Governing AI-Assisted Fracture Detection on X-Rays in Viet Nam: Missed Fractures, Radiology Report Reconciliation and Post-Discharge Follow-Up

Governing AI-Assisted Fracture Detection on X-Rays in Viet Nam: Missed Fractures, Radiology Report Reconciliation and Post-Discharge Follow-Up

IMPLEMENTATION INSIGHTS

Last reviewed: 31 October 2026

A woman arrives at an emergency department in Viet Nam after falling and injuring her wrist. She has pain, swelling, and difficulty using her hand. The initial X-ray does not show an obvious fracture, and an artificial intelligence (AI) application used to support image interpretation does not identify a suspected fracture either. After considering the examination findings and available imaging, the clinician decides that she can return home with advice about symptom management and follow-up.

Later that day, a radiologist completes the formal X-ray report and identifies a subtle fracture that was not recognized during the initial assessment. Depending on its location and characteristics, the finding may change the recommended treatment. The patient might need immobilization, additional imaging, or orthopedic evaluation.

By then, she has already left the hospital. The radiology report may have reached the hospital's information system, but that does not necessarily mean anyone has reviewed what it means for her care.

Who recognizes that the report differs from the assessment made before discharge? Who decides whether the patient needs to return? And if the hospital contacts her, how does the team know whether she has understood the new information and received the recommended care?

This hypothetical situation illustrates a patient-safety problem that cannot be solved by improving fracture detection accuracy alone. AI may help healthcare professionals recognize abnormalities during the initial assessment, but it cannot ensure that a later radiology finding reaches the right clinician or leads to appropriate follow-up.

For healthcare organizations in Viet Nam, responsible implementation requires attention to the entire pathway, from initial X-ray interpretation and discharge decisions to finalized radiology reports, clinically significant discrepancies, patient communication, and subsequent care.

Why missed fractures remain a patient-safety concern

Musculoskeletal injuries are common reasons for emergency department and urgent care visits. X-rays are frequently used to investigate suspected fractures, but interpretation can be challenging when abnormalities are subtle, anatomical structures overlap, or image quality is less than ideal.

Some fractures are readily visible. Others may involve small cortical disruptions, minimally displaced fragments, or complex joint anatomy that makes the abnormality difficult to recognize. Patient positioning, the views obtained, the anatomical region, and the reader's experience can all influence interpretation.

The clinical presentation remains important. A patient with focal bony tenderness, swelling, reduced function, or a concerning mechanism of injury may still require further assessment even when the initial radiographs appear normal.

Some fractures are not adequately visible on early X-rays, while others are visible but overlooked. These situations may lead to similar subsequent care needs, but they do not necessarily represent the same diagnostic problem or professional error.

Recognizing a fracture later does not automatically establish that the original clinician made an incorrect decision. The information available at the time, the patient's clinical findings, the reasonableness of the initial assessment, and the applicable standards of care all need to be considered.

The consequences of delayed recognition also vary. Some stable fractures can be managed conservatively without significant long-term harm. Others require timely immobilization, specialist assessment, or additional treatment to reduce the risk of displacement, impaired healing, or functional complications.

From a patient-safety perspective, the important question is not whether every fracture identified later should be classified as a diagnostic error. It is whether important new information is recognized, assessed appropriately, and incorporated into the patient's care.

That distinction becomes particularly relevant when AI is introduced. A positive AI result does not establish that a fracture is present, and a negative result does not reliably exclude every clinically important injury.

What AI-assisted fracture detection can and cannot do

AI fracture detection applications analyze X-ray images to identify patterns that may indicate a fracture. Depending on the product, the software may highlight a suspicious region, display a confidence score, or classify an examination as more or less likely to contain a fracture.

These applications are developed for particular purposes and clinical settings. A system evaluated using adult wrist and ankle radiographs cannot automatically be assumed to perform equally well in children, patients with altered bone anatomy, or examinations involving other anatomical regions. Supported patient populations, fracture types, imaging requirements, and intended-use limitations vary between products.

