Governing AI-Assisted Colonoscopy in Viet Nam: Polyp Detection, Endoscopist Review and Patient Safety
IMPLEMENTATION INSIGHTS
Governing AI-Assisted Colonoscopy in Viet Nam: Polyp Detection, Endoscopist Review and Patient Safety
IMPLEMENTATION INSIGHTS
Last reviewed: 23 October 2026
During a routine colonoscopy, an artificial intelligence system highlights a small area of the colonic mucosa that may contain a polyp. The endoscopist examines the area more closely, adjusts the viewing angle and determines whether a lesion is present.
A few minutes later, another alert appears around a mucosal fold. This time, closer inspection reveals a normal anatomical structure rather than a polyp. The endoscopist continues the examination, deciding which alerts require further attention while maintaining a careful inspection of the surrounding mucosa.
These examples illustrate an important feature of computer-aided detection (CADe) during colonoscopy. Artificial intelligence can help draw attention to abnormalities that might otherwise be overlooked, but an alert is not a confirmed diagnosis. The endoscopist must still determine what is present and whether any clinical action is appropriate.
The potential value of CADe is increasingly supported by evidence showing improved adenoma detection. At the same time, professional organizations continue to differ in their recommendations because the consequences of detecting additional lesions are more complicated than they first appear. More detected adenomas do not necessarily translate into fewer colorectal cancers, and additional findings may lead to more interventions and surveillance examinations.
For hospitals and endoscopy units in Viet Nam, introducing AI-assisted colonoscopy therefore involves more than acquiring compatible technology. It requires attention to existing procedural quality, professional competence, the interpretation of AI alerts, patient information, regulatory requirements and the consequences of clinical decisions.
The central question is not simply whether AI helps detect more polyps. It is whether the technology contributes to better-quality colonoscopy while preserving professional judgment and avoiding unnecessary risks for patients.
Understanding what CADe does during colonoscopy
Computer-aided detection systems analyze endoscopic images or video to identify areas that may contain colorectal polyps. During a procedure, the system typically displays a marker around a suspected lesion, directing the endoscopist's attention to that area of the image.
The purpose of CADe is to support visual detection. It does not independently establish the histological nature of a lesion, determine its malignant potential or decide whether it should be removed. A highlighted area may represent an adenoma, a serrated lesion, another type of polyp or a normal structure that resembles an abnormality.
This distinction is particularly important because AI applications in gastrointestinal endoscopy do not all perform the same function. Computer-aided diagnosis or characterization (CADx) is intended to assist with the optical assessment of an identified lesion, potentially providing information about its likely tissue characteristics. Computer-aided quality assessment systems may address other aspects of the examination, such as bowel cleanliness, mucosal visualization or selected procedural quality measures.
Some commercial platforms incorporate more than one of these functions. However, evidence that a system improves polyp detection should not automatically be interpreted as evidence that it can accurately characterize lesions or guide decisions about tissue removal.
An endoscopy unit considering AI should therefore begin by identifying the exact functions of the proposed product. It should understand what the system is intended to detect, which imaging conditions it supports, what limitations the manufacturer identifies and what clinical evidence is available for each function.
This distinction also helps establish appropriate expectations among endoscopists and patients. CADe is an additional source of visual information during colonoscopy, not an independent clinical decision-maker.
What international guidance tells us about CADe
The strongest established evidence for CADe concerns improvements in adenoma detection during colonoscopy. Identifying and appropriately managing precancerous colorectal lesions is an important part of colorectal cancer prevention, and adenoma detection rate (ADR) has become a widely used indicator of colonoscopy quality.
ADR generally measures the proportion of eligible colonoscopies in which at least one conventional adenoma is identified and confirmed by histopathology. Its interpretation depends on the population, indication, inclusion criteria and measurement definition being used.
Higher ADR has been associated with a lower subsequent risk of post-colonoscopy colorectal cancer. However, ADR remains a surrogate outcome. An increase in ADR following the introduction of CADe does not, by itself, demonstrate that the technology reduces colorectal cancer incidence or mortality.
