help generate the necessary evidence for regulatory approval and wider adoption.

Current challenges and practical hurdles

Despite promising research, significant barriers hinder clinical adoption of AI in paediatric neuroradiology. These challenges include (a) cost and access, (b) standardisation and interoperability, and (c) safety, governance, and ethics [ 47 – 49 ].

Cost and access

Implementing AI in radiology requires substantial investment in software, hardware (high-performance graphical processing units (GPUs) or servers), and integration into picture archive and communications systems (PACS)/radiology information systems (RIS) workflows [ 50 , 51 ]. Budget constraints in paediatric imaging departments can make acquiring advanced AI tools prohibitively expensive, particularly for smaller hospitals or low-resource settings. Available commercial AI tools often focus on high-volume adult conditions (stroke or chest imaging), limiting economic justification for smaller paediatric caseloads. Thus, disparities arise where major academic paediatric hospitals may trial AI tools, while regional centres struggle to afford them [ 52 – 55 ].

Internal development of bespoke paediatric AI models is equally resource intensive, demanding data scientists and computational infrastructure often unavailable in smaller centres. Demonstrating cost effectiveness is also challenging; administrators require robust evidence of efficiency improvements or clinical outcomes (e.g. reduced scanning times or misdiagnoses) to justify investment. Such evidence remains limited in paediatric neuroradiology. Moreover, AI models may require regular recalibration or retraining, incurring continuous costs, an issue that is complicated by the added variability of age-related brain development [ 56 ]. Without clear reimbursement pathways or external funding, economic barriers remain significant [ 47 – 49 ].

Standardisation and interoperability

A fundamental barrier is the paucity of large, standardised datasets and established validation frameworks. While many medical imaging datasets exist [ 57 ], access can be siloed and region dependent. AI performance heavily depends on large-scale data, yet paediatric neuroimaging datasets remain fragmented and scarce [ 13 , 44 ]. Contributing factors include lower patient volumes, rare conditions, stringent privacy regulations, and variability in imaging protocols. Consequently, AI models frequently train on single-centre datasets, limiting generalisability. The lack of standardised paediatric image repositories and benchmarks hinders objective algorithm comparison and development [ 5 , 9 ].

Efforts like the FL-PedBrain federated learning platform (aggregating data from 19 sites for paediatric brain tumours) [ 44 ] are emerging, although they remain preliminary, and there have been explicit calls for the inclusion of paediatric data for AI research [ 13 ] and recognition of the crucial role of this inclusion [ 58 , 59 ]. Interoperability also presents challenges: diverse scanner manufacturers, imaging protocols, and PACS systems complicate uniform AI deployment. AI tools developed for specific settings may underperform when transferred elsewhere unless robustly generalisable [ 5 , 16 , 50 , 60 , 61 ].

Additionally, paediatric-specific AI standards and guidelines are limited [ 62 ]. Existing validation and reporting frameworks (Food and Drug Association (FDA), Radiological Society of North America (RSNA)) do not explicitly address paediatric factors such as age stratification. Integration into existing radiology workflows further complicates matters. Radiologists expect AI results seamlessly integrated within PACS, yet achieving this requires standardised interfaces (e.g. DICOM structured reporting), which vary among vendors. Multiple AI tools might present outputs differently (visual highlights vs. textual reports), potentially fragmenting workflows. While standards such as Imaging Data Commons and Fast Healthcare Interoperability Resources are evolving, their adoption is not yet widespread. Comprehensive standardisation of datasets, evaluation metrics, and integration protocols remains essential for wider AI adoption in paediatric neuroradiology.

Safety, governance, and ethics

Ensuring AI is safe, unbiased, and ethically implemented for children remains paramount [ 59 , 63 – 65 ]. Algorithmic bias is a significant concern; models predominantly trained on adult data or limited paediatric populations may not generalise across diverse age groups, ethnicities, or developmental stages. This is especially of concern in mixed adult and paediatric centres where AI models (for example, intended for the triage of CT scans for intracranial bleeding in adults in the emergency department) could be accidentally used off-label in children, potentially leading to incorrect results that could falsely alarm or reassure the clinical teams [ 43 ].

Age-related anatomical variations (e.g. myelination, cranial morphology) differ significantly, increasing the risk of misclassification of normal developmental variations as pathology or vice versa [ 58 , 59 , 66 ]. Paediatric datasets often disproportionately represent specific demographics (tertiary referral populations, regional biases), amplifying these risks [ 11 ]. To mitigate healthcare disparities, rigorous validation across age ranges and diverse populations is essential.

Data privacy represents another challenge. Children are legally protected under stringent privacy frameworks (General Data Protection Regulation (GDPR), Health Insurance Portability and Accountability Act (HIPAA)), complicating large-scale data aggregation. Techniques like federated learning (where models train without raw data sharing) show promise but remain under refinement [ 44 , 51 , 55 ].

The complexity of these issues is compounded by the need for regulatory approval. Agencies (FDA, conformité européenne (CE) marking) have been cautious regarding paediatric-specific AI device authorisations; only approximately 3–4% of FDA-cleared radiology AI devices currently hold explicit paediatric indications [ 11 , 62 ]. Consequently, many tools applied clinically in paediatrics remain “off-label”, necessitating rigorous internal validation. Professional bodies like the American College of Radiology advocate explicit labelling of validated age groups for AI products [ 11 , 62 ].

Medico-legal accountability also warrants attention. Misdiagnoses or false positive findings from AI raise liability concerns. Current guidelines emphasise AI as decision support, maintaining ultimate accountability with the interpreting radiologist. Thus, AI systems require transparency and explainability. While many methods of explainability exist [ 67 ], many advanced deep learning models are inherently opaque (“black boxes”), complicating clinical and parental acceptance. To build trust, governance frameworks involving rigorous prospective validation, continuous post-deployment monitoring, and protocols for radiologist oversight and override are necessary. Ethically, involving families and communities in AI implementation discussions (including explicit consent processes and transparent communication about AI roles) is vital [ 59 , 66 , 68 – 71 ].

Overcoming bias with diverse data, safeguarding privacy, adhering to paediatric-specific regulatory standards, and maintaining human oversight are essential to ensure safe, equitable, and ethical AI deployment in paediatric neuroradiology.

Future potential and emerging applications

Several promising directions exist for AI to significantly impact paediatric neuroradiology. Emerging applications align with critical clinical needs, from safer imaging techniques to advanced diagnostic support. Here, we discuss future potential in key areas: (a) image optimisation and safety, (b) quantitative analysis and automated metrics, and (c) advanced clinical decision support. We also consider the (d) regulatory, reimbursement, and medico-legal pathways; (e) ethics, consent, and communication; and (f) workflow integration and clinician trust issues, which are central to adoption.

Image optimisation and safety

AI is expected to rapidly advance image acquisition techniques, improving patient safety through reduced contrast use and radiation exposure. Synthetic contrast-enhanced MRI exemplifies this potential [ 72 – 74 ]. AI models trained on paired non-contrast and contrast-enhanced MRI scans can predict virtual contrast images, minimising or eliminating gadolinium-based contrast agent risks [ 75 ]. Preliminary adult studies show promising results; extending this approach to paediatrics would significantly enhance safety and comfort during follow-up scans, particularly for oncological surveillance.

