Submit to this Journal Review for this Journal Propose a Special Issue

Article Menu

Share Help Cite Discuss in SciProfiles

Open AccessReview

Peer-Review Record

Current State and Future of Artificial Intelligence in Pediatric Interventional Radiology: A Narrative Review

Diagnostics 2026, 16(12), 1918; https://doi.org/10.3390/diagnostics16121918

by Abdulaziz Mohammad Al-Sharydah

Reviewer 1: Anonymous

Reviewer 2:

Michael Friebe

Reviewer 3:

Michael W. Vannier

Diagnostics 2026, 16(12), 1918; https://doi.org/10.3390/diagnostics16121918

Submission received: 1 May 2026 / Revised: 18 June 2026 / Accepted: 19 June 2026 / Published: 20 June 2026

(This article belongs to the Section Machine Learning and Artificial Intelligence in Diagnostics)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This article comprehensively reviews the current state of artificial intelligence and robotic systems in pediatric interventional radiology, their potential clinical implications, and future research directions. It is recommended that you revise the article considering the comments listed below, which will help to more clearly highlight the validity, clinical limitations, and research priorities specific to pediatric applications.
1) Although the abstract states that it summarizes the current state of Artificial Intelligence applications in Pediatric Interventional Radiology, it is unclear which clinical applications were examined or which databases were used. Therefore, a brief sentence explaining the scope of the review and the main data sources would help the reader better understand the scope of the article.
2) Although the abstract states that the study will present “future perspectives,” the original contribution of the review to the literature is unclear. Therefore, it would be beneficial to state the original contribution of the study to the literature or its key findings in a sentence in the abstract. 3) The introduction highlights the rapid advancement of AI in diagnostic radiology, but while its current limited use in PIR is described as "fragmented use," no concrete data or examples are provided. It is suggested that the introduction strengthen the argument by adding examples or numerical data of the limited use in PIR.
4) Although the abbreviations used (AI, ML, DL, PIR) are given in their full forms when first mentioned, some specific terms (e.g., AI-enabled robotics, federated learning) are used without brief explanations. Brief explanations of such terms should be added to make it easier for the reader to follow the subject.
5) The second part of the study describes the technical details of the technologies (ML, DL, CNN, etc.) very comprehensively, but a short paragraph on "contribution to PIR" should be added after each technological explanation to balance both the technical and clinical perspectives.
6) The second part presents applications such as OptiqAI, Smart Collimation Thorax, and CNN-based segmentation with examples. However, most systems were developed with adult data and have not been validated for children. Therefore, risks and limitations should be more clearly stated so that the reader understands which findings are truly applicable to pediatric patients.
7) The post-procedure monitoring section describes risk estimation and monitoring systems, but it is unclear whether these models have been clinically tested or what their limitations are. It is important for this study to specify which systems are still in the research phase and in which areas validation is needed.
8) The "Future Directions" section comprehensively presents advanced AI and robotic systems, but most examples are based on adult data or preclinical/prototype data. It should be clarified which systems have undergone pediatric validation and which findings are still in the research/prediction phase.
9) The potential benefits of AI and robotic systems are well described in the "Future Directions" section, but limitations, cost, accessibility, and training requirements are not sufficiently discussed. These discussions should be added.
10) The integration of AI and robotic tools into existing PIR software and workflows, training requirements, and trust issues should be addressed.

Author Response

Reviewer 1

Comments and Suggestions for Authors

1) Although the abstract states that it summarizes the current state of Artificial Intelligence applications in Pediatric Interventional Radiology, it is unclear which clinical applications were examined or which databases were used. Therefore, a brief sentence explaining the scope of the review and the main data sources would help the reader better understand the scope of the article.

Response: We thank the reviewer for this insightful suggestion and agree that, while the abstract clearly states that this is a narrative review of AI in PIR and lists several application domains, it does not explicitly state which databases were searched or the time frame, which might leave the scope less transparent to readers.

The Introduction already contains a detailed description of the narrative search strategy (MEDLINE/PubMed, Embase, Scopus, and relevant society publications up to the first quarter of 2026); however, this information was not summarized in the Abstract.

In response, a brief methods‑style sentence has been added to the Abstract as follows:

"Peer‑reviewed literature and position statements identified through MEDLINE/PubMed, Embase, Scopus, and major society publications up to the first quarter of 2026 are synthesized, focusing on AI applications across the PIR care pathway, including dose‑sparing image acquisition and reconstruction, automated image interpretation and computer‑aided diagnosis, data‑driven procedural planning and navigation, and post‑procedural risk prediction and monitoring."

2) Although the abstract states that the study will present “future perspectives,” the original contribution of the review to the literature is unclear. Therefore, it would be beneficial to state the original contribution of the study to the literature or its key findings in a sentence in the abstract.

Response: The Abstract, as currently written, describes the topics covered and mentions “future perspectives,” but it does not clearly state the distinctive contribution or key take‑home messages of this review.

To address this concern, the following sentence has been added to the Abstract:

"In doing so, this review delineates the current evidence gaps and priority directions for clinically meaningful AI adoption in PIR."

3) The introduction highlights the rapid advancement of AI in diagnostic radiology, but while its current limited use in PIR is described as "fragmented use," no concrete data or examples are provided. It is suggested that the introduction strengthen the argument by adding examples or numerical data of the limited use in PIR.

Response: In the current Introduction, it is stated that IR, particularly PIR, “has seen limited, fragmented use,” but I did not directly anchor this phrase to a specific PIR‑focused source. To address this, this statement is now explicitly linked to the review by Desai et al., which notes that only a handful of current and emerging AI applications have been described for PIR and that these are largely early and exploratory in nature [[Ref 3: “Current and emerging artificial intelligence applications for pediatric interventional radiology.”]]. Based on this evidence, the sentence in the Introduction has been slightly revised to read as follows:

"In medical imaging, AI has progressed rapidly in diagnostic radiology, especially in high‑volume adult imaging, whereas interventional radiology (IR), particularly pediatric IR (PIR), has seen limited, fragmented use, consistent with the findings of prior reports describing only a small number of early, exploratory PIR AI applications [1-3]."

This modification retains the original flow and reference order while providing a concrete, PIR‑specific basis for the characterization of limited and fragmented AI use in PIR.

4) Although the abbreviations used (AI, ML, DL, PIR) are given in their full forms when first mentioned, some specific terms (e.g., AI-enabled robotics, federated learning) are used without brief explanations. Brief explanations of such terms should be added to make it easier for the reader to follow the subject.

Response: I agree that, although core acronyms such as AI, ML, DL, and PIR are defined at first use, terms including “AI‑enabled robotics,” “explainable AI,” and “federated learning” may be unfamiliar to some readers and benefit from concise explanations.

In response, one‑sentence definitions have been added at the point of first mention for these concepts, while keeping the text compact to preserve flow. These clarifications are intended to make the review more accessible to clinicians and trainees who may not have a technical background in AI or computer science:

AI‑enabled / AI‑adjacent robotics:

" In IR and PIR, AI has been explored for workflow optimization, intraprocedural imaging, dose management, wire and catheter navigation, and postprocedural outcome prediction. Most of these implementations remain at an early or investigational stage. In this context, AI‑enabled robotics describes platforms that incorporate AI‑derived image analysis or decision‑support modules, while AI‑adjacent robotics refers to systems that generate rich procedural data streams but whose control logic remains predominantly rule‑based and operator‑directed."

Federated learning

"Federated learning is a distributed training approach in which centers jointly train models by sharing model parameters or gradients rather than raw patient data, thereby preserving local data privacy [6]."

Explainable artificial intelligence (XAI)

"Explainable artificial intelligence (XAI) refers to methods that make ML predictions interpretable to humans, for example, through visualizations, feature‑importance rankings, or local surrogate models [4,5,14]."

5) The second part of the study describes the technical details of the technologies (ML, DL, CNN, etc.) very comprehensively, but a short paragraph on "contribution to PIR" should be added after each technological explanation to balance both the technical and clinical perspectives.