This distinction helps explain why recent international guidance has not reached a single position on routine CADe use.
In March 2025, the American Gastroenterological Association (AGA) published a living clinical practice guideline on CADe-assisted colonoscopy. After considering randomized trials, modeling evidence, patient preferences, potential harms and resource implications, the AGA made no recommendation either for or against CADe use in adults undergoing colonoscopy [1].
The evidence reviewed by the AGA showed an absolute increase of approximately eight percentage points in ADR. However, the increase in the detection of advanced adenomas and/or sessile serrated lesions was much smaller, at approximately two percentage points. The certainty of evidence concerning long-term outcomes, including colorectal cancer incidence, colorectal cancer mortality and post-colonoscopy colorectal cancer, remained very low.
The AGA also considered the possibility that additional lesion detection could result in more intensive surveillance colonoscopies. Estimates of potential long-term benefits were derived partly from modeling rather than direct observation of reductions in cancer or mortality.
A separate living clinical practice guideline published in The BMJ in March 2025 took a different position. Its international panel issued a weak recommendation against routine CADe use in adults undergoing colonoscopy within the guideline's defined scope [2].
The panel considered that the long-term benefits remained uncertain and that detecting additional lesions could increase polypectomy, surveillance and associated burdens. The recommendation did not apply to certain situations, including colonoscopy related to a history of inflammatory bowel disease, abnormal imaging findings or therapeutic interventions.
A weak recommendation against routine use should not be interpreted as a conclusion that CADe has no clinical value. Rather, it reflects uncertainty about whether the expected benefits justify the potential harms, costs and additional procedures for the populations considered.
The European Society of Gastrointestinal Endoscopy (ESGE) published a position statement in 2025 that reached a different conclusion. ESGE expressed the view that most well-informed patients who had already decided to undergo colonoscopy for colorectal cancer screening or polyp surveillance would favor CADe assistance [3].
This position acknowledged potential benefits while recognizing that the evidence concerning long-term outcomes remained uncertain. It was not an unconditional endorsement of CADe in every clinical situation.
In March 2026, the United Kingdom's National Institute for Health and Care Excellence (NICE) adopted an evidence-generation approach. Its HealthTech guidance HTG773 stated that six specified technologies—CADDIE, CAD EYE, ENDO-AID, EndoScreener, GI Genius and MAGENTIQ-COLO—could be used in the NHS during an evidence-generation period to support colorectal polyp detection in people without diagnosed inflammatory bowel disease or Lynch syndrome [4].
The recommendation is conditional on appropriate regulatory approval, including relevant NHS digital technology requirements, and the generation of further evidence. NICE expects the evidence-generation period to last four years before reconsidering whether routine adoption is justified.
Importantly, NICE distinguished lesion detection from AI-assisted lesion characterization. The guidance called for further research on characterization rather than endorsing its routine clinical use under the same conditions as detection.
These positions are not necessarily disagreements about whether CADe can identify additional adenomas. They reflect different assessments of clinical importance, patient preferences, uncertainty, healthcare resources and the conditions under which a technology may reasonably be introduced.
For Vietnamese healthcare organizations, none of these international recommendations should be treated as a universal requirement to adopt or reject CADe. They provide evidence and professional perspectives that should inform decisions about specific products and local clinical circumstances.
What newer evidence from 2026 adds
Research published during 2026 has added important information about CADe performance in routine clinical practice, although the central uncertainty about long-term cancer outcomes remains.
In June 2026, Dominitz and colleagues reported a large cluster-randomized study involving the United States Veterans Health Administration [5]. The analysis included 334,200 colonoscopies performed by 816 endoscopists across 139 facilities, comparing centers where CADe devices were made available with control facilities.
At facilities equipped with CADe, ADR increased from 50.7% before deployment to 54.9% afterward. At control facilities, ADR decreased slightly, from 51.8% to 51.1%. After adjustment for relevant characteristics, CADe availability was associated with higher odds of adenoma detection, with an adjusted odds ratio of 1.22.