Detecting a suspected fracture is also different from interpreting the entire radiographic examination. An image may show a dislocation, another bone or joint abnormality, or an incidental finding outside the application's validated function. A fracture-focused AI system may perform its intended task correctly without recognizing every other clinically relevant feature.

Performance also depends on the images supplied to the software. Missing views, inadequate positioning, incomplete image transmission, or processing failures may limit the usefulness of its output. AI cannot reliably identify an injury that is not adequately represented on the acquired radiographs.

Clinical assessment therefore remains broader than the software result. The examining clinician must consider the patient's symptoms, mechanism of injury, physical findings, imaging, and other relevant information before deciding whether treatment, further investigation, or follow-up is needed.

AI can provide supplementary information during this process, but its output should not be interpreted as confirmation that the entire examination is complete or that discharge is necessarily safe.

What the October 2026 evidence tells us

Two studies published in npj Digital Medicine on 27 October 2026 provide useful perspectives on AI-assisted fracture detection. They address different questions: how accurately AI can identify fractures across the existing research literature, and what happens when healthcare professionals use AI during clinical practice.

Kuo and colleagues conducted an umbrella review and second-order meta-analysis of AI performance in fracture detection across medical imaging modalities [1]. The review included 50 reviews, comprising 25 systematic reviews and 25 meta-analyses, covering 387 primary studies. Twenty-one meta-analyses, representing 269 primary studies, contributed to the second-order quantitative analysis.

The pooled sensitivity of AI was 89.7%, with a 95% confidence interval of 88.2% to 91.1%. Pooled specificity was 90.4%, with a 95% confidence interval of 89.1% to 91.6%.

These results suggest favorable diagnostic accuracy in the populations and settings represented by the underlying studies. However, they should not be treated as performance estimates for every AI application used to detect fractures on emergency department X-rays. The review included different imaging modalities, clinical settings, and algorithms.

Another limitation concerns the timing of the evidence. Although the umbrella review was published in October 2026, its literature search extended to 4 June 2025. The findings therefore summarize research identified within that search period rather than all studies published up to October 2026.

The review also found that comparative evidence involving AI, clinicians, and AI-assisted clinicians did not establish a consistent advantage across settings. The available comparisons were limited, the certainty of evidence was low or very low for almost all evaluated outcomes, and the methodological quality of the included meta-analyses was generally rated as low or critically low.

Importantly, none of the included reviews reported adverse events or other patient-safety outcomes. The pooled accuracy estimates therefore do not establish that introducing AI reduces fracture complications, prevents delayed treatment, or improves long-term functional recovery.

A separate prospective study by Tan and colleagues provides a closer look at clinical implementation [2]. The investigators evaluated 2,496 consecutive musculoskeletal X-ray examinations during a three-month deployment in one emergency department. Thirteen radiographers participating in a preliminary interpretation program each assessed assigned examinations before reviewing the output of a commercial AI application, then recorded their final decisions.

The investigators compared these interpretations with a radiologist reference standard.

Standalone AI sensitivity was 75.6%, falling to 46.2% for subtle fractures. The radiographers' baseline sensitivity was 91.6%, and their overall accuracy before AI assistance was already high, at 97.6%. After reviewing AI output, overall accuracy increased modestly to 98.0%.

AI assistance helped correct 11 interpretation errors but introduced one additional error, producing a net improvement of 10 cases.

The study also identified an operational limitation. No AI output or an incomplete result was returned for 8.8% of acquired examinations. This matters because the absence of a valid output is not equivalent to a negative result.

These findings do not suggest that AI has no clinical value. Rather, they show that its contribution depends on the existing performance of healthcare professionals, how those professionals use the additional information, and whether the system operates reliably within the clinical workflow.

The benefit was relatively small in a setting where the participating radiographers already demonstrated high accuracy. Results may differ in departments with other staffing arrangements, professional experience, imaging practices, or patient populations.