AI techniques also promise substantial reductions in radiation dose for paediatric CT and PET imaging. Deep learning–based denoising and super resolution methods enable diagnostic quality images from ultra-low dose protocols, adhering closely to ALARA principles. Initial adult implementations have demonstrated dose reductions exceeding 50%; similar or greater benefits are expected in paediatrics due to higher radiosensitivity [ 1 , 76 ].

Additionally, motion correction and accelerated imaging through AI could reduce sedation requirements in paediatric MRI. Prospective AI-driven motion tracking (navigator signals or real-time video tracking) and retrospective AI-based k-space corrections (e.g. DISORDER technique) [ 18 ] represent significant advances. AI-enhanced image reconstruction also markedly reduces scan duration, further reducing patient discomfort and the need for sedation. Real-time AI quality control could automatically flag and correct motion degraded sequences or dynamically adjust scanner parameters (“closed loop imaging”) [ 11 ]. These techniques together promise safer, faster, and higher quality paediatric neuroradiology exams.

Quantitative analysis and automated volumetry

Future AI systems will increasingly deliver quantitative metrics as standard outputs, moving radiology from qualitative to precise quantitative assessment. Automated volumetric analysis could replace subjective descriptions (e.g. “mild ventriculomegaly”) with precise measurements (ventricular volume, hippocampal size, lesion growth rates). For instance, in paediatric hydrocephalus, AI could accurately quantify ventricular changes, informing shunt management decisions [ 16 ]. This is especially relevant in the paediatric context considering the need for age- and sex-stratified normative atlases, as volumetric interpretation in children requires developmental context. Similarly, AI-derived tumour volumetry can improve reproducibility in monitoring growth or therapeutic response, while aiding neurosurgical planning through enhanced lesion delineation [ 1 , 77 , 78 ].

AI could also generate developmental metrics such as brain maturation scores by comparing individual imaging to normative paediatric atlases [ 28 ]. Such quantitative assessments may help in the early identification of developmental delays or disorders like autism, prompting timely intervention.

Radiomics-based predictive analytics represent another emerging application [ 8 , 17 , 78 ]. AI-generated radiomic signatures could predict tumour histology or genetic mutations non-invasively, informing clinical decisions and reducing invasive diagnostic procedures [ 17 , 79 – 81 ]. For instance, predictive AI models might estimate the likelihood of specific tumour subtypes (e.g. pilocytic astrocytoma vs. medulloblastoma), providing probabilistic guidance before surgical intervention.

Longitudinal analysis powered by AI will become commonplace, tracking changes in lesion burden, myelination, or subtle disease progression (e.g. neurofibromatosis, tuberous sclerosis), ensuring consistent disease surveillance and personalised patient management [ 11 , 78 ].

Advanced clinical decision support

AI’s ultimate aim in paediatric neuroradiology is sophisticated decision support, enhancing diagnostic accuracy and clinical management. Radiomics and radiogenomics are poised to become routine clinical tools. AI-driven models predicting tumour molecular markers (e.g. H3K27M mutation in diffuse midline gliomas) or prognostic factors could provide early guidance on therapy selection or trial enrolment, even before biopsy results [ 17 , 19 , 80 , 82 ]. Machine learning could also stratify risk or predict outcomes, such as post-surgical seizure freedom in epilepsy [ 15 , 24 ] or neurodevelopmental prognosis post hypoxic ischaemic injury [ 27 ], facilitating early targeted interventions.

Automated detection systems will mature, providing real-time second reader capabilities. AI trained on abusive head trauma imaging patterns, integrating CT and retinal findings, could improve timely recognition and diagnosis. Early prototypes already indicate feasibility in flagging high suspicion abuse cases for further review [ 26 , 83 ]. Similarly, AI may predict therapeutic responses in epilepsy through combined imaging and EEG analysis [ 84 , 85 ].

Workflow prioritisation through AI-driven triage is another promising area. Urgent findings (e.g. unexpected cerebral herniation or stroke) detected by AI could trigger immediate alerts, expediting clinical response. Natural language processing could further streamline reporting, drafting preliminary radiology reports for expert verification, thus enhancing efficiency [ 86 , 87 ].

Regulatory, reimbursement, and medico-legal aspects

Future systems must be designed as high-risk medical devices with clear intended use, age range, and paediatric performance claims, rather than extrapolated adult data, in line with the new multi-society implementation framework [ 56 ]. Transparent description of datasets, subgroup performance, and paediatric safety ratings will be a precondition for approval, reimbursement, and avoidance of off-label use. Studies should move beyond accuracy metrics to include prospective implementation and health-economic outcomes, and AI outputs and user interactions should be logged in an auditable way to support quality assurance, incident review, and medico-legal scrutiny [ 56 ].

Ethics, consent, and communication

The framework emphasises child-centred deployment, public and patient involvement, and AI literacy for all staff in paediatric imaging [ 56 ]. In paediatric neuroradiology, consent processes should explicitly state when and how AI is used, with information adapted to the developmental stage. Future work should include participatory studies with children, young people, and families on how to present data in ways that support shared decision-making without increasing anxiety [ 59 , 65 ].

Workflow integration and clinician trust

Adoption will depend on tools that integrate into PACS/RIS with defined workflow touchpoints (triage, protocol selection, quantitative add-ons, MDT visualisation), coupled with phased pilots and post-market surveillance [ 56 ]. Systems should expose uncertainty, provide simple interfaces, and allow radiologists to accept, modify, or reject outputs, with feedback loops and monitoring dashboards for ongoing performance and bias auditing. Targeted AI education for paediatric neuroradiologists, radiographers, and referrers will be essential to build appropriate trust and avoid over-reliance.

Crucially, these advances maintain radiologists’ central oversight role. A “human-centric AI” approach ensures radiologists validate AI outputs contextually, thereby augmenting rather than replacing expert clinical judgement [ 11 , 59 , 65 ]. AI thus expands paediatric neuroradiologists’ capabilities, enabling more comprehensive multi-disciplinary patient care and deeper diagnostic insight.

Discussion and limitations

There is great hope that AI in paediatric and fetal neuroradiology will transition from an emerging research topic towards clinical maturity and we have discussed several avenues and potential benefits in this review. However, significant efforts are still required to realise its full potential. Despite much promise, adoption is constrained by absent prospective trials and lack of clarity regarding paediatric indications for CE-marked or FDA-cleared tools, as shown in prior device surveys [ 43 , 88 ]. Off-label deployment remains a risk, with inconsistent labelling of target populations and suitability for children [ 43 ].