This is important evidence because it shows that improved adenoma detection can be observed in a large healthcare system outside the conditions of smaller, tightly controlled clinical trials. The study also found that the effect was not confined to endoscopists with low baseline detection rates.
However, the study evaluated the availability of CADe devices rather than documenting their use during every individual procedure. It did not establish that the observed improvement in ADR resulted in fewer colorectal cancers or cancer-related deaths. The population and healthcare setting also differ from those of many endoscopy services in Viet Nam.
Another important development came in October 2026. Anthony and colleagues published a systematic review and meta-analysis of randomized trials examining the effects of real-time CADe on different categories of colorectal lesions [6].
The review identified 90 eligible randomized trial families, with 64 contributing to at least one primary quantitative analysis. Across the trials included in the relevant analysis, CADe increased overall ADR, with a pooled risk ratio of 1.16.
However, the analysis did not demonstrate a clear improvement in advanced adenoma detection. For this outcome, the pooled risk ratio was 1.03, with a 95% confidence interval of 0.97 to 1.10. The review also found increased detection of non-neoplastic lesions and suggested that some of the additional adenoma detection involved diminutive lesions.
These findings strengthen two conclusions that are important for clinical governance. First, CADe can improve adenoma detection across different settings. Second, the clinical significance of the additional lesions matters when evaluating the technology.
The absence of a clearly demonstrated improvement in advanced lesion detection across a pooled analysis does not prove that every CADe system lacks such a benefit, nor does it exclude benefits in particular populations or procedural circumstances. It does mean that an increase in overall ADR should not be presented as direct proof of improved cancer outcomes.
The newer evidence therefore supports continued evaluation rather than either uncritical adoption or rejection of AI-assisted colonoscopy.
Why detecting more polyps is not always the same as providing better care
Colorectal lesions differ considerably in size, morphology, histological characteristics and potential clinical consequences. Detecting a diminutive adenoma is not necessarily equivalent to detecting an advanced adenoma or another lesion associated with a greater risk of colorectal cancer.
Many additional findings identified with CADe may be clinically appropriate to remove or assess. Nevertheless, the value of detecting them depends partly on what happens afterward.
Additional polyp detection may lead to more polypectomies, histopathological examinations and recommendations for surveillance colonoscopy. Some of these interventions contribute to effective care. Others may increase resource use or expose patients to procedural risks without a clearly established additional long-term benefit.
Polypectomy can be associated with complications, including bleeding and perforation. Although these risks vary according to lesion characteristics, removal technique and patient factors, they should not be ignored when evaluating a technology that may influence the number of interventions performed.
The potential surveillance burden also deserves attention. Finding additional small adenomas may affect future colonoscopy recommendations, depending on the applicable clinical guidelines and the complete pathological findings. This can create costs and demands on endoscopy capacity, as well as inconvenience and anxiety for patients.
These considerations do not mean that small lesions should be disregarded. They mean that detection needs to be connected to appropriate clinical management.
A hospital should therefore avoid assessing CADe solely by counting highlighted areas, detected polyps or removed lesions. Measures such as adenomas per colonoscopy, advanced adenoma detection, sessile serrated lesion detection and the appropriateness of subsequent management can provide a more meaningful picture.
The goal is not to minimize detection. It is to ensure that additional detection leads to justified clinical decisions and benefits that are proportionate to any additional risks and burdens.
Colonoscopy quality begins before AI is switched on
An AI system can only analyze the visual information available to it. If the colonic mucosa is obscured by residual stool, fluid or inadequate exposure, a lesion may remain invisible to both the endoscopist and the software.
Bowel preparation is therefore fundamental to successful colonoscopy, regardless of whether CADe is available.
The United States Multi-Society Task Force on Colorectal Cancer updated its consensus recommendations on bowel preparation quality in 2025 [7]. The recommendations address appropriate preparation regimens, patient instructions and the importance of achieving sufficient mucosal visualization for the intended examination.