Response: I agree that the technical overview in Section 2.1 would benefit from a clearer statement of its practical contribution to PIR. Rather than inserting a separate “contribution to PIR” paragraph after each individual technical term, which would risk repetition and disrupt the flow of the review, a concise bridging paragraph at the end of the technology overview section has been added. This new paragraph explicitly links the discussed AI methods to their practical relevance in PIR, including image enhancement, segmentation, procedural planning, navigation, workflow support, and outcome prediction, thereby improving the balance between technical explanation and clinical applicability.

"From a PIR perspective, these AI methods are clinically relevant not as standalone computational techniques, but as enabling tools for dose‑sparing image enhancement, automated segmentation, device and access planning, procedural navigation, workflow optimization, and postprocedural risk prediction [3,8,16]."

6) The second part presents applications such as OptiqAI, Smart Collimation Thorax, and CNN-based segmentation with examples. However, most systems were developed with adult data and have not been validated for children. Therefore, risks and limitations should be more clearly stated so that the reader understands which findings are truly applicable to pediatric patients.

Response: While the Introduction and challenges sections highlight that most existing AI tools are developed and validated on adult data and that no AI software as a medical device currently has a labeled intended use specific to PIR, the subsections describing systems such as OptiqAI, Smart Collimation Thorax, and CNN‑based segmentation did not always restate these limitations at the point of discussion. To address this, these application‑focused paragraphs have been revised to explicitly note where systems were developed or validated predominantly in adults, to specify that pediatric applicability is extrapolated rather than proven, and to emphasize the need for pediatric‑specific validation and age‑stratified performance reporting. These additions clarify for readers which findings are directly supported in children and which represent adult‑derived concepts that are, at present, only conceptually or preliminarily transferable to PIR.

Below are concise, targeted edits that make the risks and limitations clearer without disrupting the structure or reference numbering.

OptiqAI and Smart Collimation Thorax (Section 2.3.1, Image acquisition and reconstruction)

Current text:

[[In 2025, Siemens Healthineers introduced OptiqAI for Artis interventional platforms (genio/icono), targeting optimized imaging, procedure time, and dose (Figure 1a–b) [17]. The Smart Collimation Thorax (Philips) uses an AI algorithm to adjust the detector height and proposes collimation from three-dimensional (3D) camera data, reducing the examination time and unnecessary exposure. This demonstrates the use of AI for positioning and collimation in plain film/fluoroscopy, which could conceptually extend to interventional fluoroscopy, although this tool is not marketed or validated for pediatrics (Figure 1c) [18].]]

Revision (minimal change; new clauses in italics):

“In 2025, Siemens Healthineers introduced OptiqAI for Artis interventional platforms (genio/icono/pheno), an AI‑powered imaging chain that combines real‑time image denoising with big‑data–driven automatic exposure control across fluoroscopy, acqui-sition, and digital subtraction angiography, dynamically adjusting tube voltage, tube current, copper prefiltration, focal spot size, pulse width, detector dose, and collimation to maintain the requested image quality at the lowest procedure time and reasonable dose) [17].’’

CNN‑based segmentation and pediatric relevance (Section 2.3.2, Image interpretation and diagnosis)

Current text:

[[DL‑based segmentation of organs, vessels, and lesions on CT, MRI, and ultrasonography enables volumetry, distance measurement, and risk assessment. Automated hepatic segmentation can streamline portal vein recanalization, transjugular intrahepatic portosystemic shunts, ablation, and embolization planning in children with portal hypertension, tumors, or vascular anomalies. Venous segmentation also supports stent planning for central venous occlusion or malformations, both of which are major indications for PIR [3,5]. CNN-driven segmentation powers many procedural toolkits, underpinning centerline extraction, lesion volumetry, and path‑length estimation for device selection [3,5].]]

Addition at the end of this paragraph:

"However, most currently available segmentation algorithms and procedural toolkits have been trained and validated using adult datasets, and their performance in small or anatomically variant pediatric patients remain incompletely characterized, highlighting the need for pediatric-specific training and validation before routine PIR deployment [8]."

General reminder in “Applications of AI in PIR” section

At the very end of Section 2.3 (just before “3. Future Directions”), a single summarizing sentence that echoes the reviewer’s concern across all applications has been added:

"Across these applications, an important limitation is that many commercially available or prototype systems were developed using adult or mixed‑age data; therefore, their performance and safety in children must be regarded as extrapolations until age stratified validation and PIR-specific outcome data are available."

7) The post-procedure monitoring section describes risk estimation and monitoring systems, but it is unclear whether these models have been clinically tested or what their limitations are. It is important for this study to specify which systems are still in the research phase and in which areas validation is needed.

Response: The current text emphasizes conceptual opportunities and proof‑of‑principle studies but does not always specify whether individual models are limited to research settings, developed in non‑PIR populations, or prospectively validated in clinical practice. To address this, the post‑procedure monitoring section has been revised to (i) explicitly indicate that most cited ML models in pediatric liver transplantation and hepatocellular carcinoma are research‑phase tools without PIR‑specific prospective validation, (ii) distinguish between adult IR templates and pediatric extrapolations, and (iii) highlight key unmet validation needs, such as age‑stratified performance, calibration, workflow integration, and impact on PIR‑relevant clinical endpoints. These clarifications should help readers understand which findings are currently evidence‑based in children and which remain aspirational frameworks for future PIR research.

Clarified research‑phase status of ML models in pediatric liver transplantation

Current text:

[[In pediatric liver transplantation, ML models have predicted wait-list outcomes, early graft failure, and rejection from perioperative variables, laboratory trajectories, and transcriptomic profiles, providing high‑discrimination risk stratifications [5].]]

Addition at the end of this paragraph:

"To date, these models have been evaluated primarily in retrospective or single‑center cohorts and remain research tools rather than PIR‑integrated decision support systems, with prospective validation and external benchmarking still needed [5]."

Clarified extrapolation from adult IR outcome‑prediction templates

Current text:

[[In IR, ML prediction of responses to intra-arterial hepatocellular carcinoma therapies offers a template for embolization and ablation outcome models [3,5].]]

Addition at the end of this paragraph:

"However, these outcome‑prediction frameworks were developed in predominantly adult populations, and their structure is currently used as a conceptual template rather than an already validated model for PIR [3,5,9]."

Explicit statement of validation gaps for proposed PIR applications

Current text:

[Similar frameworks in pediatric IR could estimate re‑bleeding risk after hemorrhage control, complications after vascular tumor embolization, or stenosis after venous stenting, guiding tailored surveillance and early re‑intervention. Advanced models can integrate procedural parameters, immediate post‑procedural imaging, and early laboratory markers [3,5,9].]

Addition at the end of this paragraph:

"At present, such PIR‑specific models remain largely hypothetical or at an early exploratory stage; critical gaps include prospective multicenter validation, age‑ and diagnosis‑stratified performance reporting, and demonstration of impact on surveillance strategies and clinical outcomes [3,5,15]."

Clarified status of automated imaging and electronic‑record monitoring systems

Current text:

[Automated systems can analyze follow‑up ultrasonography, CT, or MRI for stent thrombosis, bile leak, abscess, or malformation progression, flagging studies for priority review, while AI‑based monitoring of electronic records, similar to early warning systems in pediatric intensive care, can detect subtle deterioration after complex PIR [3,6,12].]

Addition at the end of this paragraph:

"Most such systems are adapted from diagnostic radiology or pediatric intensive care settings and have not been systematically evaluated in PIR workflows; their sensitivity, specificity, and false‑alarm burden in the immediate post‑interventional period remain key areas for future validation [3,6,12]."

8) The "Future Directions" section comprehensively presents advanced AI and robotic systems, but most examples are based on adult data or preclinical/prototype data. It should be clarified which systems have undergone pediatric validation and which findings are still in the research/prediction phase.

