Governing AI-Assisted Follow-Up of Incidental Pulmonary Nodules in Viet Nam: Radiology Report Review, Risk-Based Management and Closed-Loop Tracking

Governing AI-Assisted Follow-Up of Incidental Pulmonary Nodules in Viet Nam: Radiology Report Review, Risk-Based Management and Closed-Loop Tracking

A patient undergoes a chest CT scan during an emergency department visit. The examination reveals a small pulmonary nodule unrelated to the symptoms that brought the patient to the hospital. The radiologist documents the finding and recommends follow-up when appropriate. The patient receives treatment for the original condition and is discharged.

Several months later, there is no clear record of what happened to the pulmonary nodule recommendation. The radiology report was available, but it is uncertain whether the finding was reviewed, whether the patient understood its significance, or who was responsible for arranging further assessment.

This situation illustrates a patient-safety problem that artificial intelligence (AI) may help healthcare organizations address. AI-assisted systems can examine radiology reports, identify findings that may require attention, and help track outstanding follow-up recommendations. Yet identifying a report that needs review is not the same as ensuring that the patient receives appropriate care.

For hospitals and clinics in Viet Nam, the challenge is to use these technologies to make follow-up more reliable without creating unnecessary investigations, overlooking clinically important abnormalities, or weakening professional accountability.

Why Incidental Pulmonary Nodule Follow-Up Matters

Pulmonary nodules are often discovered on CT examinations performed for reasons unrelated to suspected lung cancer. They may appear on a chest CT requested during an emergency visit or on another CT examination that includes part of the lungs.

Many nodules are benign. Some are associated with previous infection, inflammation, or other nonmalignant conditions. Others require further assessment because of their imaging characteristics, changes over time, or the patient's clinical risk factors.

Finding a pulmonary nodule does not necessarily mean that a patient has cancer. Equally, the fact that a nodule was discovered incidentally does not mean that it can safely be ignored.

Appropriate management may involve reviewing previous imaging, arranging interval CT surveillance, referring the patient for specialist assessment, or undertaking further diagnostic investigation. In some circumstances, no additional imaging is necessary. The decision depends on the characteristics of the finding and the patient's clinical situation.

The challenge often becomes apparent after the radiologist has completed the report.

A follow-up recommendation may be clearly documented but not acted upon by the requesting clinician. A patient may leave the emergency department without understanding the finding. An outpatient referral may be issued, but the patient may never attend the appointment. Previous imaging may be stored at another healthcare facility, making comparison difficult.

These problems can occur even when the original radiology report is accurate. They demonstrate why incidental pulmonary nodule management is not solely a radiology issue. It also depends on communication, coordination between services, and the ability to recognize when an agreed clinical action has not occurred.

The American College of Radiology (ACR) addresses this problem through its Recommendations Follow-up Improvement Collaborative. The initiative focuses on whether radiology reports contain appropriate, actionable follow-up recommendations and whether recommended examinations are completed within clinically suitable timeframes.

For healthcare organizations considering AI-assisted follow-up, this distinction matters. The objective should be to reduce missed opportunities for necessary clinical assessment while avoiding investigations that offer little benefit.

Understanding the Role and Limitations of AI-Assisted Report Review

Healthcare organizations need to distinguish between AI applications that analyze medical images and those that review written radiology reports.

An image-analysis system processes CT images and may detect or characterize pulmonary nodules directly from imaging data. Its performance depends on factors such as image quality, acquisition protocols, lesion characteristics, and the patient population in which it is used.

A radiology report-review system performs a different task. It uses natural language processing or other AI techniques to examine written reports, identify potentially relevant findings, and recognize recommendations that may require follow-up.

For example, a report-review system might identify a CT report describing a newly detected pulmonary nodule with a recommendation for interval imaging. The system could then place the case on a worklist for a designated clinical team.

This function may be useful in a busy hospital where radiology reports are generated across multiple departments and patient encounters. It can help bring relevant information to the attention of the people responsible for follow-up.

However, the limitations of the technology must be understood.

A system that reads only the radiology report cannot be expected to identify a pulmonary nodule that was never described in that report. It also cannot automatically establish that a correctly extracted recommendation is appropriate for the patient's current circumstances.