The ASGE/ACG quality indicators for colonoscopy, updated in 2024, also emphasize bowel preparation adequacy as an important component of procedural quality [8].
These are professional references rather than automatically binding legal performance standards for Vietnamese healthcare organizations. Nevertheless, they provide useful guidance for local quality-improvement activities.
Before purchasing CADe, an endoscopy unit should understand how often bowel preparation is adequate and whether the existing preparation process is reliable. If inadequate cleansing frequently compromises procedures, improving patient education, preparation instructions and follow-up arrangements may offer substantial benefits independently of AI.
The same principle applies to examination technique.
A high-quality colonoscopy requires appropriate insertion and withdrawal practices, adequate visualization, systematic mucosal inspection and careful assessment of any suspected lesion. Cecal intubation, when indicated and technically achievable, remains an important element of examination completeness.
CADe may help identify a polyp that appears within the camera's field of view, but it cannot guarantee that all relevant mucosal surfaces have been examined. A lesion hidden behind a fold or located outside the visible field may remain undetected.
This limitation becomes especially important during withdrawal. If the endoscopist relies excessively on AI markers, attention may shift toward highlighted areas at the expense of surrounding mucosa. The technology could then influence inspection behavior in ways that were not intended.
Withdrawal time should also be interpreted carefully. Additional inspection prompted by AI may increase procedure duration, but a longer examination is not necessarily a better examination. Relevant quality definitions, therapeutic activities and procedural circumstances must be considered.
The practical question for an endoscopy unit is whether CADe supports thorough, consistent inspection while preserving the techniques that are already essential to safe colonoscopy.
False-positive alerts and the risk of misplaced attention
False-positive alerts are a familiar challenge in computer-aided detection. A CADe system may highlight a normal mucosal fold, a bubble, residual material, reflected light or another visual feature that resembles a polyp.
An incorrect alert does not necessarily harm a patient. Often, the endoscopist can inspect the area briefly, recognize that no lesion is present and continue the examination.
The concern is what happens when such alerts become frequent or repeatedly interrupt the procedure.
A large number of unhelpful alerts may increase cognitive workload and cause the endoscopist to spend additional time examining normal structures. In some situations, an alert may lead to unnecessary manipulation, biopsy or removal of tissue that would otherwise not have required intervention.
Different false-positive alerts also create different burdens. A marker that briefly appears and disappears over a fold may be less disruptive than an alert that persists or repeatedly directs attention to the same area. Simply counting alerts may not capture these differences.
There are also important human factors to consider.
When clinicians repeatedly encounter unhelpful alerts, they may become less responsive to future alerts, including those associated with genuine lesions. Conversely, clinicians who develop excessive confidence in the technology may pay insufficient attention to areas that have not been highlighted.
Neither response is desirable. CADe should support an endoscopist's observation rather than become the primary determinant of visual attention.
A related concern is the absence of an alert. A system may fail to highlight a visible lesion because of its appearance, viewing angle, image quality or other limitations. Endoscopists must not interpret a silent screen as confirmation that the mucosa is normal.
For these reasons, evaluation of CADe should include how alerts influence operator behavior, not merely whether the software correctly identifies lesions in selected images.
Endoscopist judgment must remain central
The endoscopist remains responsible for the clinical assessment and decisions made during colonoscopy, within the applicable professional and institutional framework.
When CADe highlights a suspected lesion, the operator may need to examine it more closely, improve visualization, adjust the scope position or use another clinically appropriate inspection technique. The decision to proceed with biopsy, removal or other management should depend on the finding itself, the patient's clinical circumstances and relevant professional guidance.
An AI marker alone is not sufficient justification for tissue removal.
This distinction is particularly important when a system includes both CADe and CADx. An application that has demonstrated improved detection performance has not automatically demonstrated reliable optical characterization.
NICE's 2026 assessment specifically recognized uncertainty concerning AI-assisted characterization, including the ability to distinguish different types of colorectal lesions [4]. Evidence supporting lesion detection should therefore not be used to justify additional diagnostic functions without appropriate evaluation.