This review demonstrates that while numerous studies have established technical feasibility for AI applications, from tumour segmentation to lesion identification and outcome prediction, routine clinical adoption remains limited (Tables 1 and 2 ). An important consideration is that the old adage that children are not simply “small adults” is especially true in clinical AI development. Algorithms must be explicitly trained and validated on paediatric datasets; adult-trained AI models frequently underperform in paediatric populations due to distinct developmental and disease presentation differences; however, we acknowledge that work is still needed to establish the minimum dataset size/diversity needed for generalisability [ 11 , 56 ]. Encouragingly, paediatric-focused collaborations (e.g. data-sharing consortia, specialised conferences, American College of Radiology and the European Society of Paediatric Radiology Paediatric AI Workgroups) increasingly acknowledge this distinction, advocating for tailored solutions. Expanding curated and annotated paediatric datasets and leveraging federated learning or other privacy-preserving technologies are essential next steps to facilitate AI training without compromising privacy [ 13 , 44 ]. Over the next decade, integration into registries, prospective multi-centre evaluations, and harmonised labelling will be essential to move beyond anecdotal reports. This will include systematic assessment of vendor-provided reconstruction tools in reducing sedation requirements, federated and privacy-preserving training to overcome data scarcity, and benchmarking of paediatric-specific diagnostic AI against clinical outcomes. Without these steps, the gap between technical capability and safe, regulated clinical adoption will persist.

Addressing identified barriers (cost, standardisation, ethics) requires multi-faceted solutions. Sustainable economic models involving research grants, public-private partnerships, or demonstrated cost effectiveness must be developed. Standardisation initiatives should continue, incorporating paediatric-specific guidelines into validation and reporting frameworks, including explicit age-based labelling of AI products [ 62 ]. Regulatory bodies could introduce incentives analogous to paediatric pharmaceutical legislation, encouraging AI developers to validate and market paediatric indications explicitly.

Foe clinical adoption to be successful, usability and trust remain critical. Active involvement of paediatric neuroradiologists in AI design e promotes clinical relevance, smooth integration into workflows, and improved acceptance. User-friendly interfaces, intuitive visualisations (such as AI-annotated MRI overlays), and explainable AI outputs facilitate clinician adoption. Ongoing education and training in interpreting AI results and understanding limitations are equally important. Within multi-disciplinary paediatric teams, neuroradiologists increasingly serve as critical interpreters of AI-derived insights for clinical colleagues, emphasising the importance of maintaining rigorous standards and healthy scepticism. Discrepancies between AI and human assessments should be actively analysed as learning opportunities, either highlighting subtle findings overlooked by radiologists or exposing AI errors.

Our review has several limitations. The review was narrative, rather than systematic, which may introduce selection bias and limit reproducibility. Screening was conducted by a single reviewer, which may risk omissions or subjective inclusion bias. Inclusion of preprints not only increases coverage but also introduces the risk of citing non–peer-reviewed, unverified findings. Although our review maps a decade of research, the literature base in paediatric neuroradiology is still sparse and heterogeneous, making strong generalisations difficult. Included studies often have small sample sizes (tens to hundreds of cases), and reliance on proof-of-concept and single-centre studies limits the strength of clinical translation. Furthermore, it is possible that many failed experiments or attempts at implementation have not been published, adding the potential for publication bias to the issues above. There also remains a gap between research applications and clinical translation, with limited tools commercially available specifically for children. In this context, the lack of prospective, multi-centre validation remains a recurring issue for the field.

Conclusion

In this review, we have considered 10 years of paediatric and fetal neuroradiology AI literature. However, looking forward a decade, success could lead to AI systems routinely detecting subtle brain malformations, optimising imaging protocols to reduce radiation exposure, or predicting tumour biology preoperatively. Such advances would profoundly improve clinical care, facilitating precise, proactive, and personalised management of paediatric neurological disorders. Thoughtful integration of AI, with continued primacy of expert human judgement, promises substantial benefits, ensuring improved outcomes and experiences for children and families. While challenges exist, the potential reward of healthier children enabled by smarter imaging certainly justifies dedicated pursuit of this goal.

To this end, paediatric neuroradiology should align with the new multi-society implementation framework for paediatric radiology [ 56 ], which organises action around four pillars: regulation and purchasing, implementation and integration, interpretation and post-market surveillance, and education. Practically, we call for (i) multi-centre data-sharing initiatives with paediatric-specific safety ratings and transparent reporting of dataset composition and subgroup performance; (ii) prospective, workflow-embedded studies that include health-economic and patient-centred outcomes; and (iii) structured AI literacy and governance programmes, with the involvement of children, young people, and families, to support trustworthy use.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Author contribution

B.K. wrote the manuscript. S.C. conception, review, and editing (primary). K.M. conception, outline (primary), supervision, review. G.C. conception (primary), supervision, review, and editing. All authors reviewed and approved the final submitted version.

Funding

Open Access funding provided by the IReL Consortium. No authors have funding or conflicts of interest relevant to the current manuscript.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Conflicts of interest

The authors declare no competing interests.

References

1. Pringle C Kilday J-P Kamaly-Asl I Stivaros SM The role of artificial intelligence in paediatric neuroradiology Pediatr Radiol 2022 52 2159 2172 10.1007/s00247-022-05322-w 35347371 PMC9537195 Pringle C, Kilday J-P, Kamaly-Asl I, Stivaros SM (2022) The role of artificial intelligence in paediatric neuroradiology. Pediatr Radiol 52:2159–2172 35347371 10.1007/s00247-022-05322-w PMC9537195

2. Bailey CR Bailey AM McKenney AS Weiss CR Understanding and appreciating burnout in radiologists Radiographics 2022 42 E137 E139 10.1148/rg.220037 35839137 Bailey CR, Bailey AM, McKenney AS, Weiss CR (2022) Understanding and appreciating burnout in radiologists. Radiographics 42:E137–E139 35839137 10.1148/rg.220037

3. Bastian MB Fröhlich L Wessendorf J Prevalence of burnout among German radiologists: a call to action Eur Radiol 2024 34 5588 5594 10.1007/s00330-024-10627-5 38345608 PMC11364704 Bastian MB, Fröhlich L, Wessendorf J et al (2024) Prevalence of burnout among German radiologists: a call to action. Eur Radiol 34:5588–5594 38345608 10.1007/s00330-024-10627-5 PMC11364704

4. Fawzy NA Tahir MJ Saeed A Incidence and factors associated with burnout in radiologists: a systematic review Eur J Radiol Open 2023 11 100530 10.1016/j.ejro.2023.100530 37920681 PMC10618688 Fawzy NA, Tahir MJ, Saeed A et al (2023) Incidence and factors associated with burnout in radiologists: a systematic review. Eur J Radiol Open 11:100530 37920681 10.1016/j.ejro.2023.100530 PMC10618688

5. Kelly BS, Clifford SM, Judge CS, et al (2024) A survey of paediatric radiology artificial intelligence. medRxiv 2024.09.19.24313885. 10.1101/2024.09.19.24313885

6. Kelly BS Judge C Bollard SM Radiology artificial intelligence: a systematic review and evaluation of methods (RAISE) Eur Radiol 2022 10.1007/s00330-022-08784-6 35420305 PMC9668941 Kelly BS, Judge C, Bollard SM et al (2022) Radiology artificial intelligence: a systematic review and evaluation of methods (RAISE). Eur Radiol. 10.1007/s00330-022-08784-6 35420305 PMC9668941