Response: While portions of the text and table footnotes already indicate that platforms such as LIBERTY, CorPath GRX, and Magellan lack pediatric‑specific series and that most handheld robotic ultrasound systems are preclinical prototypes, this information is not consistently summarized at the section level. To address this, the “Future Directions” section has been revised to (i) explicitly distinguish systems that currently have pediatric labeling or early pediatric use (e.g., the Mendaera Focalist system) from those supported only by adult or preclinical data, and (ii) highlight that for nearly all platforms described, PIR‑specific validation, age‑stratified performance, workflow integration, and outcome data remain outstanding needs. These clarifications ensure that readers can readily understand which examples represent clinically tested pediatric technologies versus conceptual or research‑phase directions.

Emerging AI technologies (Section 3.1)

Current text:

[For PIR, analogous models could inform choices between interventional, surgical, or medical management for portal hypertension, complex vascular malformations, and transplant-related complications [4,6,14].]

Addition at the end of this paragraph:

AI‑adjacent robotic and smart navigation systems (Section 3.3)

Current text:

[From a pediatric perspective, they illustrated how robotics and remote navigation could support safer catheterization in small tortuous vessels and even centralized expert intervention with local teams; however, no system has yet been optimized, labeled, or systematically evaluated for PIR [3,26,33].]

This text already states that none have been designed or validated specifically for children and this was mentioned as a footnote in Table 1.

AI‑enabled handheld robotic ultrasound guidance (Section 3.4)

Current text:

[The Mendaera Focalist handheld robotic system recently received FDA 510(k) clearance for ultrasound‑guided needle procedures in adult and pediatric patients, representing the first regulator‑recognized commercial implementation and underscoring the clinical need for PIR‑specific evaluation [37] (Table 2 and Figure 3).]

Addition at the end of this paragraph:

"Contrarily, AI‑GUIDE, HUMaN, and cooperative ultrasound robots remain preclinical or prototype systems without prospective pediatric clinical studies; therefore, their proposed roles in PIR should currently be viewed as investigational and hypothesis‑generating [34–36]."

9) The potential benefits of AI and robotic systems are well described in the "Future Directions" section, but limitations, cost, accessibility, and training requirements are not sufficiently discussed. These discussions should be added.

Response: The current “Future Directions” section focuses primarily on the technical capabilities and potential clinical impact of advanced AI and robotic systems, but it does not sufficiently address practical constraints such as capital and maintenance costs, unequal access between high‑ and low‑resource centers, or the training and workflow changes required for safe deployment. To address this, additional text has been added to explicitly discuss: (i) the high upfront and ongoing costs of robotic and advanced AI platforms and their implications for smaller or resource‑limited pediatric centers, (ii) infrastructure and interoperability requirements that may limit accessibility to well‑resourced institutions, and (iii) the need for structured training, credentialing, and multidisciplinary governance to ensure safe and equitable implementation. These additions complement the description of potential benefits and provide a more balanced, realistic view of future AI and robotic integration in PIR.

A short paragraph on limitations and cost has been added in Section 3.3 (AI‑adjacent robotic and smart navigation systems)

"Beyond their technical promise, these robotic platforms are constrained by substantial capital and maintenance costs, dedicated space and shielding requirements, and the need for compatible imaging infrastructure, which currently limits their deployment to a small number of well‑resourced centers. For many pediatric institutions, especially those with lower procedural volumes, such costs and logistical demands may outweigh potential benefits unless clear gains in operator safety, procedural efficiency, and patient outcomes can be demonstrated and supported by appropriate reimbursement models."

A paragraph on accessibility and training has been added in Section 3.4 (AI‑enabled handheld robotic ultrasound guidance)

"Similarly, for more compact handheld systems, broad adoption will depend on practical factors such as device affordability, integration with existing ultrasound platforms, and availability of technical support across diverse practice settings. In addition, safe use requires structured training in both ultrasound and robotic workflows, clear delineation of responsibilities between interventionalists, anesthesiologists, and bedside clinicians, and robust guidance on credentialing and ongoing competency assessment."

A brief, integrative paragraph has been added at the end of Section 3 (before Conclusions)

"Overall, the successful integration of advanced AI and robotic systems into PIR will require not only technical maturity but careful consideration of cost‑effectiveness, equity of access between high‑ and low‑resource centers, and the significant training and change‑management efforts needed to embed these tools safely into daily practice. Without explicit attention to these factors, such technologies risk widening existing disparities in pediatric interventional care rather than narrowing them."

10) The integration of AI and robotic tools into existing PIR software and workflows, training requirements, and trust issues should be addressed.

Response: The current manuscript describes several AI and robotic tools as well as procedural software platforms but gives less detail on how these technologies would be embedded into day‑to‑day PIR practice. To address this, the “Future Directions” and “Research priorities and collaborative studies” sections have been expanded to (i) describe how AI components and robotic systems can be integrated into existing PIR consoles and procedural software, (ii) outline training and credentialing requirements for interventionalists, trainees, technologists, and nursing staff, and (iii) discuss trust, transparency, and governance issues that influence clinicians’ willingness to rely on AI‑derived outputs in critical intra‑ and post‑procedural decisions. These additions are intended to complement the technical description of potential benefits with a more practical roadmap for safe and responsible adoption in real‑world PIR workflows.

A paragraph on integration into existing PIR software and workflows has been added; end of Section 3.2 (“Potential impact on clinical practice”)

"In practice, AI and robotic functions are most likely to be adopted when they are embedded into the procedural software, angiography consoles, and planning tools that PIR teams already use, rather than as stand‑alone applications. Integration should align AI outputs with specific decision points along the PIR workflow; for example, automatic segmentations and centerlines embedded within planning viewers, dose‑optimization presets integrated into protocol-selection interface, and risk scores displayed in peri‑procedural dashboards to enhance existing workflows without disrupting clinical practice."

A paragraph on training requirements has been added; section 3.5 (“Research priorities and collaborative studies”)

"In addition to usability and trust, structured training is essential: PIR attendings, fellows, technologists, and nurses will need to undergo curricula covering basic AI concepts, system limitations and failure modes, and hands‑on experience with AI‑enabled consoles and robotic interfaces. Simulation‑based training and supervised proctoring for new systems, together with clear institutional policies on credentialing and ongoing competency assessment, will be critical to ensure safe deployment."

A paragraph on trust and transparency has been added; end of Section 3.1 (“Emerging AI technologies”)

"Clinician trust in AI‑derived segmentations, navigation suggestions, and risk predictions will depend on transparent performance reporting (including age‑stratified accuracy and calibration), clear indication of uncertainty, and, where feasible, explainable interfaces that reveal the main factors driving a recommendation. In addition, governance frameworks should define how AI‑supported decisions are documented and audited, and how responsibility is shared when AI outputs contribute to complications, so that teams can incorporate these tools with confidence rather than concern."

These concise edits directly address the reviewer’s concern by explicitly describing integration (where AI fits in consoles and software, and at which decision points), highlighting training needs (curricula, simulation, credentialing), and outlining trust (transparency, uncertainty, governance, responsibility).

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for this narrative review that summarizes current and emerging AI applications along the PIR workflow (e.g. dose reduction, image reconstruction and interpretation, procedural planning and navigation, robotics, and risk prediction).

You are emphasizing pediatric‑specific challenges such as data scarcity, regulatory and ethical constraints, and the apparent lack of pediatric validation. The conclusion is a little short, but promising.

All in all it is a well written paper with some items that should be improved or better presented.

For example the narrative review methodology is only briefly described. The authors state that they followed SNARA guidance and performed “narrative searches” of multiple databases up to Q1 2026 with topic‑specific keywords, but provide no real search strings, inclusion/exclusion criteria, or explicit rationale for study selection beyond clinical relevance.

While for a narrative review this may be acceptable, transparency could be improved by providing these details.

There might also be a potential citation and selection bias, as many examples come from adult IR, cardiology, and transplantation, and some key statements on workflow impact (e.g., AI‑driven scheduling reducing outpatient wait times by one‑third) are supported by “pilot studies” without quantitative details or critical appraisal of study quality.