Radiology reports contain language that can be difficult for automated systems to interpret. A report may describe a newly detected nodule, a stable lesion documented on previous imaging, a benign calcified granuloma, or a finding that has already undergone specialist assessment.

Negation, uncertainty, and references to historical findings are particularly important. A statement that no new pulmonary nodules are present should not be interpreted as evidence of a newly detected lesion. Similarly, a recommendation mentioned in an earlier report should not automatically be treated as an outstanding clinical action.

Some reports contain several findings with different levels of concern. Others are amended after their initial release, potentially changing the interpretation of a finding or the recommended next step.

These differences matter because an incorrect AI flag can generate unnecessary clinical work, while a missed or incorrectly classified report may delay appropriate care.

Before introducing AI into routine practice, healthcare organizations should establish what the product is intended to do. A tool that identifies potentially relevant reports is different from one that evaluates clinical recommendations or automatically proposes patient management.

Each function requires evidence, evaluation, and safeguards appropriate to its potential influence on clinical decisions.

Clinical Verification and Risk-Based Management

Once an AI system identifies a report that may require attention, the next step should be clinical verification.

An appropriately qualified clinician should review the report and relevant patient information, including previous imaging and investigations where available. If the report is ambiguous or the recommendation appears inconsistent with the documented findings, clarification from the reporting radiologist may be necessary.

The review should establish whether the nodule is newly detected, previously known, already being managed, or unlikely to require further investigation. It should also consider whether information available since the original examination has changed the appropriate management decision.

Pulmonary nodules cannot be managed as a single clinical category.

Their significance depends on factors such as size, morphology, density, number, location, and changes over time. Patient-related factors, including age, smoking history, previous malignancy, and relevant clinical conditions, may also influence the assessment.

The Fleischner Society's 2017 recommendations remain an important international reference for managing incidentally detected pulmonary nodules on CT in adults. They distinguish solid from subsolid nodules and provide different management approaches according to imaging characteristics and clinical risk.

Their intended scope must be understood. The recommendations are not designed for routine application to patients younger than 35 years, patients with known primary cancer, immunocompromised patients, or individuals undergoing formal lung cancer screening. These circumstances require different clinical considerations or other applicable guidance.

Previous imaging is particularly valuable when evaluating whether a nodule has changed.

A clinical practice review by Callister and Silvestri, published in the New England Journal of Medicine in September 2026, reinforces the importance of distinguishing solid from subsolid nodules and comparing current examinations with earlier imaging.

Solid nodules that have remained stable over an appropriate period may carry a different level of concern from persistent subsolid nodules, which can grow more slowly and may require longer surveillance. The development or enlargement of a solid component within a subsolid nodule can be particularly important.

The same review describes how risk assessment can inform decisions about CT surveillance, positron emission tomography combined with CT (PET/CT), biopsy, or other management options. The appropriate approach depends on the estimated likelihood of malignancy, the patient's circumstances, and the risks associated with further investigation.

The ACR Appropriateness Criteria, Radiologic Management of Pulmonary Nodules and Masses: Update 2025, published in 2026, provide additional evidence-based guidance for selected clinical scenarios. These include stable or enlarging nodules, suspicious pulmonary lesions, and situations in which further imaging, tissue sampling, or other interventions may be considered.

The Fleischner Society recommendations and ACR Appropriateness Criteria address related but different clinical questions. They should inform clinical decisions within their intended scope rather than be treated as interchangeable instructions or converted into universal automated rules.

The Vietnamese clinical context also deserves attention.

Tuberculosis and other infectious or inflammatory lung conditions can influence the differential diagnosis of pulmonary nodules. Previous granulomatous disease may produce imaging findings that require careful interpretation, particularly when earlier examinations are unavailable.

However, a history of tuberculosis or the possibility of an infectious cause should not automatically exclude malignancy. Similarly, a suspicious imaging finding should not be treated as a confirmed cancer diagnosis without appropriate evaluation.

Clinical symptoms and the circumstances of presentation may also change the urgency of assessment. A patient with a concerning pulmonary lesion and symptoms suggestive of serious underlying disease may require a different pathway from an otherwise asymptomatic patient whose small nodule was discovered incidentally.

Both unnecessary investigation and failure to investigate clinically significant findings can harm patients.