7. Ryan M-L O’Donovan T McNulty JP Artificial intelligence: the opinions of radiographers and radiation therapists in Ireland Radiography 2021 27 S74 S82 10.1016/j.radi.2021.07.022 34454835 Ryan M-L, O’Donovan T, McNulty JP (2021) Artificial intelligence: the opinions of radiographers and radiation therapists in Ireland. Radiography 27:S74–S82 34454835 10.1016/j.radi.2021.07.022

8. da Rocha JLD Lai J Pandey P Artificial intelligence for neuroimaging in pediatric cancer Cancers (Basel) 2025 17 622 10.3390/cancers17040622 40507258 PMC12153639 da Rocha JLD, Lai J, Pandey P et al (2025) Artificial intelligence for neuroimaging in pediatric cancer. Cancers (Basel) 17:622 40507257 10.3390/cancers17111776 PMC12153587

9. Martin D Tong E Kelly B Current perspectives of artificial intelligence in pediatric neuroradiology: an overview Front Radiol 2021 1 713681 10.3389/fradi.2021.713681 37492174 PMC10365125 Martin D, Tong E, Kelly B et al (2021) Current perspectives of artificial intelligence in pediatric neuroradiology: an overview. Front Radiol 1:713681 37492174 10.3389/fradi.2021.713681 PMC10365125

10. Otjen JP Moore MM Romberg EK The current and future roles of artificial intelligence in pediatric radiology Pediatr Radiol 2022 52 2065 2073 10.1007/s00247-021-05086-9 34046708 Otjen JP, Moore MM, Romberg EK et al (2022) The current and future roles of artificial intelligence in pediatric radiology. Pediatr Radiol 52:2065–2073 34046708 10.1007/s00247-021-05086-9

11. Bhatia A Khalvati F Ertl-Wagner BB Artificial intelligence in the future landscape of pediatric neuroradiology: opportunities and challenges AJNR Am J Neuroradiol 2024 45 549 553 10.3174/ajnr.A8086 38176730 PMC11288527 Bhatia A, Khalvati F, Ertl-Wagner BB (2024) Artificial intelligence in the future landscape of pediatric neuroradiology: opportunities and challenges. AJNR Am J Neuroradiol 45:549–553 38176730 10.3174/ajnr.A8086 PMC11288527

12. Yu X Li S Mai W Pediatric diffuse intrinsic pontine glioma radiotherapy response prediction: MRI morphology and T2 intensity-based quantitative analyses Eur Radiol 2024 34 7962 7972 10.1007/s00330-024-10855-9 38907098 PMC11557687 Yu X, Li S, Mai W et al (2024) Pediatric diffuse intrinsic pontine glioma radiotherapy response prediction: MRI morphology and T2 intensity-based quantitative analyses. Eur Radiol 34:7962–7972 38907098 10.1007/s00330-024-10855-9 PMC11557687

13. Muralidharan V Burgart A Daneshjou R Rose S Recommendations for the use of pediatric data in artificial intelligence and machine learning ACCEPT-AI NPJ Digit Med 2023 6 166 10.1038/s41746-023-00898-5 37673925 PMC10482936 Muralidharan V, Burgart A, Daneshjou R, Rose S (2023) Recommendations for the use of pediatric data in artificial intelligence and machine learning ACCEPT-AI. NPJ Digit Med 6:166 37673925 10.1038/s41746-023-00898-5 PMC10482936

14. Baethge C Goldbeck-Wood S Mertens S SANRA—a scale for the quality assessment of narrative review articles Res Integr Peer Rev 2019 4 5 10.1186/s41073-019-0064-8 30962953 PMC6434870 Baethge C, Goldbeck-Wood S, Mertens S (2019) SANRA—a scale for the quality assessment of narrative review articles. Res Integr Peer Rev 4:5 30962953 10.1186/s41073-019-0064-8 PMC6434870

15. Jeong J-W Lee M-H Kuroda N Multi-scale deep learning of clinically acquired multi-modal MRI improves the localization of seizure onset zone in children with drug-resistant epilepsy IEEE J Biomed Health Inform 2022 26 5529 5539 10.1109/JBHI.2022.3196330 35925854 PMC9710730 Jeong J-W, Lee M-H, Kuroda N et al (2022) Multi-scale deep learning of clinically acquired multi-modal MRI improves the localization of seizure onset zone in children with drug-resistant epilepsy. IEEE J Biomed Health Inform 26:5529–5539 35925854 10.1109/JBHI.2022.3196330 PMC9710730

16. Quon JL Han M Kim LH Artificial intelligence for automatic cerebral ventricle segmentation and volume calculation: a clinical tool for the evaluation of pediatric hydrocephalus J Neurosurg Pediatr 2021 27 131 138 10.3171/2020.6.PEDS20251 33260138 PMC9707365 Quon JL, Han M, Kim LH et al (2021) Artificial intelligence for automatic cerebral ventricle segmentation and volume calculation: a clinical tool for the evaluation of pediatric hydrocephalus. J Neurosurg Pediatr 27:131–138 33260138 10.3171/2020.6.PEDS20251 PMC9707365

17. Zhang M Wong SW Wright JN MRI radiogenomics of pediatric medulloblastoma: a multicenter study Radiology 2022 304 406 416 10.1148/radiol.212137 35438562 PMC9340239 Zhang M, Wong SW, Wright JN et al (2022) MRI radiogenomics of pediatric medulloblastoma: a multicenter study. Radiology 304:406–416 35438562 10.1148/radiol.212137 PMC9340239

18. Vecchiato K, Egloff A, Carney O, et al (2020) Evaluation of DISORDER: retrospective image motion correction for volumetric brain MRI in a pediatric setting. medRxiv 2020.09.09.20190777. 10.1101/2020.09.09.20190777 PMC8040996 33602745

19. Chang F-C Wong T-T Wu K-S Magnetic resonance radiomics features and prognosticators in different molecular subtypes of pediatric medulloblastoma PLoS One 2021 16 e0255500 10.1371/journal.pone.0255500 34324588 PMC8321137 Chang F-C, Wong T-T, Wu K-S et al (2021) Magnetic resonance radiomics features and prognosticators in different molecular subtypes of pediatric medulloblastoma. PLoS One 16:e0255500 34324588 10.1371/journal.pone.0255500 PMC8321137

20. Yepes-Calderon F McComb JG Accurate image-based CSF volume calculation of the lateral ventricles Sci Rep 2022 12 12115 10.1038/s41598-022-15995-w 35840587 PMC9287564 Yepes-Calderon F, McComb JG (2022) Accurate image-based CSF volume calculation of the lateral ventricles. Sci Rep 12:12115 35840587 10.1038/s41598-022-15995-w PMC9287564