The paper provides some examples though, but these are often just descriptive and rarely contrasts positive studies with negative or neutral results. Also, it does not systematically address risk of bias, and seldom quantifies effect sizes beyond a few selected instances (e.g., 50% CT dose reduction).

Concepts such as data scarcity, ethical issues, and lack of pediatric validation recur across sections with similar wording. This should be corrected and streamlined. Several subsections (e.g., generic AI in pediatric cardiology or perinatology) are only indirectly linked back to PIR, and the strength of extrapolation to interventional workflows is not always discussed.

The central conclusion that AI is technically feasible across the PIR workflow, that dedicated pediatric design/validation are lacking, and that future work should focus on pediatric endpoints (dose, fluoroscopy time, anesthesia duration, complications, LOS, patient‑reported outcomes) are well aligned with the cited literature and appropriately cautious.

Clarifying the text when statements are aspirational (hypothesis‑generating) rather than evidence‑based would strengthen the scientific tone.

The literature base is broad and generally well chosen. The inclusion of major white papers (ACR Pediatric AI Workgroup, multisociety safety statement) and disease‑specific AI frameworks (AI‑RAPNO, pediatric liver transplantation) supports a reasonably complete view of the space as of early 2026. The selection is however clearly purposive rather than systematic with the authors primarily citing illustrative, often favorable examples and reviews, with little explicit discussion of negative studies, failed implementations, or conflicting results. Some potentially relevant domains, e.g. AI‑based dose tracking registries, real‑world post‑market surveillance of pediatric AI tools, or more granular economic evaluations are mentioned only briefly or not at all. For a narrative review, the breadth is good, but the paper would benefit from explicitly noting where evidence is anecdotal or based on small series, and possibly adding a table differentiating adult, mixed, and truly pediatric cohorts in the cited AI studies.

Some detailed comments on the sections and figures:

Figure 1B looks the same than 1A. This is the standard “hardware”. Please focus on the AI part. OptiqAI - great to mention that, but what is it and why is it relevant? What is the purpose of putting a Thorax Unit (1C) in here - if then put the fitting system in here —> https://www.usa.philips.com/healthcare/brand/azurion-image-guided-therapy-system? What about the units from GE that actually were first with AI enabled capabilities — https://www.gehealthcare.com/en/products/image-guiding-solutions?appId=aemshell — and of course CANON and UNITED IMAGING has
Figure 2 - what about the unit from INTUITIVE - https://www.intuitive.com/en-us/products-and-services/ion/how-ion-works
Figure 4 - I am not sure what the value of that figure is … the only thing that is a highlight are the paintings on the wall ... everything else is what a newly build state of the art interventional radiology suite should have.
Please check for minor typographical issues (spacing, hyphenation, reference formatting) that will improve readability, but there are no major language issues.

Overall, this is a timely and well‑written narrative on an under‑served but important topic. With relatively minor revisions focused on methodological transparency, clearer differentiation of PIR‑specific evidence, and some streamlining/clarification of speculative elements, editing the figures I would support acceptance.

Author Response

Reviewer 2

Comments and Suggestions for Authors

All in all it is a well written paper with some items that should be improved or better presented.

While for a narrative review this may be acceptable, transparency could be improved by providing these details.

Response: In response to the comments received, a separate subsection titled “Search strategy and selection” has been added to the Introduction.

"1.1. Search strategy and selection

Consistent with published guidance on the structured reporting of narrative reviews, this work was designed as a narrative, clinically oriented review rather than a formal systematic review, given the heterogeneity and early, largely feasibility‑level nature of PIR–related AI studies. Narrative searches of MEDLINE/PubMed, Embase, and Scopus were performed up to the first quarter of 2026 using combinations of terms such as “pediatric interventional radiology,” “pediatric cardiac catheterization,” “artificial intelligence,” “machine learning,” “deep learning,” “robotic,” “navigation,” “dose reduction,” and “risk prediction,” and were supplemented by key guidelines, position statements, and white papers from professional societies including the ACR, the Society of Interventional Radiology (SIR), the Society for Pediatric Radiology (SPR), and the Cardiovascular and Interventional Radiological Society of Europe (CIRSE), as well as related pediatric subspecialty groups in cardiology, transplantation, and neuro‑oncology. English‑language, peer‑reviewed articles and conference papers describing AI or robotic methods with direct or conceptual relevance to PIR workflows in pediatric or mixed‑age populations were prioritized, whereas purely adult diagnostic studies without interventional implications, non‑peer‑reviewed material, and non‑medical technical reports were generally excluded. Candidate articles were screened by title and abstract, followed by full‑text review where appropriate. When overlapping publications were identified, the most recent, influential (higher‑impact), or more comprehensive was preferentially cited. This process yielded approximately 70–80 primary and review articles, along with several society statements, which were synthesized into a single clinically oriented framework spanning diagnostic, interventional, technical, and policy perspectives. Accordingly, this review should be interpreted as a hypothesis‑generating narrative overview rather than a systematic evidence inventory."

Response: The review intentionally drew on examples from adult IR, cardiology, and transplantation because PIR‑specific AI studies remain scarce; however, the original text did not always make this extrapolation explicit and included a statement about AI‑driven scheduling reducing PIR outpatient wait times by approximately one‑third without detailing the pilot nature and limitations of the underlying work. To address this, the revised manuscript now (i) explicitly labels adult and non‑PIR examples as conceptual templates rather than directly validated PIR evidence, acknowledging the resulting selection bias, and (ii) softens and qualifies the scheduling statement to emphasize that it is based on small, single‑center pilot experience and requires confirmation in larger, prospective pediatric IR cohorts. These changes aim to provide a clearer and more balanced appraisal of the strength and generalizability of the cited data, while still illustrating plausible future benefits.

Section 3.2 (“Potential impact on clinical practice”)

Current text:

[AI‑driven scheduling has reduced pediatric IR outpatient wait times by approximately one‑third in pilot studies, likely improving patient satisfaction and throughput, while smart assistants using natural language processing can triage calls, answer common questions, and retrieve device information, reducing the cognitive burden and potentially mitigating burnout [3,7,15].]

Revision:

"Early pilot experience suggests that AI‑assisted scheduling can meaningfully shorten PIR outpatient waiting times and improve clinic throughput; nonetheless, these reports are based on small, single‑center cohorts without controlled comparison and should therefore be interpreted with caution [3,7,15]. Smart assistants using natural language processing may similarly help triage calls, answer common questions, and retrieve de-vice information, potentially reducing cognitive burden and mitigating burnout, although these applications also remain at an early developmental stage [3,15,26]."

Response: This review was intentionally designed as a narrative, clinically oriented synthesis rather than a formal systematic review, and therefore did not apply structured risk‑of‑bias tools or attempt a quantitative meta‑analysis. As a result, many examples are presented in a qualitative manner, and only selected findings—such as the approximately 50% dose reduction with DL‑based pediatric CT reconstruction and the underperformance of adult‑trained algorithms in children reported by Sammer et al. —are explicitly quantified or highlighted as negative. In response to this comment, a brief paragraph has been added to the section on research priorities to (i) acknowledge these methodological limitations of the review, (ii) note that current pediatric and PIR‑relevant AI studies are heterogeneous and often small, and (iii) emphasize that future PIR work should report effect sizes (with uncertainty) and incorporate basic bias assessment so that positive, neutral, and negative findings can be compared more rigorously.

Addition at the end of Section 3.5. Research priorities and collaborative studies:

"As a narrative, clinically oriented review, this article summarizes illustrative examples from heterogeneous AI and robotic studies without applying a formal, tool‑based risk‑of‑bias assessment. Many cited reports are small, single‑center or retrospective series, and effect sizes are quantified only in selected instances where robust pediatric data are available, such as low‑dose CT reconstruction, while other examples remain primarily descriptive [10,13–15]. Positive findings are therefore not systematically contrasted with neutral or negative results, and extrapolations from adult IR, cardiology, and transplantation introduce an inherent citation and selection bias. Future PIR‑focused AI research should not only explore whether proposed tools improve diagnostic, procedural, or workflow metrics, but also report standardized quantitative measures (including effect sizes and uncertainty) and incorporate basic bias assessment, so that benefits, limitations, and failures can be compared and synthesized across centers in a more rigorous manner [10,13–15]."

