Improving patient safety in diagnostic imaging. A systematic review of the rational use of ionizing radiation.

 

 

Nº de DOI: 10.34896/RSI.2026.15.38.001

 

 

AUTHORS

1. Angie Mabel Pesantez Morales. General Practitioner. Affiliated with Hospital San Marcos. Graduate of Universidad Técnica de Machala. Based in Pasaje, Ecuador. https://orcid.org/0009-0002-9371-9171
2. Brigette Valeria Pérez Guallichico. General Practitioner and Surgeon. Affiliated with Hospital Urgencias Médicas Tumbaco (HUMET). Graduate of Universidad de las Américas. Based in Quito, Ecuador. https://orcid.org/0009-0006-5457-4490
3. Ana Gabriela Caicedo Toro. General Practitioner and Surgeon. Affiliated with Hospital General Alfredo Noboa Montenegro. Graduate of Universidad Regional Autónoma de Los Andes. Based in Guaranda, Ecuador. https://orcid.org/0009-0003-5817-6770
4. Gissel Alejandra Izurieta García. General Practitioner. Affiliated with Centro de Diálisis Contigo Cendialcon CIA. LTDA. Graduate of Universidad Técnica de Ambato. Based in Latacunga, Ecuador. https://orcid.org/0009-0003-7248-2168
5. Dakmar Doménica Molina Maldonado. General Practitioner with a Master’s Degree in Community Nutrition and Dietetics. Affiliated with Hospital Básico SERMES. Graduate of Universidad de las Américas. Based in Latacunga, Ecuador. https://orcid.org/0009-0009-5298-0964

ABSTRACT

The review comprehensively reports on contemporary best practices and futurist trends in patient safety that will help challenge healthcare providers to balance the trade-off of diagnostic quality and radiation safety within the changing complexities of imaging.

KEY WORDS

Radiation protection, patient safety, diagnostic imaging, radiation dosage, dose optimization and artificial intelligence.

RESUMEN

Esta revisión presenta un informe exhaustivo sobre las mejores prácticas actuales y las tendencias futuras en materia de seguridad del paciente, con el fin de instar a los profesionales sanitarios a equilibrar la calidad diagnóstica y la seguridad radiológica frente a las complejidades cambiantes del diagnóstico por imagen.

PALABRAS CLAVE

Radioprotección, seguridad del paciente, diagnóstico por imagen, dosis de radiación, optimización de la dosis e inteligencia artificial.

INTRODUCTION

The rapid increase of the use of diagnostic imaging modalities has significantly changed clinical medicine, but this has been met with the equivalent increase in the public’s exposure to ionizing radiation, that warrants a focus on patient safety. This systematic review synthesizes the current best evidence on rational use of radiation, specifically as it relates to justification and optimization. The literature review, includes studies published over the past 10 years and indicated a significant change from the original ALARA-based model and more towards an ALADA-based model. The most notable point is the potential of technology to lower the radiation dose whilst maintaining diagnostic quality, particularly with the introduction of artificial intelligence (AI) and advanced deep-learning reconstruction algorithms, that have shown the potential to reduce performance dose by as much as 60%. We also purposefully reviewed the role of Clinical Decision Support (CDS) and Diagnostic Reference Levels (DRLs) as organizational tools to help patients avoid excess radiation. The review summarizes evidence on technological means to reduce dose within a safety culture, interdisciplinary communication, and patient engagement to obtain optimal safety outcomes.

OBJECTIVE

To systematically evaluate current strategies for radiation dose optimization and patient safety in diagnostic imaging, with emphasis on justification, dose reduction technologies, artificial intelligence, and organizational safety measures

METHODOLOGY

To ensure a robust and comprehensive synthesis of the evidence, a systematic search strategy was framed adhering to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. The search strategy was designed to identify peer-reviewed literature, clinical practice guidelines, and technical reports published between 2015 and 2026. The selected range was purposely envisioned to capture the most current advancements in technology, specifically the introduction of artificial intelligence in diagnostic imaging and advanced reconstruction algorithms.