For hospitals implementing AI-assisted follow-up, an important safeguard is to separate identification of a potential follow-up need from the professional decision about what care is appropriate.

AI may help bring a case to attention. The responsible clinician must determine the next step using the patient's clinical information, applicable guidance, and professional judgment.

Making Follow-Up a Defined Clinical Responsibility

A reliable follow-up process requires more than adding an alert to an electronic worklist.

Consider a CT report recommending interval imaging. An AI system identifies the report, and a clinician reviews the finding. The clinician confirms that surveillance is appropriate and determines when the next examination should occur.

At this point, the follow-up process has started, but it has not been completed.

The patient still needs to understand the recommendation. The examination must be arranged or an appropriate plan agreed upon. When the examination takes place, its results need to be reviewed, and any further decisions must be communicated to the patient and the professionals responsible for ongoing care.

An appointment that has been scheduled is not evidence that the recommended clinical assessment has been completed.

Hospitals should therefore distinguish between recommendations awaiting review, agreed management plans, scheduled actions, completed investigations, and cases requiring further escalation.

Completing one examination does not necessarily resolve the underlying clinical question. If repeat imaging leads to another surveillance recommendation, the next examination and the team responsible for it need to remain traceable.

Responsibility can become particularly difficult to establish when patients receive care from more than one organization.

In Viet Nam, a patient may undergo CT imaging at one hospital and consult a respiratory specialist at another. Some patients return to their home province after receiving emergency or specialist care in a larger city. Others move between public and private healthcare facilities or seek further assessment independently.

A referral letter may communicate the recommendation, but sending the letter does not necessarily establish that the patient has reached the receiving service or that responsibility for the next step has been accepted.

Where care is transferred to another healthcare organization, a documented handoff process should be used when appropriate. The level of confirmation required should reflect the clinical risk, available communication arrangements, and what can reasonably be achieved.

For patients who remain under the original organization's care, an identified clinician or team should retain responsibility for managing outstanding actions.

The follow-up system should make the status of each case understandable. Staff should be able to determine whether a management decision has been made, whether the patient has been informed, whether the next investigation is pending, and whether further action is required.

Case closure also needs a clear meaning.

A clinician may determine that no additional imaging is indicated. The recommended episode of assessment may have been completed, or responsibility may have been adequately transferred to another service. These situations can justify closing a particular follow-up task under an approved organizational procedure.

However, closing a completed task should not cause a new surveillance recommendation to disappear from the tracking process.

Patients who cannot be contacted, decline further investigation, or experience barriers to care require particular attention.

Repeated automated reminders may not help a patient who cannot afford travel, has difficulty obtaining an appointment, or does not understand the significance of the finding. Such situations require a proportionate clinical and organizational response.

Hospitals should establish how unsuccessful contact attempts are documented, when further communication is appropriate, and which circumstances require escalation. The approach should reflect the level of clinical concern rather than applying the same process to every pulmonary nodule.

Where a patient declines further investigation after an informed discussion, the record should reflect the information provided, the patient's decision, and any appropriate advice about when to seek further assessment.

An informed decision to decline is different from a recommendation that was never reviewed or communicated.

Closed-loop tracking should help organizations understand what happened to each clinically relevant recommendation. It should not become an indefinite reminder system that continues regardless of the patient's circumstances or the clinical value of further investigation.

Communicating Findings Without Creating Unnecessary Alarm

Pulmonary nodules can be difficult for patients to understand.

Some patients may interpret the term as a diagnosis of lung cancer. Others may assume that the finding is unimportant because it was unrelated to the symptoms that led to the CT examination.

Neither assumption is necessarily correct.

The explanation should describe what was found, what remains uncertain, why further assessment may or may not be necessary, and what the patient should expect next.

Communication should be understandable and proportionate to the clinical concern. A small nodule for which no routine surveillance is indicated should not be described in the same way as an enlarging suspicious lesion requiring prompt specialist assessment.

Where follow-up is recommended, patients should know which service is responsible, the intended timeframe, and how to obtain help if they cannot arrange the next appointment or investigation.

Automated messages may support this process, particularly when patients need reminders about previously agreed care. However, the information should be based on an appropriately reviewed clinical plan.

An unreviewed AI-generated message should not independently tell a patient that they have cancer or recommend invasive testing.