Response: The current structure intentionally foregrounds these constraints because they are central themes in pediatric AI; however, this has resulted in similar sentences appearing in multiple sections and in some generic pediatric AI domains (e.g., cardiology, perinatology, neuro‑oncology) not being explicitly re‑connected to PIR workflows each time. To address this, overlapping sentences on data scarcity, ethics, and validation have been consolidated into a single, more comprehensive paragraph in the “Challenges in implementing AI in pediatric patients” section, and later sections refer back to this paragraph rather than restating the same concepts. In parallel, a short integrative paragraph has been added at the end of the “Emerging AI technologies” and “Research priorities” content to explicitly explain how generic pediatric AI advances in cardiology, perinatology, and neuro‑oncology can—and cannot—be extrapolated to PIR, including a brief statement on the limitations and assumptions of such extrapolation.

Section 2.2 (Challenges in implementing AI in pediatric patients), at the end of the section

"Collectively, these issues—limited pediatric data and labeling resources, ethical and legal concerns around algorithmic bias and opacity, and the predominance of adult‑trained models—form a common constraint that recurs across PIR‑relevant AI applications. Rather than being repeated in each subsection, these considerations should be considered as common backdrop: most of the tools and examples discussed in this review inherit the same risks of under‑representation, uncertain generalizability to small or atypical children, and the need for dedicated pediatric validation before routine use."

Section 3.1 (Emerging AI technologies), end of section: Clarifying extrapolation from generic pediatric AI domains

"At present, most literature on XAI, federated learning, and multimodal integration in pediatrics is related to cardiology, transplantation, perinatology, or neuro‑oncology, with little direct PIR‑specific validation; in PIR, these concepts should therefore be regarded as promising research directions rather than established clinical tools [4–6,14]. These models often assume disease phenotypes, imaging protocols, and outcome measures that differ from interventional workflows, thus their relevance to PIR is best viewed as conceptual—highlighting feasible architectures and collaboration models—rather than ready‑made solutions for interventional practice [4–6,14]."

Response: In light of the reviewer’s comment, the linkage between this external evidence and the concluding sentences have been slightly strengthened by explicitly citing the pediatric AI gap and the limited PIR research infrastructure where future research priorities are summarized.

Section 3.5 (Research priorities and collaborative studies), end of the section:

"Dedicated PIR collaboratives with central coordination, standardized protocols, and shared data/statistical infrastructure are key strategies for overcoming these barriers. Aligning PIR collaboratives with broader pediatric AI initiatives from radiology and pediatric societies could ensure that PIR needs are represented in funding, regulatory, and reimbursement frameworks. Prospective multicenter comparisons of AI‑augmented versus standard workflows will be essential to demonstrate real‑world benefits and guide implementation. In this context, the pediatric AI gap described by the ACR Pediatric AI Workgroup [10]—where only a small fraction of cleared AI tools are explicitly useful in children and most are trained predominantly using adult datasets—underscores our conclusion that AI for PIR remains technically feasible but rarely designed or validated specifically for pediatric endpoints [3,10,11,15]."

Clarifying the text when statements are aspirational (hypothesis‑generating) rather than evidence‑based would strengthen the scientific tone.

Response: The sections on multimodal integration, personalized treatment planning, robotics, and future collaborative frameworks deliberately extrapolate from adjacent pediatric domains (cardiology, transplantation, perinatology, and neuro‑oncology) using published data. However, explicitly labeling these extrapolations as forward‑looking will sharpen the scientific tone and avoid over‑interpreting the current evidence base [4–6,14]. To address this, selected sentences in Sections 3.2–3.5 have been revised and brief qualifiers (e.g., “conceptual,” “hypothesis‑generating,” “prospective but not yet demonstrated in PIR”) have been added so that readers can readily distinguish between established findings and proposed future applications, without altering the core conclusions.

Section 3.2, end of the first paragraph:

"When extrapolated from experience in pediatric cardiology, transplantation, perinatology, or neuro‑oncology, these examples are intended as hypothesis‑generating concepts for PIR rather than as evidence of fully validated interventional tools, and should therefore be interpreted as forward‑looking clinical opportunities [4–6,14]."

Section 3.5, end of the first paragraph:

"These research themes should be interpreted as hypothesis‑generating priorities for PIR, extrapolated from current experience in pediatric cardiology, transplantation, perinatology, and neuro‑oncology rather than from mature, PIR‑specific evidence, and are therefore presented as an aspirational framework to guide future endpoint‑driven studies [3–6,14,15]."

Response: I fully agree that the adopted approach is purposive rather than systematic and that, as a result, the cited studies skew toward illustrative, often favorable examples, with relatively little explicit discussion of negative or conflicting findings and limited coverage of AI‑based dose registries, real‑world post‑market surveillance, and granular economic analyses. In keeping with the reviewer’s suggestion and to maintain brevity, a revision in the Search strategy and selection subsection has been actioned to (i) explicitly state that many of the included studies are small, single‑center, or feasibility series, (ii) acknowledge that negative or neutral results and failed implementations are under‑represented in the current pediatric AI literature, and (iii) note that many of the PIR‑relevant concepts discussed in this review are extrapolated from adult or mixed cohorts, with relatively few large pediatric datasets and minimal registry‑ or economic‑level evidence. This clarification directly addresses concerns about the nature of the evidence base and the purposive selection.

Revision:

"1.1. Search strategy and selection

Some detailed comments on the sections and figures:

Figure 1B looks the same than 1A. This is the standard “hardware”. Please focus on the AI part. OptiqAI - great to mention that, but what is it and why is it relevant? What is the purpose of putting a Thorax Unit (1C) in here - if then put the fitting system in here —> https://www.usa.philips.com/healthcare/brand/azurion-image-guided-therapy-system? What about the units from GE that actually were first with AI enabled capabilities — https://www.gehealthcare.com/en/products/image-guiding-solutions?appId=aemshell — and of course CANON and UNITED IMAGING has

Response: In line with these comments, it is concluded that the AI capabilities of interventional platforms such as Siemens Artis/OptiqAI and GE Allia/AutoRight cannot be meaningfully conveyed by static images of C‑arms, and that the previous panels (including the Smart Collimation Thorax unit) risked over‑emphasizing generic hardware and non‑interventional equipment. Figure 1 has therefore been entirely removed, in turn shifting the emphasis to a textual description of AI imaging chains from multiple vendors. Specifically, OptiqAI is now explained as an AI‑powered imaging chain that combines real‑time denoising with big‑data–driven automatic exposure control across fluoroscopy, acquisition, and DSA, dynamically adjusting multiple exposure parameters to maintain the requested image quality at the lowest reasonable dose. In parallel, a short description of GE’s Allia platforms with the AutoRight intelligent image chain has been added, which uses artificial intelligence to optimize imaging and dose parameters in real time during fluoroscopy and acquisition to provide consistent, low‑dose image quality across a broad procedure mix (see the generated image above). The previous text related to Smart Collimation Thorax has been deleted, and all figures have been renumbered. These changes are believed to address the reviewer’s concerns by focusing on clinically relevant AI functions from more than one vendor, while avoiding hardware‑centric images that add little beyond what the text already conveys.

Revisions:

Section 2.1, paragraph describing current AI technologies in PIR (replacement of the sentences around the former Figure 1 reference) was amended according to the suggestion by the reviewer.