The primary data sources comprised the four most common electronic databases: PubMed/MEDLINE, Scopus, the Cochrane Library, and Web of Science. The search strings consisted of a combination of Medical Subject Headings (MeSH) and related keywords, such as, «Radiation Protection», «Patient Safety», «Diagnostic Imaging», «Radiation Dosage», «Dose Optimization» and «Artificial Intelligence». Boolean operators (AND, OR) were used to structure and refine the search. An example structure is as follows:(Radiation Protection OR Patient Safety) AND (Diagnostic Imaging OR Tomography, X-Ray Computed) AND (Dose Optimization OR Justification)

Furthermore, a manual reference review from the selected international guidelines (International Commission on Radiological Protection (ICRP) and American College of Radiology (ACR)) and authoritative review papers was conducted to identify additional literature that satisfy the inclusion criteria. The search was limited to English language articles to ensure terminology and regulation frameworks were consistent in the analysis³.

The selection process involved inclusion and exclusion criteria which had been predefined to sustain relevance and quality of synthesized data. The defined inclusion criteria was satisfied if studies had to be:

1. Providing strategies for dose optimization, justification, or implement safety protocols in regard to ionizing radiation in diagnostic modalities, such as, Computed Tomography (CT), Interventional Radiology (IR) and Fluoroscopy.
2. Evaluating the use of technological interventions, such as AI-driven reconstruction, automated exposure control, or dose management software.
3. Evaluating organizational or human factors, such as safety culture or health education, or communication strategies.

Exclusion criteria removed studies that did not contribute to the overall objectives of the review, however, these studies might be interesting to explore further:

1. Studies that focused exclusively on non-ionizing modalities (e.g., MRI, Ultrasound), unless used as a comparative alternative for radiation reduction.
2. Case reports, editorials, and pilot studies with sample sizes that are not generalizable.
3. Studies that focused solely on therapeutic radiation (Radiotherapy) rather than diagnostic imaging.
4. Technical reports that were outdated and did not reflect current standards in clinical practice or hardware capabilities.

The preliminary screening phase commenced with reviewing titles and abstracts, followed by full-text evaluations to identify studies that met eligibility criteria. Any variations that arose within the selection process were reconciled via consensus within the research team.

Data extraction was conducted using a standardized electronic template developed to collateralize important data from each included study. Data points included the study aim, study design/methodology, imaging features, particular safety intervention (such as DRL implementation, AI algorithm), the primary outcome measure (such as dose reduction percentage or image quality metric), and major conclusions. The qualitative indices of radiation exposure are presented in this review, including Computed Tomography Dose Index (CTDIvol) and Dose-Length Product (DLP), to permit comparisons across protocols presented⁴.

The quality of the included studies was appraised through relevant frameworks, depending on the study design. The Risk of Bias in Non-randomized Studies of Interventions (ROBINS-I) and similar tools were used to evaluate clinical trials and observational studies. The technical evaluation of the AI models underscored validation methods, generalizability across multiple vendor platforms, and the clarity of «ground truth» benchmarks.

The review addressed the intersection of the study and the «system safety» perspective; this notion considers radiation protection as both a sociotechnical problem. This evaluation included the technical effectiveness of the dose-reduction tool implementation, as well as consideration of the organizational context in which it occurs. This includes user training, compliance with regulations (such as the European Medical Device Regulation), and a «Just Culture» within the radiology department as essential elements of evidence synthesis⁵. This multi-dimensional perspective comprehensively demonstrates how the results from studies reflected the real-world implications to improve patient safety in varying clinical settings.

RESULTS

The rapid uptake of advanced diagnostic imaging as part of routine clinical practice has dramatically enhanced the accuracy of disease detection and monitoring of treatment response. However, the preceding progress has, conversely, raised significant issues associated with the management of ionizing radiation exposure. The biological effects of radiation (i.e., deterministic and stochastic risks) should be treated with caution and care within the clinical setting. Furthermore, we should be mindful that patient safety in radiology is not a technical obligation but ethical consideration that crosses the imaging “spectrum” including requests for imaging, patient experience prior to scanning, and finally the reporting of the imaging results.

The concept of rational use is the primary objective of modern radiation protection, which requires justification of any exposure to ensure a net benefit to the patient and then the dose should be optimized to be as low as reasonably possible for the purpose of a definitive diagnosis. Historically, the concept of ALARA has been used as the guiding principle of rational use; however, imaging physicians are beginning to practice based on the basis of ALADA, which has more emphasis on the goal of diagnostic sufficiency rather than the exclusive goal of «as low as reasonably achievable dose.» This represents a paradigm shift where it is understood that the imaging task is becoming ever more complex and in some situations, inadequate image quality with excessive dose reduction could lead to a misdiagnosis and therefore an alternative patient safety risk.