Some findings may require direct communication by a healthcare professional rather than reliance on routine electronic notification. A suspicious lesion requiring prompt assessment should not remain in an ordinary reminder queue when the clinical circumstances call for more urgent action.

Healthcare organizations also need to recognize the difference between delivering a message and communicating effectively.

An electronic system may confirm that a text message was sent, but it cannot establish that the patient understood the recommendation. Language, health literacy, contact preferences, family support, and access to healthcare services may all influence the result.

Patient involvement becomes especially important when follow-up is recommended several months after the initial CT examination. A clear explanation can help the patient understand why the next examination remains relevant even if symptoms have resolved or the patient feels well.

Validating AI-Assisted Report Review in Vietnamese Hospitals

International research can inform product selection, but published performance results do not establish that a particular AI application will work reliably in every Vietnamese healthcare organization.

Radiology departments differ in their reporting practices and information systems. Some use structured templates, while others rely mainly on narrative reports. Clinical documentation may be written in Vietnamese, English, or a combination of both.

Terminology, abbreviations, punctuation, and the placement of recommendations can vary between departments and individual radiologists.

An AI system developed primarily with English-language radiology reports may behave differently when processing Vietnamese reports. Translating reports before AI analysis also introduces potential risks if clinical meaning, uncertainty, or negation is not preserved.

Local evaluation should therefore use representative reports from the intended clinical environment.

The evaluation should establish whether the system correctly identifies reports requiring review and whether it distinguishes them from findings that do not require new action. Missed cases and unnecessary alerts should be assessed separately.

Particular attention should be given to reports containing negated findings, stable nodules, benign-appearing abnormalities, multiple lesions, historical recommendations, and references to earlier investigations.

A suitable reference standard should be established through qualified human review, with a process for resolving clinically meaningful disagreement.

Reviewing only reports flagged by AI is insufficient. The evaluation should also examine a representative selection of reports that the system did not identify, because clinically important false-negative results might otherwise remain unnoticed.

Patient identification and record linkage are equally important.

A report-review system may extract the correct recommendation but associate it with the wrong patient record. Duplicate reports, corrected reports, inconsistent identifiers, and incomplete clinical information can introduce errors that are independent of the AI model's ability to interpret language.

These risks should be evaluated before the system is introduced into routine clinical use.

Technical performance is only part of the assessment. The organization also needs to know whether the AI-assisted workflow supports safe follow-up in practice.

A system might accurately identify relevant reports but generate a worklist that staff cannot manage. Alerts may reach the wrong clinical team, remain unreviewed, or become difficult to distinguish from less important notifications.

Testing should therefore extend beyond report extraction to include clinical verification, case assignment, patient communication, scheduling, escalation, result review, and documentation.

The organization should also establish what happens when the AI system is unavailable or produces unreliable results. Existing clinical responsibilities and essential follow-up processes must continue.

Performance may change after implementation. New reporting templates, software updates, modifications to clinical terminology, or changes in information-system integration can affect the way an application behaves.

Hospitals should monitor these changes, investigate significant errors, and determine when additional evaluation is needed.

A successful pilot in one department should not be treated as proof that the same system will perform equally well across other hospitals or specialties.

What the Evidence Shows About AI-Assisted Follow-Up

Recent studies suggest that AI-assisted report review, when combined with organized clinical follow-up processes, may improve the management of incidental pulmonary nodules.

A 2026 study by Olivieri and colleagues, published in Respiratory Medicine, evaluated an AI-supported workflow at an academic-affiliated community hospital in the United States.

The researchers examined patients who underwent chest CT during emergency department visits. Following implementation, an AI-based natural language processing system identified reports containing potential incidental pulmonary nodules, and patients were contacted to facilitate follow-up.

The proportion of patients notified about their pulmonary nodule increased from 75.0% before implementation to 88.4% afterward. Nodule-specific follow-up increased from 53.5% to 67.9%.

These results suggest that combining automated report identification with organized patient outreach may help reduce gaps in communication and follow-up.

The researchers also identified difficulty contacting patients by telephone after discharge as an important barrier. Even an accurately functioning AI system cannot complete the clinical process when patients cannot be reached or connected with appropriate services.