The study involved a specific radiographer-led interpretation model. Its findings should not be taken to imply that the same professional responsibilities or interpretation arrangements automatically apply to radiographers and other healthcare personnel in Viet Nam. Local practice must remain consistent with the relevant professional qualifications and authorized scope of practice.

The study also did not measure subsequent treatment decisions, patient recalls, fracture complications, or functional outcomes. It therefore cannot establish whether the observed improvement in interpretation translated into better patient care.

Together, these studies support careful evaluation of AI-assisted fracture detection. They also reinforce a practical concern: diagnostic accuracy is only one part of a safe emergency care pathway.

Timely radiology reporting remains an important safeguard

International guidance offers a useful perspective on where AI should fit within fracture assessment.

In January 2025, the United Kingdom's National Institute for Health and Care Excellence (NICE) published HealthTech guidance HTG739 on AI technologies for detecting fractures on X-rays in urgent care [3]. NICE conditionally recommended four technologies for use within the National Health Service during an evidence-generation period, subject to requirements including appropriate regulatory approval and further evidence collection.

The four technologies are Rayvolve, TechCare Alert, BoneView, and RBfracture. The recommendations include product-specific age restrictions and are not unconditional endorsements for permanent routine adoption.

NICE identified uncertainties involving clinical performance, treatment decisions, additional imaging, specialist referrals, implementation costs, software failures, and patient outcomes. Its guidance also recognizes that AI may offer different benefits depending on how a clinical service is organized.

A separate NICE clinical guideline, NG38, recommends that a radiologist, radiographer, or other appropriately trained reporter provide a definitive written report of emergency department X-rays for suspected fractures before the patient is discharged [4].

This approach, commonly described as hot reporting, addresses an important source of risk. When the definitive interpretation is available before discharge, a clinically significant finding can be considered while the patient remains in the department. Any necessary change in management may then be made without relying on a later patient recall.

Emergency departments, however, differ in their reporting capacity, staffing, and access to specialist interpretation. Some can provide definitive reporting promptly, while others make initial clinical decisions before the formal report is completed.

Where reporting occurs after discharge, a reliable process is needed to identify clinically significant new findings and determine whether additional action is required. This should be regarded as a safeguard for delayed reporting, not as a substitute for timely professional interpretation where that can reasonably be provided.

The Royal College of Radiologists and the Academy of Medical Royal Colleges have also published recommendations on alerts and notification of imaging reports [5]. Their guidance emphasizes the timely communication of critical, urgent, and unexpected significant findings so that appropriate clinical action can follow.

Similarly, the US Centers for Disease Control and Prevention's Core Elements of Hospital Diagnostic Excellence, updated in February 2026, identifies reliable diagnostic testing, interpretation, communication, follow-up, and learning from diagnostic safety events as important components of hospital diagnostic excellence [6].

These recommendations provide useful international reference points. They do not automatically establish legal requirements for Vietnamese healthcare facilities, and their application should reflect local professional responsibilities and organizational circumstances.

When the final radiology report changes the clinical picture

An emergency clinician may review an X-ray and make a treatment or discharge decision before the radiologist completes the formal report. In many cases, the final interpretation confirms the initial assessment or provides additional detail without changing management.

Occasionally, however, the difference is clinically important. A fracture initially considered absent may be identified, or a suspected fracture may be characterized differently. An unexpected finding may change the need for immobilization, further imaging, orthopedic consultation, or the urgency of follow-up.

The difficulty is not always recognizing that two interpretations differ. It is determining whether the new information changes what should happen to the patient.

In this article, radiology report reconciliation refers to an organizational process that compares the finalized imaging interpretation with the information used for an earlier clinical decision and identifies differences that may require further action.

This is a proposed operational description, not a claim that Vietnamese legislation establishes a single nationally mandated procedure under that exact name.

A report may be finalized and stored in the radiology information system without anyone reviewing whether it changes the patient's treatment. An electronic notification may reach a clinician who is no longer on duty. A finding may be acknowledged without a decision about whether further care is needed.

These problems can arise even when the radiologist has interpreted the images correctly and the information system is functioning as designed.