21. Hatamizadeh A, Nath V, Tang Y, et al (2022) Brainlesion: glioma, multiple sclerosis, stroke and traumatic brain injuries, 7th International Workshop, BrainLes 2021, Held in Conjunction with MICCAI 2021, Virtual Event, September 27, 2021, Revised Selected Papers, Part I. Lect Notes Comput Sci 272–284. 10.1007/978-3-031-08999-2_22

22. Spitzer H Ripart M Whitaker K Interpretable surface-based detection of focal cortical dysplasias: a multi-centre epilepsy lesion detection study Brain 2022 145 3859 3871 10.1093/brain/awac224 35953082 PMC9679165 Spitzer H, Ripart M, Whitaker K et al (2022) Interpretable surface-based detection of focal cortical dysplasias: a multi-centre epilepsy lesion detection study. Brain 145:3859–3871 35953082 10.1093/brain/awac224 PMC9679165

23. Wagstyl K Whitaker K Raznahan A Atlas of lesion locations and postsurgical seizure freedom in focal cortical dysplasia: a MELD study Epilepsia 2022 63 61 74 10.1111/epi.17130 34845719 PMC8916105 Wagstyl K, Whitaker K, Raznahan A et al (2022) Atlas of lesion locations and postsurgical seizure freedom in focal cortical dysplasia: a MELD study. Epilepsia 63:61–74 34845719 10.1111/epi.17130 PMC8916105

24. Ripart M Spitzer H Williams LZJ Detection of epileptogenic focal cortical dysplasia using graph neural networks JAMA Neurol 2025 82 397 406 10.1001/jamaneurol.2024.5406 39992650 PMC11851297 Ripart M, Spitzer H, Williams LZJ et al (2025) Detection of epileptogenic focal cortical dysplasia using graph neural networks. JAMA Neurol 82:397–406 39992650 10.1001/jamaneurol.2024.5406 PMC11851297

25. Azad Z Aiman U Shaheen S Artificial intelligence in paediatric head trauma: enhancing diagnostic accuracy for skull fractures and brain haemorrhages Neurosurg Rev 2024 47 641 10.1007/s10143-024-02897-w 39294484 Azad Z, Aiman U, Shaheen S (2024) Artificial intelligence in paediatric head trauma: enhancing diagnostic accuracy for skull fractures and brain haemorrhages. Neurosurg Rev 47:641 39294484 10.1007/s10143-024-02897-w

26. Heo S Ha J Jung W Decision effect of a deep-learning model to assist a head computed tomography order for pediatric traumatic brain injury Sci Rep 2022 12 12454 10.1038/s41598-022-16313-0 35864281 PMC9304372 Heo S, Ha J, Jung W et al (2022) Decision effect of a deep-learning model to assist a head computed tomography order for pediatric traumatic brain injury. Sci Rep 12:12454 35864281 10.1038/s41598-022-16313-0 PMC9304372

27. Gui L Loukas S Lazeyras F Longitudinal study of neonatal brain tissue volumes in preterm infants and their ability to predict neurodevelopmental outcome Neuroimage 2019 185 728 741 10.1016/j.neuroimage.2018.06.034 29908311 Gui L, Loukas S, Lazeyras F et al (2019) Longitudinal study of neonatal brain tissue volumes in preterm infants and their ability to predict neurodevelopmental outcome. Neuroimage 185:728–741 29908311 10.1016/j.neuroimage.2018.06.034

28. Smyser CD Dosenbach NUF Smyser TA Prediction of brain maturity in infants using machine-learning algorithms Neuroimage 2016 136 1 9 10.1016/j.neuroimage.2016.05.029 27179605 PMC4914443 Smyser CD, Dosenbach NUF, Smyser TA et al (2016) Prediction of brain maturity in infants using machine-learning algorithms. Neuroimage 136:1–9 27179605 10.1016/j.neuroimage.2016.05.029 PMC4914443

29. Khalighi S Reddy K Midya A Artificial intelligence in neuro-oncology: advances and challenges in brain tumor diagnosis, prognosis, and precision treatment NPJ Precis Oncol 2024 8 80 10.1038/s41698-024-00575-0 38553633 PMC10980741 Khalighi S, Reddy K, Midya A et al (2024) Artificial intelligence in neuro-oncology: advances and challenges in brain tumor diagnosis, prognosis, and precision treatment. NPJ Precis Oncol 8:80 38553633 10.1038/s41698-024-00575-0 PMC10980741

30. Wu J, Chen L, Li Z, et al (2025) 3D isotropic high-resolution fetal brain MRI reconstruction from motion corrupted thick data based on physical-informed unsupervised learning. IEEE J Biomed Heal Inform PP:1–14. 10.1109/jbhi.2025.3586049 40663667

31. Lajous H, Roy CW, Hilbert T et al (2022) A fetal brain magnetic resonance acquisition numerical phantom (FaBiAN). Sci Rep 12:8682 10.1038/s41598-022-10335-4 PMC9127105 35606398

32. Ciceri T, Squarcina L, Giubergia A et al (2023) Review on deep learning fetal brain segmentation from magnetic resonance images. Artif Intell Med 143:102608 10.1016/j.artmed.2023.102608 37673558

33. Khalili N Turk E Benders MJNL Automatic extraction of the intracranial volume in fetal and neonatal MR scans using convolutional neural networks NeuroImage: Clinical 2019 24 102061 10.1016/j.nicl.2019.102061 31835284 PMC6909142 Khalili N, Turk E, Benders MJNL et al (2019) Automatic extraction of the intracranial volume in fetal and neonatal MR scans using convolutional neural networks. NeuroImage: Clinical 24:102061 31835284 10.1016/j.nicl.2019.102061 PMC6909142

34. Zhao L Asis-Cruz JD Feng X Automated 3D fetal brain segmentation using an optimized deep learning approach AJNR Am J Neuroradiol 2022 43 448 454 10.3174/ajnr.A7419 35177547 PMC8910820 Zhao L, Asis-Cruz JD, Feng X et al (2022) Automated 3D fetal brain segmentation using an optimized deep learning approach. AJNR Am J Neuroradiol 43:448–454 35177547 10.3174/ajnr.A7419 PMC8910820

35. Lim A Lo J Wagner MW Motion artifact correction in fetal MRI based on a generative adversarial network method Biomed Signal Process Control 2023 81 104484 10.1016/j.bspc.2022.104484 Lim A, Lo J, Wagner MW et al (2023) Motion artifact correction in fetal MRI based on a generative adversarial network method. Biomed Signal Process Control 81:104484

36. Yaqub M Kelly B Papageorghiou AT Noble JA A deep learning solution for automatic fetal neurosonographic diagnostic plane verification using clinical standard constraints Ultrasound Med Biol 2017 43 2925 2933 10.1016/j.ultrasmedbio.2017.07.013 28958729 Yaqub M, Kelly B, Papageorghiou AT, Noble JA (2017) A deep learning solution for automatic fetal neurosonographic diagnostic plane verification using clinical standard constraints. Ultrasound Med Biol 43:2925–2933 28958729 10.1016/j.ultrasmedbio.2017.07.013