However, the study used a retrospective before-and-after design at a single hospital. The intervention combined AI-supported identification with patient outreach and follow-up coordination.

It is therefore not possible to attribute the entire improvement to AI alone.

The researchers also found no statistically significant difference in lung cancer stage at diagnosis between the study cohorts. Improvements in notification and follow-up are encouraging process outcomes, but they should not be interpreted as demonstrated reductions in advanced cancer or mortality.

Evidence from organized quality-improvement programs provides additional context.

In a 2025 study published in the Journal of the American College of Radiology, Wandtke and colleagues reported results from the ACR Learning Network Recommendations Follow-Up Improvement Collaborative across seven participating healthcare sites.

Across these sites, average adherence to the Fleischner Society guidelines improved from 62.3% at baseline to 89.0% after the interventions. The average proportion of patients completing recommended follow-up imaging increased from 41.4% to 61.1%.

The participating sites achieved different levels of improvement. The researchers found that the most successful approaches combined reliable tracking systems with dedicated care coordination teams.

These findings support the importance of organizational arrangements alongside technology. The study was a quality-improvement initiative rather than a controlled trial demonstrating that AI alone was responsible for the reported changes.

A broader 2025 systematic review by Aunger and colleagues, published in CHEST, examined 31 studies of interventions intended to improve pulmonary nodule guideline adherence and follow-up.

The interventions included tracking systems, process improvement, natural language processing, radiology reporting templates, clinical decision-support tools, and greater patient involvement.

Tracking systems appeared particularly promising because they addressed several parts of the care pathway and could reduce reliance on individual memory or manual follow-up.

Nevertheless, the studies varied considerably in design, intervention, and outcome measurement. Most nonrandomized studies had serious or critical risk of bias, limiting the strength of conclusions that could be drawn.

Taken together, the available evidence supports a cautious interpretation.

Technology may help identify patients and organize outstanding recommendations. Dedicated clinical teams, clear responsibility, and effective communication are also important. The evidence does not establish that every AI product will improve patient outcomes or that results achieved in one healthcare system can be reproduced without local evaluation.

For Vietnamese healthcare organizations, purchasing an AI application should not be confused with implementing an effective pulmonary nodule follow-up service.

Measuring Whether Follow-Up Is Safer

Hospitals introducing AI-assisted follow-up should evaluate whether the process improves clinical care, not simply whether the software processes reports efficiently.

The number of reports reviewed, nodules flagged, or alerts generated may be useful operational measures. However, these figures do not establish whether clinically important findings were addressed or whether patients received appropriate care.

A meaningful evaluation should begin with the reliability of report identification. This includes the proportion of relevant reports correctly identified, the frequency of unnecessary alerts, and the number of clinically important reports missed by the system.

The next consideration is whether identification leads to an appropriate clinical response.

Organizations should examine whether relevant findings receive documented clinical review, whether management decisions are recorded, whether patients are appropriately informed, and whether agreed investigations or consultations occur within clinically justified timeframes.

The definition of success matters.

A patient whose clinician determines that no additional imaging is necessary should not be counted as a follow-up failure simply because no repeat CT was performed. Similarly, a patient who declines an investigation after an informed discussion should be distinguished from someone whose recommendation was never addressed.

Tracking completion rates without considering these differences can create misleading results.

The evaluation should also consider potential harm.

An increase in follow-up CT examinations may reflect better care for patients whose recommendations were previously overlooked. However, it may also indicate unnecessary surveillance or investigation.

Avoidable imaging can expose patients to radiation, additional costs, and anxiety. Further investigations may create risks of invasive procedures, additional incidental findings, and complications without sufficient clinical benefit.

The workload created for clinicians, radiologists, and follow-up coordinators also deserves attention. An AI system that produces more alerts than staff can reasonably review may become less reliable over time.

Hospitals should consider differences between patient groups and healthcare settings. Patients who live far from specialist centers may have more difficulty completing recommended investigations than those receiving care in major cities. Insurance arrangements, financial barriers, communication difficulties, and access to transportation may also affect follow-up.

Where reliable information is available, longer-term outcomes such as delayed cancer diagnosis and stage at diagnosis may be evaluated. These outcomes require careful interpretation, appropriate comparison groups, and sufficient follow-up.