To further this discussion, we find ourselves at a unique time in health care with a moment in time where medical physics, clinical radiology, and computational science have come together in convergence. AI presented unique opportunities for automated dose modulation and image denoising, where there is the opportunity for the dichotomy between radiation dose and image quality to be delinked¹. At a minimum, the establishment of global safety standards and regional DRLs has provided an opportunity for a performance metric, which has enabled the identification and corrective action of outlier performance in radiation delivery.

While there has been progress, obstacles remain for universal adoption of safety standards. Equipment variability, differences in institutional safety cultures, and levels of health literacy by both providers and patients lead to inconsistent radiation practice. Therefore, through this systematic review, we aimed to address issues by integrating the most current literature and practice on effective strategies in radiation protection. Through the evaluation of technology and organizational structures, we strive to represent how to advance patient safety through the judicious practice of ionizing radiation in health care in the current day².

Core Principles of Radiation Protection and Rational Utilization of Radiation Exposure:

The Principle of Justification Pertinent to Clinical Practice:

The rationale of justification is the first and arguably most salient area of radiation protection in diagnostic imaging. The operational premise is that no imaging procedure involving ionizing radiation should be performed unless a substantial net benefit is given to the patient, that outweighs potential harm. It can be inferred that justification for obtaining an imaging procedure, including the modality or advancement, is a shared responsibility between the referring physician and the radiologist. To meet this premise, the referring physician is responsible for determining if the clinical question can be answered with an imaging procedure, while the radiologist is responsible for the justification of the selected modality as the best option, and lowest risk⁶.

One factor that threatens to compromise the effective justification is what is sometimes termed «defensive medicine,» where imaging is ordered to mitigate the risk of a medical malpractice lawsuit rather than addressing an explicit clinical question. This practice can lead to an increase in the use of CT exams and the use of higher dose procedures in a context of emergency. One long-term countermeasure to this practice is the incorporation of Clinical Decision Support (CDS) systems. CDS systems provide guidelines that improve the ability to deliver evidence-based practice at the point of care to help decide which imaging workflow is appropriate. For example, using ACR Appropriateness Criteria within an electronic health record (EHR) may reduce the number of unnecessary scans through the recommendation of a non-ionizing imaging exam, such as ultrasound or MRI in case of clinical indication.

Justification also has a longitudinal element based on the imaging history of the patient. With the prevalence of chronic diseases requiring serial follow-up comparisons, the accumulation of dose must be analyzed. In this case, justification requires a determination of whether the prior imaging is sufficient, or if the new exam will provide additional actionable information. «Evidence-Linked Radiology» is being proposed for consideration to help reform the justification process by embedding structured clinical data in every requesting imaging exam⁷.

Optimization of Protection and the ALARA Concept:

When a procedure is justified, the next immediate focus is to optimize, which has been defined as the practice of being based on the ALARA (As Low As Reasonably Achievable) principle. Optimization does not mean obtaining the lowest possible dose, rather it means obtaining the dose necessary to achieve the diagnostic purpose. The importance of this distinction is paramount, as an under-dosed image that is deemed to be non-diagnostic is unsafe, i.e., the patient has been subjected to radiation with no clinical benefit and thus may require repetition².

The physics of optimization is a balance of some complex factors among radiation dose, image resolution, and the image noise. The basic principles of medical physics dictate that if the photon flux (mAs) or energy (kVp) is decreased the resulting radiation dose will also decrease while image noise will increase⁸. An example of optimized imaging would include current use of advanced hardware and software to take advantage of the physics optimization balance. As an example, Automatic Exposure Control (AEC) systems have been developed using machine learning to assess in real-time the patient’s size and tissue density and modulate the radiation output to maintain consistent output of high-quality images to various patient populations.

ALARA has undergone significant evolution which has led to a proposal of ALADA (As Low As Diagnostically Acceptable). There is now addition of the concept that the dose of the imaging exam should be considered, based on the diagnostic purpose. For example, a CT exam for detection of kidney stones does not require as much detail, meaning more noise and lower dose, as a CT angiography of a body part requiring complex vascular mapping. There has been recent literature examining the idea of «End2end-ALARA» framework that suggests machine learning could be used to optimize imaging across imaging and reconstruction to allows a personal dose while maintaining a stable image quality indicator across patient populations⁹. The movement toward optimization by purpose is a developed approach to radiation protection principles from looking at «one dose fit all» case.