A reliable reconciliation process therefore needs a clear relationship between the imaging report and the earlier clinical decision. An appropriately responsible clinician must be able to determine whether the new finding changes management, whether the patient has already received suitable treatment, and whether further assessment is necessary.

Not every difference requires a recall or additional intervention. For example, a radiologist may provide a more precise description of a stable fracture without changing the initial management. In such circumstances, documenting that the report was reviewed and that the existing treatment remains appropriate may be sufficient under the organization's procedures.

Other findings require action. A fracture that was not previously suspected may mean that the patient needs immobilization or additional evaluation. A different interpretation may also affect the urgency of specialist review.

The arrangements must work across shift changes, weekends, overnight reporting, and situations where the original clinician is unavailable. Responsibility for reviewing important findings should not disappear when a particular professional completes a shift.

Amended radiology reports create another challenge. An addendum issued after the original report has been reviewed may introduce information that changes the patient's care. The hospital therefore needs to consider how clinically important amendments are recognized and whether a previous acknowledgment remains sufficient.

Receiving a report, reviewing its clinical significance, and resolving any necessary action are related but separate activities. A healthcare organization should be able to distinguish between them.

Post-discharge follow-up requires more than a telephone call

When a clinically significant fracture is identified after discharge, the appropriate response depends on the injury, the patient's condition, and the potential consequences of delayed treatment.

Some findings may require urgent reassessment, particularly where instability or another serious concern makes delay potentially harmful. Others may be addressed through a scheduled appointment, modified immobilization, further imaging, or communication with the treating service.

The decision about urgency requires professional judgment. Recalling every patient with a difference between preliminary and final interpretations would not necessarily improve care and could expose patients to unnecessary investigations, anxiety, and inconvenience.

Once a finding is judged to require action, communication becomes part of clinical care rather than merely an administrative task.

A patient may reasonably believe that an X-ray has been fully interpreted when leaving the emergency department. Being contacted later about a fracture can therefore be confusing or concerning. The explanation needs to make clear what the later report identified, why the information matters, and what the patient is advised to do next.

The communication should also avoid automatically attributing blame to the original clinician, radiologist, or AI application. A change in interpretation does not, by itself, establish professional negligence or a technical failure. Those questions require separate assessment when relevant.

For some findings, communicating the result and giving clear instructions may be sufficient. Other situations require the healthcare organization to arrange an appointment, contact another clinical service, or confirm that an urgent reassessment has occurred.

A telephone call is not necessarily the end of that process.

The hospital may attempt contact without reaching the patient. The patient may receive a message but misunderstand its significance. An appointment may be arranged but not attended. Someone may understand the advice yet be unable to return because of transportation difficulties, distance, work commitments, financial concerns, or other barriers.

These circumstances matter because a completed notification task does not necessarily mean that the patient has received appropriate follow-up.

Healthcare organizations can make this distinction clearer by documenting whether contact was attempted, whether the relevant information was successfully communicated, whether the recommended care was arranged, and whether follow-up occurred when verification was clinically appropriate.

The necessary level of follow-up verification will vary. A minor finding that does not materially change treatment may need less intensive follow-up than an injury where delay could lead to substantial harm.

Unsuccessful contact attempts also require proportionate arrangements. An incorrect telephone number, an unanswered call, or a message reaching an unintended recipient may prevent important information from reaching the patient.

Where the potential consequences of delay are significant, alternative communication or escalation may be appropriate, subject to applicable confidentiality requirements and institutional procedures. Organizations should also distinguish a patient who declines recommended care after receiving appropriate information from someone who was never successfully contacted.

These differences can help identify whether a failure arose from communication, clinical handover, scheduling, contact information, or barriers to accessing care.

The aim is not to create an identical recall process for every radiology finding. It is to ensure that clinically significant information identified after discharge is not silently lost.

Clinical suspicion still matters when X-rays and AI are negative

Not every fracture missed during the initial assessment can be recognized by reviewing the original images.

Some fractures may be radiographically occult at the time of the first examination. The injury may be present even though the radiographs do not show a sufficiently recognizable abnormality.