Where AI-assisted characterization is considered, healthcare organizations need to assess its intended purpose, diagnostic performance, limitations and relationship to established clinical practice. The availability of CADx should not be assumed to remove the need for histopathological examination when this remains indicated.
The same professional responsibility applies when an endoscopist disagrees with an AI output. A suspected lesion should not be dismissed simply because CADe has failed to identify it, and a highlighted area should not automatically be treated as abnormal.
The aim is neither automatic agreement with AI nor systematic rejection of its suggestions. It is to use the information provided by the system while maintaining independent clinical assessment.
Training for AI-assisted endoscopy needs to extend beyond installation
The effective use of CADe involves more than learning how to activate software or recognize an on-screen marker. Clinicians need to understand the technology's capabilities, limitations and potential influence on their clinical behavior.
In February 2026, ESGE published a position statement outlining a curriculum for the safe and effective use of AI in endoscopy [9]. The curriculum addresses preparation for adoption, foundational AI knowledge, practical competencies and the maintenance of competence following implementation.
An important principle is that AI training should build on established endoscopic expertise rather than replace it.
Endoscopists should remain capable of performing high-quality conventional colonoscopy and recognizing lesions independently of AI assistance. They should understand that a functioning CADe system may still miss lesions and that incorrect alerts can occur during ordinary procedures.
Practical training can use examples of genuine lesions, false-positive alerts, difficult-to-visualize abnormalities and cases in which endoscopist observations differ from AI outputs. Reviewing these situations can help clinicians understand when further inspection is warranted and when a marker does not justify additional intervention.
Training should also address the potential effects of prolonged AI use. An operator's response to alerts may change with experience, and excessive dependence on the system may develop gradually rather than during initial deployment.
For this reason, a single installation session is unlikely to address every training need. Endoscopy units should consider whether periodic review, professional feedback and selected case discussions are appropriate as the technology becomes part of routine care.
The ESGE curriculum is a valuable international professional reference. It should not, however, be presented as a mandatory Vietnamese Ministry of Health training protocol or as a substitute for applicable local professional requirements.
What research involving Viet Nam tells us
Research involving artificial intelligence in gastrointestinal endoscopy has been developing in Viet Nam for several years. This provides a useful foundation for understanding local technical capabilities and the practical challenges of clinical implementation.
In December 2024, Dao and colleagues published a narrative review describing research and development activities involving AI in gastrointestinal endoscopy in Viet Nam [10]. The review discussed work on endoscopic image datasets, lesion detection, image segmentation and classification.
Some Vietnamese research groups have developed annotated image datasets and algorithms intended to identify colorectal polyps or assess their characteristics. These activities demonstrate progress in local technical development and collaboration between clinical and technology researchers.
However, many early evaluations have relied on selected images or datasets rather than complete real-time procedures. Performance measured under these conditions cannot automatically be generalized to routine clinical practice.
Real colonoscopy involves changing camera angles, variations in preparation quality, different endoscope platforms, long periods without visible lesions and substantial differences in operator technique. These factors may influence both lesion detection and false-positive alerts.
There is also clinical research involving Vietnamese institutions.
In 2023, Lui and colleagues published a prospective multicenter randomized tandem-colonoscopy trial involving centers in Hong Kong, Ha Noi and Singapore [11].
The study enrolled 223 patients, of whom 216 completed the tandem examination and were included in the comparison. CADe was evaluated during the initial withdrawal examination of the proximal colon, followed by a second conventional white-light examination to assess lesions missed during the first inspection.
The proximal ADR was higher in the CADe group than in the conventional colonoscopy group, at 44.7% versus 34.6%. However, the trial did not demonstrate a statistically significant reduction in proximal adenoma miss rate. The reported miss rates were 20.0% with CADe and 14.0% with conventional colonoscopy, with a P value of 0.07.