Use of Diagnostic Reference Levels (DRLs):

Diagnostic Reference Levels (DRLs) provide a usable method for optimization of radiation protection in medical imaging. DRLs are not to be taken as specific dose limits for a given patient; instead they signal when a facility or procedure is delivering an unusually high or low dose of radiation. DRLs are generally defined at the 75th percentile of the distribution of observed doses in standard-sized patients, across several facilities. If a facility’s dose on average exceeds the DRL, this triggers a formal review of imaging procedures⁴.

The use of DRLs has been key to consistently managing the delivery of radiation internationally. In Computed Tomography, the key indices for the DRLs are Volume CT Dose Index (CTDIvol) which measures the radiation beam intensity and the Dose-Length Product (DLP) which describes the total energy deposited into the patient. As contrast-enhanced examinations are commonly performed for chest, abdomen and other imaging, and typically lead to increased exposure from multi-phase scanning, DRLs help manage these additional exposures.

There is also growing interest in using adaptive DRLs which account for patient size and weight. Traditional DRLs are defined in relation to a «standard» 70kg patient – which is often not provide suitable benchmarks for pediatric or obese populations. Institutions can develop localized, size-DRLs by utilizing Dose Management Systems (DMS) which will automatically record and analyze exposure data from thousands of exams. This data can be used to drive continuous quality improvement since optimization efforts can be targeted to the procedures or facilities where they are needed⁴.

Technological Solutions to Reduce Dose:

CT Protocol Innovations:

Computed Tomography is the largest contributor to medical radiation, establishing it as the primary target for technological dose reduction solutions. With this in mind, a major development in this field has been the transition from Filtered Back Projection (FBP) to Iterative Reconstruction (IR) and now Deep Learning Reconstruction (DLR). FBP is mathematically straightforward but is sensitive to noise. In comparison, IR algorithms iterate many loops of calculations to refine the image and subsequently eliminate noise without diminishing structural detail of the scan. This active imaging development enables scans to be acquired at considerably lower tube currents².
DLR builds on this by using neural networks that were trained on high-quality, full-dose «ground truth» images. These models can identify and remove noise patterns in ultra-low dose scans with reductions in dose by as much as 60% while maintaining image quality, or better than standard protocols⁴. Another hardware-based innovation is the arrival of Photon-Counting CT (PCCT), which utilizes a detector system based on counting single X-ray photons and also determining energy of photons in contrast to the standard, and common system, of energy integrating detectors. The advantages are negative factors of high spatial resolution, decreased electronic-noise, and the ability to perform bio-efficient imaging in multi-energy (spectral) imaging.

Protocol optimization involves the utilization of high-pitch scanning and «flash» modes, which are specifically useful in the radiology of the heart and pediatric imaging. Both of these modes function by extending the gurney mover that the patient of interest is on as the table moves in relation to the gantry rotation, which also decreases the amount of time the patient is exposed to radiation dose, and decreases the amount of motion-based artifacts. By taking the high-pitch scanning mode, or flash mode, continues with the automatic kVp selection, the increases the kVp that the scanner will deliver the dose in order to optimize the radiation dose to the patient, which is suited carefully to the body habitus of that patient, for the patients interest.

Dose Optimization in Interventional Radiology and Fluoroscopy

Interventional Radiology (IR) and fluoroscopic-guided procedures, must provide distinct safety challenges with respect to the potential long-term, over long durations, exposure time. The exposure times and proximity of the medical staff to the radiation beam create a unique risk for deterministic effects (skin injury) of complex and long procedures. The optimization for IR is about dose reduction, or to monitor the dose with optimization.

For the technology of IR, this includes pulsed fluoroscopy – versus guaranteed continuous exposure. By pulsing the frame rate down to sensible frames per second, you are maximizing the total dose without losing the temporal resolution required when advancing catheters and needles. Newer IR suites are also «dose-aware» with improved displays that give the interventionalist real-time feedback on cumulative air kerma, and dose-area product (DAP). The physician can make a change in the technique, like «spin» the C-arm or, maybe add more collimation to the field, to ensure that they are carefully monitoring, and not exceeding, the thresholds for safety.

Lastly, to reduce «scout» shots and long exposures due to fluoroscopic time, routine and common applications of 3-D roadmap technology, and bioplastic imaging (doing fluoroscopy while overlaying pre-procedural CT or MRI), can support less fluorescent time. This way the intervention is more accurate and complete in a shorter time, with a lower dose to the patient, and a lower dose to the aid team⁶.