Changes in clinical outcomes should not be attributed to AI without considering other factors, including differences in patient populations, staffing, clinical pathways, and the availability of diagnostic services.

The central measure of success is whether more patients receive the care they actually need, within an appropriate timeframe, while unnecessary investigations are minimized.

Vietnamese Legal and Organizational Responsibilities

AI-assisted pulmonary nodule follow-up needs to be considered within Viet Nam's existing framework for medical examination and treatment, artificial intelligence, medical devices, and personal data protection.

These regulatory areas address different responsibilities. Compliance with one framework does not automatically establish compliance with another or demonstrate that a clinical workflow is safe.

Professional Responsibility and Clinical Decision-Making

The Law on Medical Examination and Treatment No. 15/2023/QH15, effective January 1, 2024, as subsequently amended and reflected in Consolidated Document No. 26/VBHN-VPQH dated February 26, 2026, provides an important legal foundation for professional healthcare practice.

AI-generated information does not independently establish a diagnosis or replace the professional responsibilities of appropriately licensed healthcare practitioners.

Healthcare organizations should determine who is authorized to review pulmonary nodule findings, establish management plans, communicate clinical decisions, and oversee outstanding follow-up actions.

The use of an AI worklist should support these responsibilities rather than introduce an informal decision-making process in which clinical accountability is unclear.

The organization should also distinguish between the clinical responsibilities of practitioners and its own responsibilities for staffing, systems, documentation, and patient safety.

Artificial Intelligence Regulation

Viet Nam's Law on Artificial Intelligence No. 134/2025/QH15 took effect on March 1, 2026. Decree No. 142/2026/NĐ-CP, effective May 1, 2026, provides implementing provisions, including those relating to AI risk classification and other applicable responsibilities.

Decision No. 33/2026/QĐ-TTg, effective August 15, 2026, establishes the List of High-Risk Artificial Intelligence Systems.

The healthcare-specific entries expressly identified in the list concern defined applications of AI-assisted surgery and robotic surgical systems. AI-assisted review of pulmonary nodule reports is not separately named among those entries.

However, this does not mean that every report-review application is automatically classified as low-risk, exempt from applicable AI obligations, or clinically safe.

The legal assessment of a specific system depends on its intended purpose, functions, degree of automation, and actual deployment, together with the relevant statutory criteria.

A system that flags reports for subsequent clinical review may present a different risk profile from one that independently recommends investigations, determines follow-up intervals, or influences whether patients receive further assessment.

Decree No. 142/2026/NĐ-CP also addresses responsibilities relating to AI risk classification and reassessment when changes in deployment, integration, functionality, or intended use introduce new or greater risks.

Healthcare organizations should therefore evaluate both the applicable legal classification and the potential clinical consequences of errors, while clarifying the respective responsibilities of technology providers and deploying organizations.

Meaningful human oversight remains important even where the system is not expressly included in a particular high-risk category.

Medical Device Requirements

Some AI applications used in healthcare may fall within Viet Nam's medical device regulatory framework, depending on their intended medical purpose and functions.

Decree No. 98/2021/NĐ-CP on medical device management, as amended, including by Decree No. 07/2023/NĐ-CP and Decree No. 04/2025/NĐ-CP, provides part of the applicable framework.

Ministry of Health Circular No. 24/2026/TT-BYT, effective July 1, 2026, addresses risk determination and management measures for products classified as medical devices and amends provisions related to the implementation of Decree No. 98/2021/NĐ-CP.

Healthcare organizations should establish whether a proposed AI application is regulated as medical device software and identify the relevant requirements before deployment.

The presence of AI does not, by itself, determine whether software is a medical device. Equally, describing a product as an administrative or workflow tool does not necessarily exclude medical device requirements when its actual intended functions have a medical purpose.

Authorization or certification in another jurisdiction may provide relevant information for product assessment, but it does not automatically establish compliance with Vietnamese law.

Personal Data Protection

Pulmonary nodule follow-up involves health information that may identify individual patients. This can include imaging findings, clinical history, medical records, contact details, and information about subsequent investigations.

The Law on Personal Data Protection No. 91/2025/QH15 and Decree No. 356/2025/NĐ-CP, both effective January 1, 2026, are particularly relevant to these activities.