37. Xie H Wang N He M Using deep learning algorithms to classify fetal brain ultrasound images as normal or abnormal Ultrasound Obstet Gynecol 2020 10.1002/uog.21967 31909548 Xie H, Wang N, He M et al (2020) Using deep learning algorithms to classify fetal brain ultrasound images as normal or abnormal. Ultrasound Obstet Gynecol. 10.1002/uog.21967 31909548

38. Yun HJ Lee H-J You S Deep learning–based brain age prediction using MRI to identify fetuses with cerebral ventriculomegaly Radiol Artif Intell 2025 7 e240115 10.1148/ryai.240115 39969279 PMC11950871 YYun HJ, Lee H-J, You S et al (2025) Deep learning–based brain age prediction using MRI to identify fetuses with cerebral ventriculomegaly. Radiol Artif Intell 7:e240115 39969279 10.1148/ryai.240115 PMC11950871

39. Shen L Zheng J Lee EH Attention-guided deep learning for gestational age prediction using fetal brain MRI Sci Rep 2022 12 1408 10.1038/s41598-022-05468-5 35082346 PMC8791965 Shen L, Zheng J, Lee EH et al (2022) Attention-guided deep learning for gestational age prediction using fetal brain MRI. Sci Rep 12:1408 35082346 10.1038/s41598-022-05468-5 PMC8791965

40. Kwon H You S Yun HJ The role of cortical structural variance in deep learning-based prediction of fetal brain age Front Neurosci 2024 18 1411334 10.3389/fnins.2024.1411334 38846713 PMC11153753 Kwon H, You S, Yun HJ et al (2024) The role of cortical structural variance in deep learning-based prediction of fetal brain age. Front Neurosci 18:1411334 38846713 10.3389/fnins.2024.1411334 PMC11153753

41. Wilson S Yun HJ Sadhwani A Foetal cortical expansion is associated with neurodevelopmental outcome at 2-years in congenital heart disease: a longitudinal follow-up study EBioMedicine 2025 114 105679 10.1016/j.ebiom.2025.105679 40158387 PMC11994330 Wilson S, Yun HJ, Sadhwani A et al (2025) Foetal cortical expansion is associated with neurodevelopmental outcome at 2-years in congenital heart disease: a longitudinal follow-up study. EBioMedicine 114:105679 40158387 10.1016/j.ebiom.2025.105679 PMC11994330

42. Ouyang M, Whitehead MT, Mohapatra S et al (2024) Machine-learning based prediction of future outcome using multimodal MRI during early childhood. Semin Fetal Neonatal Med 29:101561 10.1016/j.siny.2024.101561 PMC11654837 39528363

43. Kelly BS, Lee J, Antram E, et al (2025) Safe for kids? AI medical devices in radiology overlook paediatric suitability. Eur Radiol 1–8. 10.1007/s00330-025-11975-6 41060417

44. Lee EH Han M Wright J An international study presenting a federated learning AI platform for pediatric brain tumors Nat Commun 2024 15 7615 10.1038/s41467-024-51172-5 39223133 PMC11368946 Lee EH, Han M, Wright J et al (2024) An international study presenting a federated learning AI platform for pediatric brain tumors. Nat Commun 15:7615 39223133 10.1038/s41467-024-51172-5 PMC11368946

45. Antonissen N, Tryfonos O, Houben IB et al (2025) Artificial intelligence in radiology: 173 commercially available products and their scientific evidence. Eur Radiol 11. 10.1007/s00330-025-11830-8 PMC12711992 40707732

46. An FDA guide on indications for use and device reporting of artificial intelligence-enabled devices: significance for pediatric use. https://brandon-j-nelson.com/posts/assets/Nelson%20et%20al.%20-%202023%20-%20An%20FDA%20Guide%20on%20Indications%20for%20Use%20and%20Device%20Rep.pdf . Accessed 22 Jul 2024 10.1016/j.jacr.2023.06.004 37400046

47. Brady AP Allen B Chong J Developing, purchasing, implementing and monitoring AI tools in radiology: practical considerations. A multi-society statement from the ACR, CAR, ESR, RANZCR & RSNA Can Assoc Radiol J 2024 75 226 244 10.1177/08465371231222229 38251882 Brady AP, Allen B, Chong J et al (2024) Developing, purchasing, implementing and monitoring AI tools in radiology: practical considerations. A multi-society statement from the ACR, CAR, ESR, RANZCR & RSNA. Can Assoc Radiol J 75:226–244 38251882 10.1177/08465371231222229

48. Brady AP Loewe C Brkljacic B Guidelines and recommendations for radiologist staffing, education and training Insights Imaging 2025 16 57 10.1186/s13244-025-01926-6 40074928 PMC11904080 Brady AP, Loewe C, Brkljacic B et al (2025) Guidelines and recommendations for radiologist staffing, education and training. Insights Imaging 16:57 40074928 10.1186/s13244-025-01926-6 PMC11904080

49. Brady AP Bello JA Derchi LE Radiology in the era of value-based healthcare: a multi-society expert statement from the ACR, CAR, ESR, IS3R, RANZCR, and RSNA Can Assoc Radiol J 2021 72 208 214 10.1177/0846537120982567 33345576 Brady AP, Bello JA, Derchi LE et al (2021) Radiology in the era of value-based healthcare: a multi-society expert statement from the ACR, CAR, ESR, IS3R, RANZCR, and RSNA. Can Assoc Radiol J 72:208–214 33345576 10.1177/0846537120982567

50. Harvey H, Glocker B (2019) Artificial intelligence in medical imaging, opportunities, applications and risks. 61–72. 10.1007/978-3-319-94878-2_6

51. Willemink MJ Koszek WA Hardell C Preparing medical imaging data for machine learning Radiology 2020 295 4 15 10.1148/radiol.2020192224 32068507 PMC7104701 Willemink MJ, Koszek WA, Hardell C et al (2020) Preparing medical imaging data for machine learning. Radiology 295:4–15 32068507 10.1148/radiol.2020192224 PMC7104701

52. He J Baxter SL Xu J The practical implementation of artificial intelligence technologies in medicine Nat Med 2019 25 30 36 10.1038/s41591-018-0307-0 30617336 PMC6995276 He J, Baxter SL, Xu J et al (2019) The practical implementation of artificial intelligence technologies in medicine. Nat Med 25:30–36 30617336 10.1038/s41591-018-0307-0 PMC6995276

53. Recht MP Dewey M Dreyer K Integrating artificial intelligence into the clinical practice of radiology: challenges and recommendations Eur Radiol 2020 30 3576 3584 10.1007/s00330-020-06672-5 32064565 Recht MP, Dewey M, Dreyer K et al (2020) Integrating artificial intelligence into the clinical practice of radiology: challenges and recommendations. Eur Radiol 30:3576–3584 32064565 10.1007/s00330-020-06672-5

54. Daye D Wiggins WF Lungren MP Implementation of clinical artificial intelligence in radiology: who decides and how? Radiology 2022 305 555 563 10.1148/radiol.212151 35916673 PMC9713445 Daye D, Wiggins WF, Lungren MP et al (2022) Implementation of clinical artificial intelligence in radiology: who decides and how? Radiology 305:555–563 35916673 10.1148/radiol.212151 PMC9713445