Suspected scaphoid fracture is a familiar example. A patient may have clinical findings suggestive of scaphoid injury despite normal initial wrist X-rays. Depending on the circumstances, further assessment may involve immobilization, additional imaging, or planned reassessment.

NICE NG38 recommends considering MRI as first-line imaging for suspected scaphoid fractures following a thorough clinical examination [4]. This is a United Kingdom clinical recommendation rather than a mandatory Vietnamese imaging protocol.

The important principle is that a negative radiograph does not always exclude a clinically significant injury. A negative AI result based on the same images does not remove that uncertainty.

Discharge instructions should therefore reflect the patient's actual clinical findings and the level of diagnostic uncertainty. Where appropriate, patients need to understand when further assessment is necessary if symptoms persist, function remains impaired, or new concerns arise.

There is also a technical distinction between a negative AI result and an examination for which the AI system did not provide a valid output.

The prospective emergency department study published in October 2026 illustrates this problem [2]. Some examinations received no AI result or an incomplete output. If clinicians assume that every examination has been analyzed successfully, the absence of an alert may be misunderstood.

An organization using AI should ensure that unsuccessful or incomplete processing is distinguishable from a valid negative result. When AI output is unavailable, an appropriate alternative interpretation pathway must remain available.

Neither an apparently normal X-ray nor a negative AI classification should create greater diagnostic certainty than the available clinical evidence supports.

Evaluating AI within the actual emergency care pathway

Before adopting fracture detection software, a healthcare organization should understand how its current imaging and reporting processes work.

Who interprets X-rays initially? When is the formal report normally available? Are patients sometimes discharged before that report is completed? How are clinically significant differences identified, and what happens if the original clinician is no longer available?

These questions help establish whether AI is being introduced to address a specific clinical need or simply because the technology is available.

A hospital with prompt definitive radiology reporting may derive a different benefit from AI than a facility where preliminary interpretations are routinely made before the formal report becomes available. The experience of the professionals interpreting images, the types of examinations performed, and the timing of clinical decisions will influence the potential value.

Local evaluation should therefore include examinations representative of the hospital's intended patient population, anatomical regions, fracture types, imaging equipment, and acquisition practices. Evidence from another healthcare setting or software version should not automatically be treated as confirmation that the same performance will be achieved locally.

The organization also needs to consider how AI output reaches clinicians. An alert generated before discharge may influence immediate treatment, while one received after discharge may require an additional review process. If the output is delayed, incomplete, or difficult to access, its clinical usefulness may be reduced regardless of the algorithm's measured accuracy.

Performance monitoring should extend beyond sensitivity and specificity. Hospitals may examine the proportion of examinations receiving valid AI output, the timing of preliminary and definitive reports, the frequency of clinically significant differences between interpretations, and the time required to review findings that could change management.

For patients discharged before the final report, monitoring may include successful communication of important findings, completion of recommended reassessment, delays in follow-up, and relevant unplanned return visits.

These indicators need careful interpretation. More patient recalls after implementation could mean that previously overlooked fractures are being recognized earlier. It might also reflect increased false-positive findings or changes in reporting practice. Fewer recalls do not necessarily demonstrate improved safety.

The hospital should also consider whether AI contributes to unnecessary imaging, specialist referrals, immobilization, or additional workload. Avoiding missed fractures is important, but excessive investigations and inappropriate treatment are also relevant patient-safety concerns.

When feasible, case reviews can help explain the findings. A delayed report may reveal a communication weakness rather than an AI failure. A clinician may have appropriately rejected an incorrect AI alert. A patient recall may have been necessary even though the original assessment was reasonable based on the information available.

These distinctions support learning without assuming that every discrepancy represents professional error.

The measures discussed here are proposed considerations for organizational evaluation and quality improvement. They are not a mandatory Vietnamese AI fracture detection scorecard or an independently validated Digital Medicine Vietnam assessment instrument.