These findings illustrate why detection rate and miss rate should not be treated as interchangeable outcomes. Greater detection during an initial examination does not necessarily establish a reduction in the proportion of lesions missed when evaluated using tandem colonoscopy.
The result should not be interpreted as evidence that CADe generally increases missed lesions. It is a finding from a particular trial with a defined population, procedural method and study design.
Because the trial involved three countries, its pooled results should also not be described as outcomes specifically attributable to the Vietnamese participating center.
A different perspective comes from a survey published by Dao and colleagues in the Vietnam Medical Journal in September 2024 [12].
Among 59 participating Vietnamese endoscopists, 57.6% reported experience with AI-integrated endoscopy systems, while 59.3% demonstrated a high level of acceptance according to the study's assessment.
Accurate lesion detection was identified as an important advantage by 72.9% of respondents. Frequent false-positive alerts and associated interruptions were identified as an important perceived disadvantage by 57.6%.
These results provide insight into professional perceptions, but the study involved a relatively small survey group. The percentage of respondents identifying false-positive alerts as a disadvantage is not a measured false-positive rate for a particular CADe product.
Taken together, the available Vietnamese research demonstrates growing clinical and technical interest in AI-assisted gastrointestinal endoscopy. It also highlights the need to distinguish algorithm development, clinician acceptance and clinical effectiveness.
Participation in international research and successful performance on selected datasets do not establish that every commercially available CADe system has demonstrated improved patient outcomes in routine Vietnamese practice.
Evaluating CADe under local clinical conditions
Before introducing CADe into routine care, an endoscopy unit should understand how the proposed system performs within its intended clinical environment.
This assessment begins with the technology itself. Compatibility with the existing endoscopy platform, video-processing configuration, imaging modes and supporting equipment needs to be established. Evidence generated using one combination of hardware and software may not necessarily apply to a different configuration.
The patient population also matters. A system evaluated mainly in screening colonoscopies may not perform identically in diagnostic examinations or in patients with different lesion characteristics and clinical conditions.
Where the manufacturer's intended use excludes certain populations or procedural situations, those limitations should be reflected in local arrangements.
Clinical teams should also distinguish technical testing from evaluation during live procedures. Recorded images and video may help identify obvious compatibility problems or evaluate selected detection characteristics. However, retrospective testing cannot fully reproduce the interaction between an endoscopist and real-time AI alerts.
During live colonoscopy, the operator responds to findings, changes the viewing angle, washes or suctions the mucosa and makes immediate clinical decisions. These interactions can affect both detection and procedural workflow.
A carefully supervised introduction may therefore be useful when evaluating a new system, provided it is conducted within the applicable clinical, institutional, regulatory and research requirements.
Local assessment should pay attention to false-positive alerts, procedural interruptions, system latency, technical reliability and the consequences of equipment or software changes. It should also examine whether clinicians remain comfortable performing the procedure without AI assistance.
There is a broader issue of generalizability. Evidence from one hospital, one patient population or one group of endoscopists does not guarantee identical performance elsewhere.
This does not mean every hospital must conduct its own randomized trial before considering CADe. It means that technology selection and implementation should be proportionate to the intended use, available evidence, clinical risks and local circumstances.
Start with the existing quality of the endoscopy service
A hospital is in a stronger position to evaluate CADe when it already understands the quality of its colonoscopy service.
Relevant baseline measures may include ADR, bowel preparation adequacy, cecal intubation, sessile serrated lesion detection, withdrawal practices and adherence to appropriate surveillance recommendations. The selection of measures should reflect the services provided, available information and relevant professional standards.
Reliable baseline data are essential because changes observed after CADe implementation may have several explanations.
An increase in ADR could reflect AI assistance, but it could also be influenced by improved bowel preparation, additional endoscopist training, changes in patient characteristics or more systematic quality feedback.
Similarly, an apparent difference between endoscopists may reflect case mix and procedural indications as well as individual detection performance.
Organizations should therefore avoid interpreting simple before-and-after comparisons as definitive evidence of AI effectiveness.