55. Kelly BS Quinn C Belton N Cybersecurity considerations for radiology departments involved with artificial intelligence Eur Radiol 2023 33 8833 8841 10.1007/s00330-023-09860-1 37418025 PMC10667413 Kelly BS, Quinn C, Belton N et al (2023) Cybersecurity considerations for radiology departments involved with artificial intelligence. Eur Radiol 33:8833–8841 37418025 10.1007/s00330-023-09860-1 PMC10667413

56. Shelmerdine SC, Naidoo J, Kelly BS, et al (2025) AI implementation in pediatric radiology for patient safety: a multi-society statement from the ACR, ESPR, SPR, SLARP, AOSPR, SPIN. Pediatr Radiol 1–14. 10.1007/s00247-025-06386-0 41288670

57. Jiménez-Sánchez A, Avlona N-R, Boer S de, et al (2025) In the picture: medical imaging datasets, artifacts, and their living review. Proc 2025 ACM Conf Fairness, Account, Transpar 511–531. 10.1145/3715275.3732035

58. Shelmerdine SC Revolutionising paediatric radiology: the future impact of artificial intelligence Eur Radiol 2024 34 2294 2296 10.1007/s00330-023-10288-w 37819275 Shelmerdine SC (2024) Revolutionising paediatric radiology: the future impact of artificial intelligence. Eur Radiol 34:2294–2296 37819275 10.1007/s00330-023-10288-w

59. Lee L Salami RK Martin H How I would like AI used for my imaging”: children and young persons’ perspectives Eur Radiol 2024 10.1007/s00330-024-10839-9 38900281 PMC11557655 Lee L, Salami RK, Martin H et al (2024) How I would like AI used for my imaging”: children and young persons’ perspectives. Eur Radiol. 10.1007/s00330-024-10839-9 38900281 PMC11557655

60. Katal S York B Gholamrezanezhad A AI in radiology: from promise to practice − a guide to effective integration Eur J Radiol 2024 181 111798 10.1016/j.ejrad.2024.111798 39471551 Katal S, York B, Gholamrezanezhad A (2024) AI in radiology: from promise to practice − a guide to effective integration. Eur J Radiol 181:111798 39471551 10.1016/j.ejrad.2024.111798

61. dos Santos DP Dietzel M Baessler B A decade of radiomics research: are images really data or just patterns in the noise? Eur Radiol 2021 31 1 4 10.1007/s00330-020-07108-w 32767103 PMC7755615 dos Santos DP, Dietzel M, Baessler B (2021) A decade of radiomics research: are images really data or just patterns in the noise? Eur Radiol 31:1–4 32767103 10.1007/s00330-020-07108-w PMC7755615

62. Brewster RCL Nagy M Wunnava S Bourgeois FT US FDA approval of pediatric artificial intelligence and machine learning–enabled medical devices JAMA Pediatr 2025 179 212 214 10.1001/jamapediatrics.2024.5437 39680415 PMC11791695 Brewster RCL, Nagy M, Wunnava S, Bourgeois FT (2025) US FDA approval of pediatric artificial intelligence and machine learning–enabled medical devices. JAMA Pediatr 179:212–214 39680415 10.1001/jamapediatrics.2024.5437 PMC11791695

63. Shelmerdine SC Rethinking our relationship with AI: for better or worse, richer or poorer? Eur Radiol 2024 10.1007/s00330-024-11007-9 39095603 Shelmerdine SC (2024) Rethinking our relationship with AI: for better or worse, richer or poorer? Eur Radiol. 10.1007/s00330-024-11007-9 39095603

64. O’Neil C, Gunn H (2020) Ethics of artificial intelligence. 237–270. 10.1093/oso/9780190905033.003.0009

65. Kelly B Kirwan A Quinn M The ethical matrix as a method for involving people living with disease and the wider public (PPI) in near-term artificial intelligence research Radiography 2023 10.1016/j.radi.2023.03.009 37062673 Kelly B, Kirwan A, Quinn M et al (2023) The ethical matrix as a method for involving people living with disease and the wider public (PPI) in near-term artificial intelligence research. Radiography. 10.1016/j.radi.2023.03.009 37062673

66. Ciet P Eade C Ho M-L The unintended consequences of artificial intelligence in paediatric radiology Pediatr Radiol 2024 54 585 593 10.1007/s00247-023-05746-y 37665368 Ciet P, Eade C, Ho M-L et al (2024) The unintended consequences of artificial intelligence in paediatric radiology. Pediatr Radiol 54:585–593 37665368 10.1007/s00247-023-05746-y

67. Borys K Schmitt YA Nauta M Explainable AI in medical imaging: an overview for clinical practitioners – beyond saliency-based XAI approaches Eur J Radiol 2023 162 110786 10.1016/j.ejrad.2023.110786 36990051 Borys K, Schmitt YA, Nauta M et al (2023) Explainable AI in medical imaging: an overview for clinical practitioners – beyond saliency-based XAI approaches. Eur J Radiol 162:110786 36990051 10.1016/j.ejrad.2023.110786

68. Stogiannos N Malik R Kumar A Black box no more: a scoping review of AI governance frameworks to guide procurement and adoption of AI in medical imaging and radiotherapy in the UK Br J Radiol 2023 96 20221157 10.1259/bjr.20221157 37747285 PMC10646619 Stogiannos N, Malik R, Kumar A et al (2023) Black box no more: a scoping review of AI governance frameworks to guide procurement and adoption of AI in medical imaging and radiotherapy in the UK. Br J Radiol 96:20221157 37747285 10.1259/bjr.20221157 PMC10646619

69. Gastounioti A Kontos D Is it time to get rid of black boxes and cultivate trust in AI? Radiol Artif Intell 2020 2 e200088 10.1148/ryai.2020200088 32510520 PMC7259191 Gastounioti A, Kontos D (2020) Is it time to get rid of black boxes and cultivate trust in AI? Radiol Artif Intell 2:e200088 32510520 10.1148/ryai.2020200088 PMC7259191

70. Ursin F Timmermann C Steger F Explicability of artificial intelligence in radiology: is a fifth bioethical principle conceptually necessary? Bioethics 2021 10.1111/bioe.12918 34251687 Ursin F, Timmermann C, Steger F (2021) Explicability of artificial intelligence in radiology: is a fifth bioethical principle conceptually necessary? Bioethics. 10.1111/bioe.12918 34251687

71. Malamateniou C Knapp KM Pergola M Artificial intelligence in radiography: where are we now and what does the future hold? Radiography 2021 27 S58 S62 10.1016/j.radi.2021.07.015 34380589 Malamateniou C, Knapp KM, Pergola M et al (2021) Artificial intelligence in radiography: where are we now and what does the future hold? Radiography 27:S58–S62 34380589 10.1016/j.radi.2021.07.015