The Role of Artificial Intelligence in Exposure Control:

Artificial Intelligence (AI) is quickly becoming an important component of the radiation safety ecosystem, moving from image reconstruction to the entire imaging chain. In the pre-acquisition phase, AI auto-positioning systems use 3D cameras to establish the patient’s body habitus and ensure that the patient is centered in the gantry. As miscentering is a common cause of unnecessary radiation exposure by the interfering of AEC system accuracy, by automating the process AI ensures that the optimization hardware is functioning at a high level of efficiency.

In the acquisition phase AI algorithms can produce dose modulation in real time. The system analyzes the initial scout image with greater accuracy than previously available, predicting the best dose for each slice based on the patient’s anatomy and suspected pathology. Intelligent exposure control is especially useful in pediatric radiology, where radiation sensitivity is higher and there is higher variability of anatomy^10.

In the post-acquisition phase, AI can play an important role in quality assurance and dose auditing. AI powered dose management systems can automatically scan records of thousands of doses to identify «outlier» events, where the DRL was exceeded. These systems can also relate the dose to image quality, establishing a feedback loop that can be used to inform radiologists of necessary protocol adjustments. While the benefits of AI can be useful in safety-critical capacities, there are also risks to consider. Two examples are the «hallucination» of AI (generating artifacts that appear like real anatomy) and algorithmic bias, which require thorough validation of processes, along with human judgement^1.

The use of large language models (LLMs) in radiology reporting can also indirectly affect patient safety by ensuring that findings are clearly communicated, and that follow-up recommendations are in alignment with the appropriate guidelines, preventing the «cascade» of unnecessary repeat imaging¹¹. As AI further evolves towards multimodal systems that will ingest imaging, genomic and clinical data, the goal of truly personalized radiation protection (optimizing radiologic doses based on an individual’s biological susceptibility to radiation), becomes an achievable aim^12. However, this will require an approach based on «system safety,» which combines technical reliability with ethical governance and human involvement^5.

Clinical and Organizational Strategies for Patient Safety:

Ensuring patient safety in diagnostic imaging involves more than just technical changes to radiation parameters: this requires a solid set of clinical and organizational strategies to mitigate inherent risks. Risk management in medical imaging aims to address the conflicting tension between the proven diagnostic benefits of ionizing radiation and the possible biological risk of injury, particularly in vulnerable populations such as children and pregnant women¹³. Although new hardware and software solutions provide the means of lowering patient exposure, the organization context—policy, culture and decision support tools—will dictate whether these tools are applied in practice. Managing effective safety applies a nuanced multi-faceted plan that includes the application of advanced technology, such as iterative reconstruction and artificial intelligence (AI), coupled with adherence to safety protocol practices and continued education for staff¹³. In addition, it is essential that hospital or institutional policies prioritize the routine maintenance and calibration of equipment to prevent unnecessary patient exposure due to breakdown in the technology used¹³.

Implementation of Clinical Decision Support Systems:

Clinical Decision Support Systems (CDSS) are an essential strategy of intervention in the «upstream» phase of radiation protection as it concerns the principle of justification. By embedding evidence-based guidelines into the ordering pathway of imaging, the CDSS assists clinicians in navigating through the complexities of modality selection while ensuring the chosen imaging method offers the greatest diagnostic yield with the least possible risk. The implementation of AI into CDSS has revolutionized the domain from a static rule-based alert to a dynamic data driven decision model. Diagnostic Decision Support Systems (DDSS) (a type of CDSS) have improved diagnostic accuracy while providing a safer triaging process when implemented using simulated or clinical environments¹⁴.

In the context of radiation oncology and high-dose imaging procedures, AI CT DSS will improve the safety and efficiency while enhancing accuracy for treatment planning and treatment delivery¹⁵. Reducing variability in the process of image reconstruction and dose calculation improves quality and standardization across the healthcare institution¹⁵. In addition, the integration of digital health technology now allows for the more effective monitoring of cumulative radiation exposure alongside planning of dosage, while establishing a longitudinal perspective of a patient’s radiological history to better informed future justification decisions¹³. The use of these systems are not without risk: clinicians must remain keenly aware of potential biases or «hallucinations» provided by AI that lead to incorrect diagnostic meaning-making and more broadly to patient harm¹⁴.