" Desai et al. highlighted PIR AI opportunities in scheduling, fluoroscopic dose reduction, outcome prediction, equipment selection, robotics, human–computer interaction, and education [3]. AI denoising algorithms have been applied for real time fluoroscopy, acquisition, and digital subtraction angiography (DSA), and big‑data‑driven exposure control that automatically optimizes the kV, mA, filtration, pulse width, focal spot, detector dose, and collimation as the C‑arm position or angulation changes [3,17,18]. In 2025, Siemens Healthineers introduced OptiqAI for Artis interventional platforms (genio/icono/pheno), an AI‑powered imaging chain that combines real‑time image denoising with big‑data–driven automatic exposure control across fluoroscopy, acquisition, and digital subtraction angiography, dynamically adjusting tube voltage, tube current, copper prefiltration, focal spot size, pulse width, detector dose, and collimation to maintain the requested image quality at the lowest procedure time and reasonable dose) [17].

In the image‑guiding portfolio of GE Healthcare, the Allia interventional platforms incorporate the AutoRight intelligent image chain, which uses artificial intelligence to optimize multiple imaging and dose parameters in real time during fluoroscopy and acquisition, aiming to maintain consistent image quality at the lowest reasonable dose across a broad range of procedures [18]."

The entire sentence beginning "The Smart Collimation Thorax (Philips) uses an AI algorithm…" and has been deleted and the former Figure 1 and its legend have been removed from the manuscript.
Subsequent figures and figure citations have been renumbered accordingly (former Figures 2–4 now Figures 1–3).

Figure 2 - what about the unit from INTUITIVE - https://www.intuitive.com/en-us/products-and-services/ion/how-ion-works

Response: This platform was no added because Ion is a robotic bronchoscopy / endoluminal system intended for peripheral lung biopsy within the tracheobronchial tree, rather than an endovascular navigation or interventional radiology robot, and it is not currently indicated for pediatric use. Given the focus on AI‑adjacent and AI‑enabled systems for vascular and image‑guided endovascular interventions, it is believed that including Ion would broaden the scope beyond the remit of this PIR‑focused review.

Figure 4 - I am not sure what the value of that figure is … the only thing that is a highlight are the paintings on the wall ... everything else is what a newly build state of the art interventional radiology suite should have.

Response: Figure 4 is deliberately framed as a fictional near‑future PIR suite whose purpose is to integrate, in a single pediatric scenario, the AI functionalities that are otherwise discussed separately in Sections 2.3 and 3 (AI‑assisted 3D vascular planning, intraprocedural vessel segmentation and target highlighting, real‑time dose dashboards, AI‑guided vascular access, and on‑table risk prediction and decision support), rather than to showcase novel hardware. The child‑friendly environment simply anchors the scene in pediatrics; the substantive content lies in the screens and interfaces, which conceptually aggregate dose‑sparing fluoroscopy, planning toolkits, AI‑enabled navigation, and outcome prediction into one coherent workflow. Because no current commercial system offers this complete, end‑to‑end AI workflow for children, a schematic, non‑branded illustration is considered more informative for readers than additional photographs of otherwise standard angiography suites.

Please check for minor typographical issues (spacing, hyphenation, reference formatting) that will improve readability, but there are no major language issues.

Response: The manuscript has been carefully re‑checked for spacing, hyphenation, and reference‑style consistency. Although minor typographic adjustments where made, no substantive language changes were required. The manuscript remains consistent with the Editage language‑editing certificate already provided.

Response: In line with the reviewer’s guidance, targeted revisions have been implemented to (i) make the narrative search strategy and purposive selection of studies more transparent, (ii) more clearly distinguish where evidence is derived from PIR‑specific versus adult or mixed cohorts, and (iii) streamline and explicitly label speculative or hypothesis‑generating elements, alongside revising the figures as suggested.

Reviewer 3 Report

Comments and Suggestions for Authors

Manuscript ID: diagnostics-4323802
Title: Current state and future of artificial intelligence in pediatric interventional radiology: A narrative review
Author: Abdulaziz Mohammad Al-Sharydah

A single-authored narrative review of the "Current state and future of artificial intelligence in pediatric interventional radiology" is presented.

This single-authored review manuscript consists of a non-structured abstract with keywords, 4 sections (introduction, current state of AI in PIR with 6 subsections, future directions with 5 subsections, conclusions) on 13 pages of single-spaced text with 4 embedded figures and 2 tables. A list of abbreviations, 37 references, and 5 URLs (in the figure captions) are provided.

There are many reports of pediatric interventional radiology in the published literature, but most do not include AI, so this is a timely and novel topic. The sole author does not have previous publications on AI in the bibliography or readily available on PubMed. Why a solo author?

This narrative review is based on a search process outlined in lines 58-66 on page 2. There is no information on the search terms used. What does "single clinically oriented framework" mean, and how is this relevant? This is not a systematic review, but no reason is given. Why not? What are "relevant society publications"? What keywords were used? How many articles were retrieved and included in this review?

"SNARA" appears in line 61 but is not included in the list of abbreviations on page 14, and no reference is cited. Apparently, this is a misspelling of SANRA - ref.: Baethge C, Goldbeck-Wood S, Mertens S. SANRA—a scale for the quality assessment of narrative review articles. Research Integrity and Peer Review. 2019;4(1):5. Please clarify.

No flowcharts are provided. The assessment of AI technology - current and future - is very limited. A few current products are summarized, but their assessment is primarily qualitative, consisting of a brief description of their features.

Figure 4 is a "fictional" AI-enabled pediatric interventional suite. How was this generated? Is there a "real" example? The components of this scene are not identified, and the methods used for its generation are not disclosed.

The sole author is a qualified interventional radiologist, but has limited experience with AI technology, and whose publications appear unrelated to AI. Why only one author? An AI-expert co-author would be a welcome addition.

Many AI technologies are not mentioned in this report, including transformers and foundation models, which are not discussed in detail despite their high impact on current AI development work. Large language models are not mentioned. Why not? Their impact on AI in healthcare is extremely high.

No timeline is shown. There are no details on how PIR‑specific performance was measured and compared.

In this single-author report, "we" is frequently used. Rather than change it to a first-person perspective, the author may wish to use a 3rd person impersonal perspective.

All of the figures depict equipment, but AI will probably have its greatest impact on process improvements. Diagrams, tables, and flowcharts that depict this aspect would be welcome additions.

Overall, a single-authored narrative review of AI in PIR includes an up-to-date literature review. The search strategy is not specified in detail. An AI-expert co-author and process-oriented AI improvement discussion would be welcome additions.

Author Response

Reviewer 3

Comments and Suggestions for Authors

Manuscript ID: diagnostics-4323802
Title: Current state and future of artificial intelligence in pediatric interventional radiology: A narrative review
Author: Abdulaziz Mohammad Al-Sharydah

A single-authored narrative review of the "Current state and future of artificial intelligence in pediatric interventional radiology" is presented.

Response: This article was conceived as a focused, clinician‑oriented narrative review written from the perspective of a practicing pediatric and adult interventional radiologist who also serves as institutional lead for AI‑enabled imaging and procedural technologies. My prior work includes peer‑reviewed publications on procedural software toolkits and advanced image‑guided workflows in interventional radiology (e.g. Al‑Sharydah et al., Diagnostics 2023;13:765), as well as multiple papers on technology‑driven optimization of interventional practice in pediatric and adult settings. The single‑author format reflects the narrative and conceptual scope of this review, not a lack of engagement with multidisciplinary expertise; the discussion draws on ongoing collaborations with diagnostic pediatric radiologists, physicists, and data‑science colleagues, as reflected in the diversity of cited sources. Because authorship structure does not alter the methods or conclusions, the manuscript text has not been altered. Co‑authorship in future empirical PIR–AI studies where team‑based approaches are essential would be welcome.