Healthcare organizations should establish the applicable lawful basis and purpose for processing personal data, including any consent requirements or legally applicable exceptions.

Access should be restricted to authorized personnel. Appropriate safeguards should also apply when patient information is exchanged with technology providers or integrated into external software services.

Hospitals need to understand where information is stored, which parties can access it, how long it is retained, and whether it is used for purposes beyond the original clinical follow-up process.

Additional assessment may be necessary where identifiable information is processed by external AI providers or transferred outside Viet Nam.

These arrangements should be examined according to the applicable legal requirements, including those concerning personal data processing and cross-border transfers where relevant.

Data collected to support patient follow-up should not automatically be treated as available for unrelated commercial activities or AI model training.

If patient information is proposed for research, the organization should separately evaluate applicable research ethics and participant protection requirements.

These issues are best addressed during product selection, contracting, and system design, rather than after identifiable patient data have already been transferred to a technology provider.

International clinical guidelines and quality-improvement initiatives can support organizational practice, but they do not automatically become binding Vietnamese legal requirements.

The appropriate governance arrangements need to reflect the actual AI application, the healthcare setting, the applicable laws, and the clinical risks involved.

A Practical Starting Point for Vietnamese Healthcare Organizations

Hospitals do not need to begin by deploying AI across every radiology report or clinical department.

A useful first step is to examine how incidental pulmonary nodules are currently managed.

A clinical quality team could review a representative sample of CT reports and determine whether relevant follow-up recommendations were documented, communicated, and acted upon. The review should also examine whether the recommendations were appropriate and whether the responsible clinical team could be identified.

This baseline assessment may reveal problems that technology alone cannot resolve.

A hospital may already produce clear radiology recommendations but lack a dependable process for following up patients after emergency department discharge. Another organization may have an effective outpatient scheduling system but limited visibility when patients undergo imaging or specialist assessment elsewhere.

The intervention should address the weakness that has actually been identified.

If AI-assisted report review is introduced, the organization can begin with a limited clinical setting and clearly defined functions.

Radiology, the relevant clinical services, care coordination, information technology, and quality management should agree on which reports are included, who reviews AI-generated findings, how management decisions are recorded, and what evidence is needed to close a follow-up task.

The initial implementation should be supervised. AI-generated worklists can be compared with independently reviewed reports, with particular attention to missed cases, incorrect flags, and clinically important discrepancies.

The organization should also confirm that alerts reach the appropriate team and that the staff responsible for reviewing them have sufficient time and access to relevant clinical information.

Where silent-mode evaluation or retrospective testing is used, patient information should be processed under appropriate privacy and information-security arrangements. The use of identifiable clinical reports in external AI services requires its own assessment and should not be assumed to be permissible merely because the activity is described as a pilot.

Follow-up failures should be reviewed as part of routine clinical quality improvement.

A delayed examination might result from an unclear radiology recommendation, an unassigned follow-up task, unsuccessful patient contact, or difficulty obtaining specialist care. Understanding the cause is more useful than simply recording that a reminder was missed.

Some improvements may involve clearer reporting language, better discharge communication, or a more reliable handoff between departments. Others may require changes to appointment coordination, escalation procedures, or information-system integration.

Not every problem requires an AI solution.

As the organization gains experience, the process can be extended to additional departments or facilities where there is a demonstrated need.

Expansion should depend on evidence that the workflow is clinically reliable, the workload is manageable, and appropriate safeguards remain effective.

A system evaluated on English-language radiology reports in one hospital should not automatically be assumed to perform equally well in another facility using different documentation practices or patient management arrangements.

Continued monitoring is necessary because both the technology and the clinical environment can change.

The aim should be to establish a sustainable follow-up process in which AI provides useful support without becoming a substitute for clinical judgment, effective communication, or organizational responsibility.

From Identifying a Nodule to Protecting the Patient

An incidental pulmonary nodule can create a clinical question that remains unresolved long after the original CT examination has been completed.

AI may help healthcare organizations identify relevant reports, recognize outstanding recommendations, and make follow-up responsibilities more visible. These capabilities can be valuable, particularly when patients move between departments or healthcare facilities.

However, the technology cannot resolve the clinical question on its own.

A patient may still need a careful review of previous imaging, an individualized management decision, a clear explanation of the finding, and access to further assessment when appropriate.