Education and Safety Culture Among Healthcare Providers:

Developing a «safety culture» is essential for justification of regional use of radiation. A safety culture is a shared commitment to safety at all levels of the organization, including clinical, bedside providers through to senior hospital administrators. Ongoing education on the safety aspects of radiation use is a crucial component to ensure staff are educated and practice dose optimization methods; and used of personal protective equipment (PPE) properly¹³. In training and education, more recently medical education has developed the use of healthcare simulation (HCS) and AI-based virtual reality platforms to facilitate personalized coaching and self-directed learning while help establish AI literacy in medical graduates¹⁴.

To further close the gap between the field of technical development and safety engineering, several experts emphasize the importance of the use of «Safety Factories»¹⁶. This means that safety tooling and methods are integrated directly into the development pipelines of the software that operates the imaging equipment itself, and ensure that testing and validation of safety-critical functions are performed through automatic checks for consistency¹⁶. Improvements in organizational safety can also be achieved through modular multi-workflow digital platforms for incident management. For instance, digital mechanisms which automate reporting and investigation of, for example, safety incidents have shown measurable improvements in workflow traceability and response times to the incident, which would directly correspond to managing near-misses or overexposures related to radiation¹⁷.

The treating of modern imaging is becoming increasingly complex, for example, magnetic resonance-guided radiotherapy and advanced cone beam CT require a highly trained workforce capable of adapting to rapid adaptive workflows¹⁸. In order to achieve the potential of these transformative technologies, workforce requires not only technical training but also quality assurance and continuous monitoring of performance¹⁸.

Communication of Risks of Radiation to Patients:

Communication of the risks of radiation is a critical component of patient-centered care and informed consent. The difficulty is in translating complex dosimetric information into a communication that the patient understands and can engage in shared decision making. AI and digital health technologies are emerging as valuable tools in this regard. For example, the development of Digital Twins–dynamic virtual representations of the patient–help clinicians simulate disease progression and treatment responses, provides the patient with an individualized basis upon which to discuss the risks and benefits of specific imaging or therapeutic interventions¹⁹.

In oncology, AI-based digital twins improve predictive ability and can lead to individualized treatment strategies, giving the patient less abstract language for understanding the idea of radiation risk²⁰. This individualization may be more important in the administration of therapies that are considered new, e.g., targeted alpha-therapy or photodynamic therapy, in which the risk-benefit profile differs considerably compared to radiation therapies that are established²⁰. For further consideration, while these benefits accumulate through technology, the communication piece must also consider the ethical and legal implications regarding AI, particularly in relation to data privacy and accountability¹⁹. This is necessary with the understanding of how the patient’s data is being used and how the AI is determining decision making, maintaining trust in a medical imaging experience.

DISCUSSION

The synthesis of present research demonstrates a paradigm shift in diagnostic imaging, moving away from measuring simple reductions in dose, to a comprehensive model of «precision safety». This model engineering technology innovation, organizational discipline and, personalized clinical interpretation. The evidence supports that although AI and the advanced hardware provide the channels for reducing doses of radiation, the effectiveness of achieving safety doe rely on the seamless integration of these into the clinical workflow.

Dose management programs using artificial intelligence (AI) and deep learning capabilities have been shown to perform significantly better than the traditional manner of service. The use of AI for low-dose imaging in CT and X-ray, especially, has demonstrated significant improvements in noise reduction, artifact elimination, and supervised dose reductions without sacrificing diagnostic performance²¹. These technologies are valuable because they enhance image quality and address efficiency as well by minimizing the necessity to repeat scans, which reduces both the cumulative dose to the patient and their overall satisfaction²¹.

These programs have become even more robust and effective through the use of data-driven safety monitoring. Utilizing statistical models to predict future states and classify risks is a common method. This is analogous to the conformal safety monitoring used in high-uncertainty environments such as flight testing. Safety monitoring in radiology departments can allow for situations to be identified as potentially unsafe before they occur²². Likewise, geometric analysis frameworks can be used to capture the spatial aspect of safety in more complex clinical environments in a quantitative manner²³. These interdisciplinary synergies reinforce that the most effective dose management programs are those that treat radiation safety as a dynamic process that allows for real-time monitoring rather than an audited retrospective perspective.

Many barriers exist even with the recognized benefits of the modern safety strategy. First, there are the problems related to technology including model generalization and performance limitations due to computational costs associated with AI. These may impede implementation and clinical adoption, especially for smaller or less-resourced hospitals²¹. In the area of 3D-printed medical devices designed for radiation therapy, there are challenges related to material variability and that there is no workflow standard, thus complicating efforts to transition from research to large-scale clinical validation and implementation²⁴.