Response: This work was intentionally designed as a narrative, clinically oriented review rather than a formal systematic review, following SANRA guidance and reflecting the heterogeneity and early, largely feasibility‑level nature of PIR‑related AI studies. It is now clarified in Section 1.1 that narrative searches of MEDLINE/PubMed, Embase, and Scopus up to the first quarter of 2026 with combinations of terms such as “pediatric interventional radiology,” “pediatric cardiac catheterization,” “artificial intelligence,” “machine learning,” “deep learning,” “robotic,” “navigation,” “dose reduction,” and “risk prediction,” were used and that position statements and white papers from societies including ACR, SIR, SPR, CIRSE, and related pediatric subspecialty groups were additionally screened. It is stated that this purposive process yielded 70–80 primary and review articles and several society statements that were synthesized into a single clinically oriented framework covering diagnostic, interventional, technical, and policy perspectives. In addition, it has been noted that a formal systematic review with exhaustive enumeration and risk‑of‑bias assessment was beyond the intended scope of this concept‑mapping manuscript.

Revision:

"Consistent with published guidance on the structured reporting of narrative reviews, this work was designed as a narrative, clinically oriented review rather than a formal systematic review, given the heterogeneity and early, largely feasibility‑level nature of PIR–related AI studies. Narrative searches of MEDLINE/PubMed, Embase, and Scopus were performed up to the first quarter of 2026 using combinations of terms such as “pediatric interventional radiology,” “pediatric cardiac catheterization,” “artificial intelligence,” “machine learning,” “deep learning,” “robotic,” “navigation,” “dose reduction,” and “risk prediction,” and were supplemented by key guidelines, position statements, and white papers from professional societies including the ACR, the Society of Interventional Radiology (SIR), the Society for Pediatric Radiology (SPR), and the Cardiovascular and Interventional Radiological Society of Europe (CIRSE), as well as related pediatric subspecialty groups in cardiology, transplantation, and neuro‑oncology. English‑language, peer‑reviewed articles and conference papers describing AI or robotic methods with direct or conceptual relevance to PIR workflows in pediatric or mixed‑age populations were prioritized, whereas purely adult diagnostic studies without interventional implications, non‑peer‑reviewed material, and non‑medical technical reports were generally excluded. Candidate articles were screened by title and abstract, followed by full‑text review where appropriate. When overlapping publications were identified, the most recent, influential (higher‑impact), or more comprehensive was preferentially cited. This process yielded approximately 70–80 primary and review articles, along with several society statements, which were synthesized into a single clinically oriented framework spanning diagnostic, interventional, technical, and policy perspectives. Accordingly, this review should be interpreted as a hypothesis‑generating narrative overview rather than a systematic evidence inventory."

Response: The term “SNARA” in the original text was an inadvertent typographical error and should not have appeared as an unexplained abbreviation. Rather than anchoring the manuscript to a specific quality‑assessment tool, the intention was simply to indicate that the review follows established good‑practice principles for structured narrative reviews. The erroneous acronym has been removed and the sentence has been revised to read:

Response: The present manuscript was deliberately structured as a narrative, concept‑mapping review aimed at integrating pediatric‑specific challenges, technical developments, and research priorities into a coherent workflow‑oriented overview, rather than as a formal technology assessment or guideline. In this context, flowcharts were not included, because the available PIR‑specific evidence does not yet support robust algorithmic decision pathways or comparative “step‑wise” implementation schemes; most applications represent feasibility‑level or early clinical tools described in small series or mixed‑population studies. Structured qualitative assessment is provided through text, tables, and a conceptual figure: early AI products and platforms are summarized with their intended tasks, control paradigms, evidence setting, and AI contribution (Tables 1–2), and Figure 3 integrates these elements into a conceptual near‑future pediatric interventional suite. To address the reviewer’s concern more explicitly, it is now clarified in the Methods/Introduction that the review is narrative and hypothesis‑generating, and the wording in the “Current state” and “Future directions” sections has been revised to distinguish clearly between marketed products, investigational prototypes, and conceptual applications, including a brief statement on the limited availability of comparative outcome and economic data. Given the heterogeneity and immaturity of the evidence base, this structured narrative approach is believed to be more appropriate and transparent than adding flowcharts that could inadvertently overstate the maturity of AI‑driven decision pathways in PIR.

Revision:

Section 1.1 (Search strategy and selection), at the end of the existing paragraph.

"Given the heterogeneity, small sample sizes, and largely feasibility‑level nature of current PIR‑related AI reports, this work is presented as a structured, hypothesis‑generating narrative review with qualitative appraisal of available technologies, rather than as a formal technology‑assessment with flowcharts or comparative effectiveness algorithms."

Response: Figure 4 is deliberately presented as a fictional, schematic illustration rather than as a real installation, and its role is to integrate, in a single pediatric scenario, the distinct AI functions that are otherwise described separately in Sections 2.3 and 3 (AI‑assisted pre‑procedural 3D planning and path selection, intraprocedural vessel segmentation and target highlighting, real‑time radiation‑dose dashboards, AI‑guided vascular access, and on‑table risk prediction/decision support). The illustration was created by the author (using a digital illustration workflow) specifically for this review to act as a conceptual “map” of how these components could co‑exist in a near‑future PIR suite; to my knowledge, there is currently no single real‑world pediatric interventional room that incorporates all of these AI modules in the integrated manner depicted.

Revision

"Figure 3. Fictional artificial intelligence (AI)‑enabled pediatric interventional radiology suite. Concept illustration of a near‑future angiography suite, created by the author, in which multiple AI tools support the pediatric interventional radiology workflow from planning to access. The left‑wall screen depicts AI‑assisted pre‑procedural planning with three‑dimensional vascular modelling and an optimized catheter path; the central fluoroscopy display shows AI fusion with vessel segmentation and a highlighted target branch to facilitate real‑time targeting and tracking. Two small side monitors present AI‑driven radiation‑dose monitoring and procedural risk prediction, while AI‑enabled and robotic‑ready devices adjacent to the table provide intraprocedural navigation support. The right ultrasound screen illustrates an AI‑guided vascular access planner for femoral venous puncture. This figure is a schematic, non‑commercial integration of AI components already reported individually in the literature and is intended as a didactic, hypothesis‑generating depiction of a potential future PIR environment rather than as a representation of any specific existing clinical suite."

Response: This review was intentionally conceived as a single‑author, narrative, clinically oriented synthesis written from the perspective of a practicing pediatric and adult interventional radiologist, department chair, and institutional lead for AI‑enabled imaging and procedural technologies. In this role, the author is directly responsible for evaluating, implementing, and governing AI‑related tools in interventional practice, and has prior peer‑reviewed work on procedural software toolkits and advanced image‑guided workflows in interventional radiology (for example, Al‑Sharydah et al., Diagnostics 2023;13:765), as well as multiple technology‑focused articles in diagnostic and interventional radiology. Contemporary authorship guidance emphasizes that authorship should reflect substantial intellectual contribution and accountability rather than a minimum number of co‑authors, and there is a long tradition of single‑authored scholarly articles, particularly for narrative reviews, perspectives, and concept‑mapping pieces. In keeping with this, the present manuscript reflects the integrative view of a clinician who routinely bridges PIR, imaging technology, and AI implementation, while drawing extensively on the published work of AI methodologists and multidisciplinary pediatric teams cited throughout the article. As the manuscript is a conceptual, hypothesis‑generating review rather than an original AI methods paper, and given that its content has already benefited from institutional discussions with data‑science and engineering colleagues, single authorship is considered appropriate and transparent in this context; future empirical PIR–AI projects from this group will continue to be conducted and reported within formally multidisciplinary teams.