Where practical, evaluation should use consistent outcome definitions, appropriate patient groups and methods that account for relevant differences between observation periods.
The organization should also consider whether the expected benefit varies across endoscopists with different levels of experience or baseline performance. Evidence does not support assuming that CADe is useful only for inexperienced operators, nor that all experienced endoscopists will benefit equally.
The purpose of baseline assessment is not to create another administrative reporting exercise. It is to understand what problem the technology is expected to address and whether its introduction produces a meaningful improvement.
Monitoring detection, intervention and patient safety
Once CADe becomes part of routine colonoscopy, monitoring should extend beyond whether the system is functioning or how frequently it is used.
ADR remains an important measure, but it should be interpreted alongside information about the lesions identified and the consequences of their detection.
Adenomas per colonoscopy, advanced adenoma detection and sessile serrated lesion detection may provide additional information. Each measure requires a consistent definition and an appropriate denominator.
The population undergoing examination must also be considered. Screening, surveillance and diagnostic colonoscopies can involve different underlying lesion prevalence and clinical characteristics. Unadjusted comparisons between these groups may be misleading.
Monitoring should also consider the effect of CADe on procedure conduct. Additional inspection, repeated examination of normal structures, technical interruptions and changes in procedure duration may influence efficiency and patient experience.
The number of polypectomies and other interventions deserves attention, particularly when additional detected lesions are small or have uncertain clinical significance.
Histopathological results can help clarify whether increased detection is associated with clinically relevant lesions or additional removal of non-neoplastic tissue. Surveillance recommendations should also be reviewed for consistency with the applicable professional guidance.
Patient-safety monitoring should include relevant procedure-related adverse events and clinically meaningful concerns associated with AI use. Such concerns may include inappropriate intervention, distraction contributing to procedural difficulty or technical malfunction that interferes with clinical work.
Not every transient false-positive alert needs to become an incident report. However, recurrent problems, unexpected system behavior or events with potential clinical consequences should be considered through the organization's established quality and patient-safety arrangements.
Long-term outcomes, including post-colonoscopy colorectal cancer, remain important. Meaningful evaluation of these outcomes requires suitable follow-up, reliable data and careful consideration of other factors affecting cancer risk.
A short pilot study cannot establish that an early increase in ADR has prevented future colorectal cancer.
The most useful monitoring approach is therefore one that connects detection performance with procedural quality, clinical decisions, downstream interventions and patient safety.
A practical approach to CADe alerts during colonoscopy
Endoscopy units can establish a consistent approach to handling CADe alerts within their existing clinical procedures.
When a marker appears, the endoscopist should assess the highlighted area using the actual endoscopic image and determine whether further inspection is warranted. This may involve changing the viewing angle, improving mucosal visualization or using another appropriate endoscopic maneuver.
If closer inspection does not confirm a lesion, the operator may continue the examination without additional intervention. If a lesion is identified, its assessment and management should follow the applicable clinical guidance and the endoscopist's professional judgment.
The absence of an alert is equally important. If the endoscopist identifies a suspicious area that CADe has not highlighted, the finding should still receive appropriate assessment.
Where clinically significant disagreement between endoscopist observations and AI output occurs repeatedly, selected cases may be suitable for review through the unit's quality-improvement process.
Such review should focus on what happened, whether the technology behaved as expected and whether any change to training, configuration or clinical workflow is necessary.
Agreement with AI should not automatically be treated as the correct outcome. A clinician may appropriately reject an incorrect alert or identify a lesion that the system has missed.
This approach is a proposed organizational practice rather than a mandatory Vietnamese Ministry of Health protocol or a validated clinical decision instrument.
Its purpose is to help endoscopy teams use AI information consistently without allowing it to replace clinical judgment.
Patient information and communication about AI use
Patients should receive accurate information about the examinations and interventions they undergo. When CADe is introduced, healthcare organizations should consider how its role is explained within the applicable patient information and consent arrangements.
A patient may reasonably wish to know whether AI will assist the endoscopist, what the technology is intended to do and who remains responsible for clinical decisions.