Second, ethical and regulatory barriers exist. The characteristics of some AI algorithms may fall under the umbrella of being «black boxes,» causing uncertainty about explainability and accountability when AI is involved in making diagnostic judgments (20). Ensuring compliance with continuously evolving regulatory frameworks (e.g., FDA guidelines on additive manufacturing or EU changes to data privacy) may be difficult for institutions and require administrative resources¹⁹. There may be, and are examples of, «digital divide» problems in which safety technologies are at only elite institutions and the inequities of safety and quality may (will) also exist across populations¹⁸.

FUTURE DIRECTIONS

The future of radiation safety may rest in the transition of human-centered population-based standards to personalized dosimetry. The convergence of future-directed multi-omics data with functional imaging and AI-driven predictive models will facilitate adaptive and biological informed dose delivery¹⁸. Digital twins of our technology would be synonymous with the future of this direction allowing for real-time, personalized simulation of radiation effects to an individual patient²⁰.
Additionally, emerging trends include expanding AI applications to fluoroscopy and positron emission tomography (PET), as well as hybrid AI systems that combine post-processing with real-time data acquisition²¹. The relationship between 3D printing and AI-driven innovations – including automated treatment planning and real-time quality assurance – will likely also become the foundation of next-generation radiotherapy²⁴. The medical community will need to focus on disciplined validation of these technologies in multicenter studies and the development of standard quality assurance protocols that can keep pace with these rapidly evolving technologies¹⁸.

CONCLUSION

Improving patient safety within diagnostic imaging will necessitate a comprehensive, integrated strategy permitting the diagnostic benefit of ionizing radiation, while minimizing ionizing exposure. The systematic review in this journal has demonstrated that while we have the technical capabilities – in the form of AI, advanced reconstruction algorithms, etc. – that can facilitate a considerable reduction in dose, the success of these interventions will depend on the strength of clinical and administrative systems. Implementing Clinical Decision Support Systems, cultivating a proactive safety culture, and utilizing personalized interventions such as Digital Twins will be critical to assuring the rational use of radiation.

The advancement of universal safe excellence, however, will continue to be challenged by technical, ethical, and regulatory hurdles. This cannot be achieved without a multi-disciplinary partnership including healthcare providers, medical physicists, researchers, and regulators. Ultimately, personalized dosimetry and the implementation of digital health technology will allow medical imaging to move to a future where every radiological examination is individualized to the biological profile and clinical scenario for the patient, resulting in an approach to safety and care that is second to none.