Professional responsibilities and the Vietnamese regulatory context

Introducing AI-assisted fracture detection in Viet Nam requires consideration of several areas of law, including medical examination and treatment, diagnostic imaging, artificial intelligence, medical device management, and personal data protection.

The applicable requirements depend on the software's intended medical purpose, technical functions, deployment arrangements, and the clinical decisions its outputs influence.

Clinical practice and diagnostic imaging

Viet Nam's Law on Medical Examination and Treatment No. 15/2023/QH15, effective from 1 January 2024, as amended, provides an important legal foundation for healthcare services, professional practice, patient rights, and medical records [8].

The use of AI does not independently expand the legally authorized professional scope of healthcare personnel. Clinical assessment, imaging interpretation, treatment decisions, and subsequent care remain subject to the applicable professional responsibilities and institutional arrangements.

The Ministry of Health's Decision No. 622/QĐ-BYT, issued on 11 March 2026, promulgated the updated professional guidance on technical procedures for diagnostic imaging under the radiology category, Volume 1 [7]. This decision replaced Decision No. 2775/QĐ-BYT of 29 August 2025 and provides relevant national professional guidance for diagnostic imaging services.

The document should be interpreted according to the actual technical procedures and professional requirements involved. It should not be described as establishing a dedicated AI fracture detection or post-discharge recall protocol.

Healthcare organizations need to determine how the applicable requirements relate to their authorized services and the professionals involved in emergency assessment, diagnostic imaging, and follow-up.

Where a clinically significant finding becomes available after discharge, the hospital should address it through appropriate clinical and organizational procedures. The existence of a later report does not automatically establish that the earlier assessment was negligent, but important new information should not be ignored simply because the initial encounter has ended.

Artificial intelligence and medical device regulation

Viet Nam's Law on Artificial Intelligence No. 134/2025/QH15 took effect on 1 March 2026 [9]. Decree No. 142/2026/NĐ-CP, effective from 1 May 2026, provides implementing provisions addressing matters including AI risk classification and management [10].

Decision No. 33/2026/QĐ-TTg, effective from 15 August 2026, establishes a list of high-risk AI systems [11].

The healthcare-specific entries in this list concern defined AI-assisted surgical and surgical robotic systems. An application used to support fracture detection on X-rays should therefore not automatically be classified as legally high-risk solely because it performs a diagnostic function.

However, that distinction does not establish that the application has no clinical risk or is exempt from other applicable requirements. The specific functions, intended purpose, and deployment arrangements need to be assessed against the relevant legal framework.

Medical device regulation requires separate consideration.

Decree No. 98/2021/NĐ-CP on medical device management, as amended and reflected in Ministry of Health Consolidated Document No. 08/VBHN-BYT dated 6 March 2026, includes software within the definition of a medical device when the applicable criteria are satisfied [12].

There is an important software-specific qualification. Clause 8, Article 3 of the consolidated framework provides that certain provisions concerning medical device classification, issuance of circulation numbers, and declaration of eligibility for trading do not apply to software used for medical devices.

The scope of this provision should be assessed carefully. It should not be interpreted as a blanket exemption for every standalone AI application that performs a medical function, nor as confirmation that a particular software product has satisfied all other applicable legal requirements.

Its relevance depends on the nature of the software, its intended purpose, its relationship with other medical devices, and the applicable regulatory interpretation.

Circular No. 24/2026/TT-BYT, effective from 1 July 2026, further addresses risk determination and management measures for products classified as medical devices and amends related implementing provisions [13].

Consequently, a healthcare organization should not assume either that all AI fracture detection software must follow the same medical device registration pathway or that software-based products are automatically exempt from regulatory obligations.

The legal status of the specific product and version should be assessed before routine deployment. The regulatory position of an existing X-ray machine or imaging system does not necessarily determine the status of separately integrated AI software.

Foreign regulatory authorization and published clinical validation may provide useful information, but neither independently establishes compliance with all applicable Vietnamese requirements.

Personal data protection and patient confidentiality

AI-assisted fracture detection may involve identifiable X-ray images, patient information, medical records, AI outputs, and communications concerning follow-up after discharge.