72. Kim E, Cho H-H, Cho SH et al (2022) Accelerated synthetic MRI with deep learning–based reconstruction for pediatric neuroimaging. AJNR Am J Neuroradiol 43:1653–1659 10.3174/ajnr.A7664 PMC9731246 36175085

73. Nagaraj UD Meineke J Sriwastwa A Evaluation of synthetic images derived from a neural network in pediatric brain magnetic resonance imaging Magn Reson Imaging 2025 121 110427 10.1016/j.mri.2025.110427 40389019 Nagaraj UD, Meineke J, Sriwastwa A et al (2025) Evaluation of synthetic images derived from a neural network in pediatric brain magnetic resonance imaging. Magn Reson Imaging 121:110427 40389019 10.1016/j.mri.2025.110427

74. Pasquini L Napolitano A Pignatelli M Synthetic post-contrast imaging through artificial intelligence: clinical applications of virtual and augmented contrast media Pharmaceutics 2022 14 2378 10.3390/pharmaceutics14112378 36365197 PMC9695136 Pasquini L, Napolitano A, Pignatelli M et al (2022) Synthetic post-contrast imaging through artificial intelligence: clinical applications of virtual and augmented contrast media. Pharmaceutics 14:2378 36365197 10.3390/pharmaceutics14112378 PMC9695136

75. Gräfe D Simion S-H Rosolowski M Brain deposition of gadobutrol in children—a cross-sectional and longitudinal MRI T1 mapping study Eur Radiol 2023 33 4580 4588 10.1007/s00330-022-09297-y 36520178 PMC10289941 Gräfe D, Simion S-H, Rosolowski M et al (2023) Brain deposition of gadobutrol in children—a cross-sectional and longitudinal MRI T1 mapping study. Eur Radiol 33:4580–4588 36520178 10.1007/s00330-022-09297-y PMC10289941

76. Ng CKC Artificial intelligence for radiation dose optimization in pediatric radiology: a systematic review Children 2022 9 1044 10.3390/children9071044 35884028 PMC9320231 Ng CKC (2022) Artificial intelligence for radiation dose optimization in pediatric radiology: a systematic review. Children 9:1044 35884028 10.3390/children9071044 PMC9320231

77. Huang J Shlobin NA Lam SK DeCuypere M Artificial intelligence applications in pediatric brain tumor imaging: a systematic review World Neurosurg 2022 157 99 105 10.1016/j.wneu.2021.10.068 34648981 Huang J, Shlobin NA, Lam SK, DeCuypere M (2022) Artificial intelligence applications in pediatric brain tumor imaging: a systematic review. World Neurosurg 157:99–105 34648981 10.1016/j.wneu.2021.10.068

78. Pacchiano F, Tortora M, Doneda C et al (2024) Radiomics and artificial intelligence applications in pediatric brain tumors. World J Pediatr 20:747–763 10.1007/s12519-024-00823-0 PMC11402857 38935233

79. Ende E Ryan S Crain MA Makary MS Artificial intelligence, augmented reality, and virtual reality advances and applications in interventional radiology Diagnostics 2023 13 892 10.3390/diagnostics13050892 36900036 PMC10000832 Ende E, Ryan S, Crain MA, Makary MS (2023) Artificial intelligence, augmented reality, and virtual reality advances and applications in interventional radiology. Diagnostics 13:892 36900036 10.3390/diagnostics13050892 PMC10000832

80. (ESR) ES of R (2015) Medical imaging in personalised medicine: a white paper of the research committee of the European Society of Radiology (ESR). Insights Imaging 6:141–155 10.1007/s13244-015-0394-0 PMC4376812 25763994

81. Saxena S Jena B Gupta N Role of artificial intelligence in radiogenomics for cancers in the era of precision medicine Cancers 2022 14 2860 10.3390/cancers14122860 35740526 PMC9220825 Saxena S, Jena B, Gupta N et al (2022) Role of artificial intelligence in radiogenomics for cancers in the era of precision medicine. Cancers 14:2860 35740526 10.3390/cancers14122860 PMC9220825

82. Liu Z Duan T Zhang Y Radiogenomics: a key component of precision cancer medicine Br J Cancer 2023 129 741 753 10.1038/s41416-023-02317-8 37414827 PMC10449908 Liu Z, Duan T, Zhang Y et al (2023) Radiogenomics: a key component of precision cancer medicine. Br J Cancer 129:741–753 37414827 10.1038/s41416-023-02317-8 PMC10449908

83. Shahi N Shahi AK Phillips R Using deep learning and natural language processing models to detect child physical abuse J Pediatr Surg 2021 56 2326 2332 10.1016/j.jpedsurg.2021.03.007 33838900 Shahi N, Shahi AK, Phillips R et al (2021) Using deep learning and natural language processing models to detect child physical abuse. J Pediatr Surg 56:2326–2332 33838900 10.1016/j.jpedsurg.2021.03.007

84. Tveit J Aurlien H Plis S Automated interpretation of clinical electroencephalograms using artificial intelligence JAMA Neurol 2023 80 805 812 10.1001/jamaneurol.2023.1645 37338864 PMC10282956 Tveit J, Aurlien H, Plis S et al (2023) Automated interpretation of clinical electroencephalograms using artificial intelligence. JAMA Neurol 80:805–812 37338864 10.1001/jamaneurol.2023.1645 PMC10282956

85. Kim H Hwang H Application of artificial intelligence (AI) and machine learning (ML) in pediatric epilepsy: a narrative review Pediatr Med Chir 2021 0 0 0 Kim H, Hwang H (2021) Application of artificial intelligence (AI) and machine learning (ML) in pediatric epilepsy: a narrative review. Pediatr Med Chir 0:0–0

86. Davendralingam N Sebire NJ Arthurs OJ Shelmerdine SC Artificial intelligence in paediatric radiology: future opportunities Br J Radiol 2021 94 20200975 10.1259/bjr.20200975 32941736 PMC7774693 Davendralingam N, Sebire NJ, Arthurs OJ, Shelmerdine SC (2021) Artificial intelligence in paediatric radiology: future opportunities. Br J Radiol 94:20200975 32941736 10.1259/bjr.20200975 PMC7774693

87. Katzman BD Pol CB Soyer P Patlas MN Artificial intelligence in emergency radiology: a review of applications and possibilities Diagn Interv Imaging 2023 104 6 10 10.1016/j.diii.2022.07.005 35933269 Katzman BD, Pol CB, Soyer P, Patlas MN (2023) Artificial intelligence in emergency radiology: a review of applications and possibilities. Diagn Interv Imaging 104:6–10 35933269 10.1016/j.diii.2022.07.005

88. Nelson BJ Zeng R Sammer MBK An FDA guide on indications for use and device reporting of artificial intelligence-enabled devices: significance for pediatric use J Am Coll Radiol 2023 20 738 741 10.1016/j.jacr.2023.06.004 37400046 Nelson BJ, Zeng R, Sammer MBK et al (2023) An FDA guide on indications for use and device reporting of artificial intelligence-enabled devices: significance for pediatric use. J Am Coll Radiol 20:738–741 37400046 10.1016/j.jacr.2023.06.004