REFERENCES

1. Magomedova Z, Pershina ES, Daneshvar K, Nöldge G, Mosler F. Human and artificial intelligence in radiology: current status, evidence, regulation, and future perspectives. Swiss J Radiol Nucl Med. 2026;27(1):1-28. doi:10.59667/sjoranm.v27i1.14.
2. Dudhe SS, Mishra G, Parihar P, Nimodia D, Kumari A. Radiation dose optimization in radiology: a comprehensive review of safeguarding patients and preserving image fidelity. Cureus. 2024;16(5):e60846. doi:10.7759/cureus.60846.
3. Tołkacz A, Kwiatkowska A, Gewald M, Król A, Sobczak J, Kosikowska A. Health education, communication and safety culture in diagnostic imaging: a narrative review of up-to-date guidelines. J Educ Health Sport. 2026;88:69433. doi:10.12775/JEHS.2026.88.69433.
4. Singh SJ, Jindal V. Optimization and assessment of CT dose indices (CTDIvol and DLP) in contrast-enhanced chest and abdomen CT: integrating diagnostic reference levels for patient safety. Int J Sci Res. 2025;14(11):1756-1763. doi:10.21275/SR251126125832.
5. Dobbe R. AI safety is stuck in technical terms: a system safety response to the International AI Safety Report [preprint]. arXiv. 2025. doi:10.48550/arXiv.2503.04743.
6. Alanazi NNF, Alenzi KA, Faraj AFM, Alaianizi AA, Alanazi AAL, Alrawaily AMQ, et al. Radiation exposure prevention strategies in diagnostic imaging: a review of nursing and radiology roles. International Journal of Computational and Experimental Science and Engineering. 2024;10(4):2990-2997. doi:10.22399/ijcesen.4428.
7. Kazemzadeh H, Naderi K. Evidence-linked radiology reporting: a human-supervised reference architecture for structured imaging intelligence [preprint]. arXiv. 2026. doi:10.48550/arXiv.2605.25120.
8. Dhayie AK, Jaein MF, Ktab AH, Zayd HEAE, Kamel AAM, editors. Fundamentals of medical physics: imaging, radiation, and applications. 1st ed. Delhi: Bright Sky Publications; 2025. doi:10.62906/bs.book.319.
9. Tao X, Lin L. End2end-ALARA: approaching the ALARA law in CT imaging with end-to-end learning [preprint]. arXiv. 2025. doi:10.48550/arXiv.2504.06777.
10. Shelmerdine SC, Naidoo J, Kelly BS, Laborie LB, Toso S, Akinci D’Antonoli T, et al. AI implementation in pediatric radiology for patient safety: a multi-society statement from the ACR, ESPR, SPR, SLARP, AOSPR, SPIN. Pediatr Radiol. 2026;56(2):243-256. doi:10.1007/s00247-025-06386-0.
11. Bluethgen C, Van Veen D, Zakka C, Link KE, Fanous AH, Daneshjou R, et al. Best practices for large language models in radiology. Radiology. 2025;315(1):e240528. doi:10.1148/radiol.240528.
12. Parvin N, Joo SW, Jung JH, Mandal TK. Multimodal AI in biomedicine: pioneering the future of biomaterials, diagnostics, and personalized healthcare. Nanomaterials (Basel). 2025;15(12):895. doi:10.3390/nano15120895.
13. Adelodun MO, Anyanwu EC. Comprehensive risk management and safety strategies in radiation use in medical imaging. International Journal of Frontiers in Medicine and Surgery Research. 2024;6(1):47-63. doi:10.53294/ijfmsr.2024.6.1.0039.
14. Hamilton A. Artificial intelligence and healthcare simulation: the shifting landscape of medical education. Cureus. 2024;16(5):e59747. doi:10.7759/cureus.59747.
15. Zafar F, Vilsan J, Mani S, Al Yousif AR, Cano-Reyes SE, Abraham G, et al. AI in radiation oncology: a comprehensive review of current applications and future directions. Cureus. 2025;17(9):e92964. doi:10.7759/cureus.92964.
16. Cârlan C, Ratiu D, Wagner M. Safety factories: a manifesto [preprint]. arXiv. 2025. doi:10.48550/arXiv.2509.08285.
17. Kamble S, Dangle A, Khurud G, Kendre O, Bhatt S. Integrated digital management system for railway workshops: a modular multi-workflow architecture for machine, permit, contract, and incident management [preprint]. arXiv. 2026;2605.00840.
18. Yan Y, Alexander DA, Bednarz BP, Bronk LF, Chen H, Gladstone DJ, et al. Innovative approaches in precision radiation oncology: advanced imaging technologies and challenges which shape the future of radiation therapy. Front Med (Lausanne). 2025;12:1686593. doi:10.3389/fmed.2025.1686593.
19. Giansanti D, Morelli S. Exploring the potential of digital twins in cancer treatment: a narrative review of reviews. J Clin Med. 2025;14(10):3574. doi:10.3390/jcm14103574.
20. John A, Alhajj R, Rokne J. A systematic review of AI as a digital twin for prostate cancer care. Comput Methods Programs Biomed. 2025;268:108804. doi:10.1016/j.cmpb.2025.108804.
21. David-Olawade AC, Olawade DB, Vanderbloemen L, Rotifa OB, Fidelis SC, Egbon E, et al. AI-driven advances in low-dose imaging and enhancement: a review. Diagnostics (Basel). 2025;15(6):689. doi:10.3390/diagnostics15060689.
22. Feldman AO, Harp DI, Duncan J, Schwager M. Conformal safety monitoring for flight testing: a case study in data-driven safety learning [preprint]. arXiv. 2025;2511.20811. doi:10.48550/arXiv.2511.20811.
23. Xu Z, Wang Z, Li H, Yu D, Yang Z, Wang J. A geometric analysis-based safety assessment framework for marine vehicle route decision-making [preprint]. arXiv. 2025;2501.06670. doi:10.48550/arXiv.2501.06670.
24. Pinckney R, Parupelli SK, Sandwall P, Chang S, Desai S. A review of 3D-printed medical devices for cancer radiation therapy. Bioengineering (Basel). 2026;13(1):115. doi:10.3390/bioengineering13010115.