Viet Nam's Law on Personal Data Protection No. 91/2025/QH15 and Decree No. 356/2025/NĐ-CP both took effect on 1 January 2026 [14,15].

Healthcare organizations should assess the relevant legal grounds and purposes for processing patient data, applicable consent requirements and exceptions, access controls, security measures, retention arrangements, and the responsibilities of technology providers.

Particular attention is needed when medical images are transmitted to external AI platforms or processed through cloud services. Hospitals need to understand where the information is processed, who can access it, and whether it may be used beyond the intended clinical purpose.

Where personal data are transferred outside Viet Nam, the applicable cross-border transfer requirements, procedures, and exceptions need separate assessment.

Secondary uses of medical images or clinical information for algorithm development, model retraining, research, or commercial product improvement also require consideration of the relevant legal and institutional requirements.

Removing visible identifiers does not necessarily make an image or associated dataset fully anonymous. Identifying information may remain within metadata or other linked records, and the possibility of reidentification may need to be assessed.

The same attention to confidentiality applies to post-discharge communication. A patient recall may require telephone calls, electronic messages, or communication through an authorized representative, but the organization should avoid unnecessary disclosure of medical information to unintended recipients.

Privacy safeguards need to support both confidentiality and timely clinical follow-up. Protecting patient information and ensuring that important findings receive an appropriate response are complementary responsibilities.

From fracture detection to reliable care after discharge

AI-assisted fracture detection may help healthcare professionals identify fractures that would otherwise be overlooked during initial X-ray interpretation. Recent evidence demonstrates potential diagnostic value, but also shows that its contribution varies according to clinical setting, professional performance, and operational reliability.

For Vietnamese hospitals, the important consideration is how the technology fits within the existing emergency care pathway.

Timely and competent radiology reporting remains a fundamental safeguard. Where a definitive report is available before discharge, important findings can be considered while the patient is still present. When reporting occurs later, the healthcare organization needs a reliable way to identify clinically significant new information, review its implications, and arrange appropriate follow-up.

A fracture may be recognized during the first clinical assessment, identified through AI assistance, discovered in the finalized radiology report, or confirmed only after additional imaging. Each situation may require a different clinical response.

The hospital needs a dependable connection between diagnostic information and patient care. That requires more than installing software or generating alerts. It involves clear professional responsibilities, appropriate handovers, reliable report review, effective communication, and proportionate follow-up according to clinical urgency.

A completed radiology report does not necessarily mean its findings have been addressed. A clinician's acknowledgment does not always mean that a clinically significant discrepancy has been resolved. And a patient recall cannot be considered successful merely because a telephone call or electronic message was attempted.

The value of AI-assisted fracture detection should not be judged only by how accurately an algorithm identifies abnormalities on X-rays. It should also be judged by whether the healthcare organization recognizes important new findings, connects them to clinical decisions, and ensures that patients receive appropriate care when their diagnosis or treatment needs to change after discharge.

Professional Education Notice

This article is published by Digital Medicine Vietnam (DMVN) for professional education, evidence interpretation, and knowledge exchange. It does not provide individualized fracture diagnosis, emergency care, radiology interpretation, orthopedic treatment, medical device compliance, legal, regulatory, or patient-safety advice.

The opening clinical scenario is hypothetical and does not describe an identifiable patient or documented clinical incident.

The discussion does not establish the safety, clinical effectiveness, regulatory authorization, or suitability of any particular commercial, open-source, or internally developed AI application. Product-specific evidence must be interpreted according to the intended purpose, software version, supported patient populations, anatomical coverage, technical limitations, and actual clinical setting.

The cited studies should be understood in the context of their designs, reference standards, professional readers, patient populations, and limitations. Improvements in diagnostic accuracy do not independently establish reduced complications, better functional recovery, or other patient-safety benefits.

International guidance, including NICE recommendations and Royal College of Radiologists publications, should be interpreted within its original jurisdiction. Such guidance does not constitute Vietnamese regulatory authorization or a mandatory Vietnamese clinical protocol.