For hospitals and clinics in Viet Nam, AI-assisted pulmonary nodule follow-up is most useful when it operates within a reliable clinical process. The system should help clinicians identify patients who need attention, help patients understand what happens next, and help the organization recognize when an agreed action has not occurred.

Its value should not be judged by how many nodules it identifies or how many reminders it sends.

The more important question is whether patients receive appropriate, timely, and proportionate care, and whether the organization can demonstrate how clinically significant findings have been addressed.

A pulmonary nodule noted in a radiology report is only the beginning of that responsibility. What matters is what happens afterward.

Selected References and Regulatory Sources

International Clinical Guidance

MacMahon H, Naidich DP, Goo JM, et al. Guidelines for Management of Incidental Pulmonary Nodules Detected on CT Images: From the Fleischner Society 2017. Radiology. 2017;284(1):228–243. DOI: 10.1148/radiol.2017161659.

Callister MEJ, Silvestri GA. Pulmonary Nodules. New England Journal of Medicine. 2026;395:894–905. DOI: 10.1056/NEJMcp2515063.

Gilyard S, Ahmed O, Kim CY, et al. ACR Appropriateness Criteria® Radiologic Management of Pulmonary Nodules and Masses: Update 2025. Journal of the American College of Radiology. 2026;23(5):915–927. DOI: 10.1016/j.jacr.2026.01.027.

American College of Radiology. Recommendations Follow-up Improvement Collaborative. ACR Learning Network.

Implementation and Clinical Evidence

Olivieri PP, Markan A, Ashbaker D, et al. Improving follow-up of incidental pulmonary nodules in the emergency department using an artificial intelligence-supported workflow. Respiratory Medicine. 2026;261:109045. DOI: 10.1016/j.rmed.2026.109045.

Wandtke B, Szekeres D, Garcia Tomkins K, et al. The ACR Learning Network Recommendations Follow-Up Improvement Collaborative: Ensuring Quality Surveillance of Pulmonary Nodules. Journal of the American College of Radiology. 2025;22(9):1071–1081. DOI: 10.1016/j.jacr.2025.05.017.

Aunger J, Yip KP, Dosanjh K, et al. Interventions to Improve Adherence to Clinical Guidelines for the Management and Follow-Up of Pulmonary Nodules: A Systematic Review. CHEST. 2025;168(1):248–268. DOI: 10.1016/j.chest.2025.02.031.

World Health Organization. Global Tuberculosis Report 2025. Geneva: World Health Organization; 2025.

Relevant Vietnamese Legal Framework

National Assembly of Viet Nam. Law on Medical Examination and Treatment No. 15/2023/QH15. Adopted January 9, 2023; effective January 1, 2024, as subsequently amended. Consolidated Document No. 26/VBHN-VPQH, dated February 26, 2026.

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

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

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

Government of Viet Nam. Decree No. 98/2021/NĐ-CP on Medical Device Management, as amended, including by Decree No. 07/2023/NĐ-CP and Decree No. 04/2025/NĐ-CP.

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 June 30, 2026; effective July 1, 2026.

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

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

Editorial and Clinical Notice

This article is intended for professional education, healthcare quality improvement, and discussion of responsible AI implementation. The opening clinical scenario is illustrative and does not describe an identified patient or documented clinical incident.

The article summarizes selected international clinical guidance, published research, and relevant Vietnamese regulatory considerations. It also describes proposed organizational approaches to clinical verification, patient communication, follow-up tracking, and quality monitoring.

These approaches do not constitute a mandatory Vietnamese clinical protocol, an officially approved AI governance standard, or an independently validated pulmonary nodule management instrument.

International clinical recommendations should be interpreted according to their intended populations, clinical limitations, and individual patient circumstances. They do not automatically establish binding legal requirements for healthcare organizations in Viet Nam.

The article does not provide individualized medical advice, determine the regulatory classification of any particular AI product, or constitute a formal legal opinion.

Clinical decisions should be made by appropriately qualified healthcare professionals using current applicable guidance, relevant patient information, and locally approved procedures. Product classification, data processing arrangements, and other legal obligations must be assessed for the specific technology and deployment context.

Digital Medicine Vietnam (DMVN) | Content reviewed against sources available as of 1 November 2026.