Response: This review was intentionally conceived as a single‑author, narrative, clinically oriented synthesis written from the perspective of a practicing pediatric and adult interventional radiologist, department chair, and institutional lead for AI‑enabled imaging and procedural technologies. Contemporary authorship guidance emphasizes that authorship should reflect substantial intellectual contribution and accountability, not a minimum number of co‑authors, and warns that excessive authorship can dilute responsibility and credit rather than enhance scientific quality.^1-4 In this spirit, there is a long‑standing and accepted tradition of single‑authored scholarly articles and narrative pieces in clinical journals, particularly for case reports, perspectives, and concept‑mapping or hypothesis‑generating reviews similar to the present work.^1-4 As a clinician who routinely bridges pediatric interventional radiology, imaging technology, and AI implementation, the author is well positioned to provide an integrated view while drawing extensively on the multidisciplinary work of AI methodologists and pediatric teams cited throughout the manuscript. Given that this article is a conceptual narrative review rather than an original AI methods or multicentre empirical study, and that its content has benefited from institutional discussions with data‑science and engineering colleagues, single authorship is considered appropriate and transparent in this context; future empirical PIR–AI projects from our group will continue to be conducted and reported within formally multidisciplinary teams.¹⁻⁵

Supporting references:

Har‐El G. Does it take a village to write a case report?. Otolaryngology–Head and Neck Surgery. 1999 Jun;120(6):787-8.
Shaffer E. Too many authors spoil the credit. Canadian Journal of Gastroenterology & Hepatology. 2014 Dec;28(11):605.
Wilcox LJ. Authorship: the coin of the realm, the source of complaints. Jama. 1998 Jul 15;280(3):216-7.
Rennie D, Yank V, Emanuel L. When authorship fails: a proposal to make contributors accountable. Jama. 1997 Aug 20;278(7):579-85.

No timeline is shown. There are no details on how PIR‑specific performance was measured and compared.

Response: AI in pediatric interventional radiology is at a genuinely nascent stage: as the review documents, there are only scattered feasibility reports, mixed adult–pediatric cohorts, and conceptual frameworks, with virtually no standardized, PIR‑specific performance datasets that would support a robust chronological timeline or formal comparative benchmarking. To this end, drawing a “timeline” of AI in PIR or tabulating performance across systems would be largely artificial and risk overstating the maturity and comparability of the available evidence. Instead, the manuscript is intentionally structured as a hypothesis‑generating, workflow‑oriented narrative that (i) maps current applications along the PIR care pathway, (ii) explicitly emphasizes the absence of rigorous PIR‑specific metrics, and (iii) defines dose, fluoroscopy time, anesthesia duration, complications, length of stay, and patient‑reported outcomes as the endpoints future studies must report to enable precisely the kind of temporal and quantitative analyses the reviewer suggests. In other words, the lack of a timeline or performance table is not a weakness of the review design, but a direct reflection—and explicit message—of how early and fragmented the PIR‑focused AI literature still is.

In this single-author report, "we" is frequently used. Rather than change it to a first-person perspective, the author may wish to use a 3rd person impersonal perspective.

Response: We thank the reviewer for this stylistic observation. The use of “we” in single‑author manuscripts has a long tradition in scientific writing, where it may denote the author together with the reader or, more generally, the research community engaging with the argument, and is considered a matter of style rather than correctness when context makes the intended meaning clear.¹⁻³

Contemporary guidance on authorial voice in academic prose increasingly supports the use of first‑person pronouns (“I” or “we”) when they improve clarity and avoid awkward passive constructions, noting that such usage is not grammatically wrong and can be particularly effective when guiding the reader through complex methodological or conceptual material.²⁻³

In this manuscript, “we” is used in this inclusive, reader‑oriented sense (for example, “we next consider…”), rather than to imply multiple authors, and it is consistently applied only to interpretive or explanatory statements, not to factual claims about who performed specific work.

Nevertheless, in light of the reviewer’s preference for a more impersonal tone, we are willing to revise selected instances to neutral constructions (e.g., “this review next considers…”, “the following section discusses…”) where this can be done without sacrificing clarity, and to retain “we” only where it clearly serves the rhetorical function of guiding the reader through the PIR AI framework. That said, the author ensured that changes were made to the manuscript to restrict the use of “we” as much as possible.

Supporting references:

Webb C. The use of the first person in academic writing: objectivity, language and gatekeeping. Journal of advanced nursing. 1992 Jun;17(6):747-52.
Tang R, John S. The ‘I’ in identity: Exploring writer identity in student academic writing through the first person pronoun. English for specific purposes. 1999 Dec 1;18:S23-39.
Rull V. Does solo publication still make sense?. EMBO reports. 2026 Feb;27(3):566-9.

All of the figures depict equipment, but AI will probably have its greatest impact on process improvements. Diagrams, tables, and flowcharts that depict this aspect would be welcome additions.

Response: As noted in response to earlier comments, the current version of the manuscript has already been revised to ensure:

Hardware‑centric content has been reduced (including removal of the original multi‑panel Figure 1).

The remaining figures—particularly the fictional AI‑enabled PIR suite—are explicitly described as conceptual, workflow‑oriented illustrations that integrate AI‑assisted planning, intraprocedural guidance, dose dashboards, and risk prediction into a single pediatric care pathway.

Process‑level effects of AI (triage, scheduling, protocol selection, planning toolkits, navigation assistance, and post‑procedural monitoring) are now emphasized primarily in the text, tables, and research‑priority sections, where they can be discussed with more nuance than is feasible in a simple flowchart.

Given these changes and the early, heterogeneous nature of PIR‑specific AI evidence, adding additional diagrams or flowcharts at this stage would risk over‑stating the maturity and prescriptiveness of AI‑driven PIR pathways without materially improving the reader’s understanding. The revised text, tables, and conceptual Figure 4 is believed to already address the core issue raised—that the major impact of AI in PIR is likely to be at the level of workflow and decision‑support rather than hardware alone—and no further modifications to the figure set are proposed.

Response: In the revised manuscript, a dedicated subsection detailing the databases, time frame, and key terms used in the literature search have been added, consistent with best practice for narrative reviews. The single authorship has also been respectfully maintained, as the work is intentionally written from the standpoint of a pediatric interventional radiologist translating AI concepts into PIR practice, while explicitly acknowledging the input of AI‑expert colleagues in the Acknowledgments. Finally, the discussion of process‑oriented AI applications has been strengthened—across triage, scheduling, planning, intraprocedural guidance, and post‑procedural follow‑up—in the “Implications for clinical practice” and “Future directions” sections, which is believed to address the reviewer’s request for more emphasis on workflow‑level improvements.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have revised the article taking all of the points mentioned into consideration, and the paper is acceptable in its current form.

Author Response

Decision Accept after minor revision Comments Thank you for this timely review article.
I am not sufficiently experienced in artificial intelligence to provide a decision on the manuscript. However, I would like to highlight the following critical issue that require revision by experts. The review places excessive emphasis on commercial influences, presenting patents and technologies while downplaying brands in figures. In particular, I suggest modifying the figures by limiting the display of commercial specifications and instead simply illustrating the underlying technological principles of the subject of the manuscript.
Please disclose conflicts of interest. Responce: The editor’s thoughtful and careful assessment of the manuscript, including the important points regarding commercial content and AI‑generated figures, is sincerely appreciated. In response, the legends of Figure 1 and Figure 2 have been revised to limit commercial specifications, de‑emphasize brand‑level details, and explicitly foreground the underlying technological principles they are intended to illustrate (e.g., AI‑adjacent robotic navigation, and AI‑enabled handheld ultrasound guidance), while retaining citations to the relevant peer‑reviewed or regulatory sources where appropriate. In addition, and in accordance with MDPI’s policies on the responsible and transparent use of AI tools, one conceptual figure is now explicitly noted as having been created with the assistance of a generative AI image tool and then manually edited by the author to ensure clinical, technical, and ethical appropriateness; the Acknowledgments section has been revised using the recommended template, and the Conflict of Interest statement updated to clarify that commercial systems are included solely for illustrative purposes, that no AI tools were used for data analysis or generation of the scientific narrative, and that the author has no financial or commercial relationships with the companies mentioned and assumes full responsibility for the integrity of the manuscript. With these targeted revisions and clarifications, the article is submitted in the strong belief that it now fully meets the scientific, ethical, and editorial standards for publication in Diagnostics.

Article Menu

Current State and Future of Artificial Intelligence in Pediatric Interventional Radiology: A Narrative Review

Further Information

Guidelines

MDPI Initiatives

Follow MDPI