The Vietnamese laws and professional documents discussed apply according to their actual scope, effective provisions, and relevant circumstances. Medical device and AI regulatory requirements must be assessed for the specific technology and deployment arrangement.

The operational approaches described, including radiology report reconciliation, clinical review, post-discharge communication, follow-up verification, and local monitoring, are proposed considerations for responsible healthcare AI implementation and quality improvement. They are not mandatory Vietnamese AI fracture detection protocols, individual legal compliance determinations, or independently validated DMVN assessment instruments.

Clinical decisions require appropriate professional assessment and must reflect the patient's circumstances and applicable standards of care. Organizations considering implementation should obtain suitable clinical, technical, regulatory, and legal review where necessary.

References

[1] Kuo R, O'Hanlon C, Freethy A, et al. Performance of artificial intelligence for fracture detection: an umbrella review and second-order-meta-analysis. npj Digital Medicine. Published 27 October 2026. DOI: 10.1038/s41746-026-03303-z.

[2] Tan JR, Tran ANT, Sim WY, et al. Radiographer fracture detection with and without AI assistance in prospective emergency department use. npj Digital Medicine. Published 27 October 2026. DOI: 10.1038/s41746-026-03313-x.

[3] National Institute for Health and Care Excellence. Artificial Intelligence (AI) Technologies to Help Detect Fractures on X-Rays in Urgent Care: Early Value Assessment. HealthTech Guidance HTG739. Published 14 January 2025.

[4] National Institute for Health and Care Excellence. Fractures (Non-Complex): Assessment and Management. NICE Guideline NG38. Published 17 February 2016.

[5] The Royal College of Radiologists and the Academy of Medical Royal Colleges. Recommendations on Alerts and Notification of Imaging Reports. October 2022.

[6] Centers for Disease Control and Prevention. Core Elements of Hospital Diagnostic Excellence (DxEx). Updated 4 February 2026.

[7] Ministry of Health of Viet Nam. Decision No. 622/QĐ-BYT Promulgating the Professional Guidance on Technical Procedures for Diagnostic Imaging under Radiology, Volume 1. Issued 11 March 2026; replacing Decision No. 2775/QĐ-BYT of 29 August 2025.

[8] National Assembly of Viet Nam. Law No. 15/2023/QH15 on Medical Examination and Treatment. Adopted 9 January 2023; effective 1 January 2024, as amended.

[9] National Assembly of Viet Nam. Law No. 134/2025/QH15 on Artificial Intelligence. Adopted 10 December 2025; effective 1 March 2026.

[10] Government of Viet Nam. Decree No. 142/2026/NĐ-CP Detailing Certain Provisions and Measures for Implementing the Law on Artificial Intelligence. Issued 30 April 2026; effective 1 May 2026.

[11] Prime Minister of Viet Nam. Decision No. 33/2026/QĐ-TTg Promulgating the List of High-Risk Artificial Intelligence Systems. Issued 30 June 2026; effective 15 August 2026.

[12] Government of Viet Nam. Decree No. 98/2021/NĐ-CP on Medical Device Management, as amended, including by Decrees No. 07/2023/NĐ-CP and No. 04/2025/NĐ-CP. See also Ministry of Health Consolidated Document No. 08/VBHN-BYT dated 6 March 2026.

[13] Ministry of Health of Viet Nam. Circular No. 24/2026/TT-BYT on Risk Determination and Management Measures for Products Classified as Medical Devices and Amendments to Circular No. 05/2022/TT-BYT. Issued 30 June 2026; effective 1 July 2026.

[14] National Assembly of Viet Nam. Law No. 91/2025/QH15 on Personal Data Protection. Adopted 26 June 2025; effective 1 January 2026.

[15] Government of Viet Nam. Decree No. 356/2025/NĐ-CP Detailing Certain Provisions and Measures for Implementing the Law on Personal Data Protection. Issued 31 December 2025; effective 1 January 2026.