The explanation should be straightforward. CADe helps identify areas that may contain polyps, but the endoscopist evaluates the findings and determines the appropriate response.
Patients should not be led to believe that AI guarantees detection of every polyp or eliminates the possibility of colorectal cancer after colonoscopy.
Nor should an AI-generated marker be presented as a confirmed diagnosis or evidence that a lesion is cancerous.
Healthcare organizations should avoid promotional statements suggesting guaranteed cancer prevention, superior long-term outcomes or complete elimination of missed lesions when such claims are not supported by appropriate clinical evidence.
The applicable disclosure and consent requirements need to be assessed according to Vietnamese law, the actual use of the technology and the organization's established procedures. It would be inappropriate to assume that every form of AI assistance requires the same additional consent process.
If AI is being evaluated as part of clinical research, rather than solely used within ordinary clinical care, the relevant research ethics, participant information and consent requirements need separate consideration.
Clear communication is important not only for legal and ethical reasons. It also helps patients understand what the technology can reasonably contribute to their care.
Ethical governance and responsible use of endoscopy data
AI-assisted colonoscopy raises questions that extend beyond diagnostic performance. These include professional responsibility, transparency, protection of patient information, equity and the relationship between healthcare organizations and technology providers.
In 2026, the World Endoscopy Organization published an international consensus statement addressing the legal and ethical framework for AI in gastrointestinal endoscopy [13].
The statement identified important considerations involving data governance, medicolegal responsibility, equity and bias. It provides a useful international perspective on the responsible development and clinical use of AI technologies.
However, an international expert consensus does not establish specific legal obligations in Viet Nam.
For Vietnamese endoscopy units, an important practical consideration is how patient data are handled by the proposed system.
Some CADe applications may process endoscopic video locally, while others may involve additional software services, external technical support or other forms of data processing. The actual architecture needs to be understood rather than assumed.
The organization should establish whether identifiable patient information, images or procedure recordings are processed, stored, accessed or transferred outside the clinical environment.
It should also clarify whether the technology provider intends to use clinical data for purposes beyond the immediate examination, including performance improvement, algorithm development or training.
Those additional purposes require their own assessment of the applicable legal and institutional requirements.
Data governance should not be limited to cybersecurity controls. It also concerns who is permitted to access information, why it is processed, how long it is retained and what happens when the service arrangement changes.
Equity is another consideration. Hospitals introducing CADe should understand whether the technology is available consistently to appropriate patients, whether additional charges are involved and how differences in access may affect service delivery.
These questions do not automatically determine whether a system should be adopted. They help ensure that the decision is evaluated within the broader responsibilities of safe and equitable healthcare.
Vietnamese legal and regulatory considerations
The use of AI-assisted colonoscopy in Viet Nam must be considered within the existing legal framework governing medical examination and treatment, artificial intelligence, medical devices and personal data protection.
These are related but distinct areas of regulation. Compliance with one framework does not automatically establish compliance with the others.
Medical examination, treatment and professional responsibility
The Law on Medical Examination and Treatment No. 15/2023/QH15, effective from 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 healthcare practice in Viet Nam [14].
Introducing CADe does not independently authorize a practitioner to perform colonoscopy, interpret endoscopic findings or make treatment decisions outside the applicable professional scope.
Healthcare facilities and practitioners remain responsible for meeting relevant requirements concerning professional practice, permitted medical services and patient safety.
AI assistance should therefore be integrated into existing clinical responsibilities rather than treated as a separate source of professional authority.
Artificial intelligence regulation
Viet Nam's Law on Artificial Intelligence No. 134/2025/QH15 took effect on March 1, 2026 [15]. Decree No. 142/2026/NĐ-CP, effective May 1, 2026, provides further implementing provisions, including provisions relating to risk classification and management [16].
Decision No. 33/2026/QĐ-TTg, effective August 15, 2026, establishes a list of high-risk AI systems [17].
The healthcare entries identified in this list concern specified AI-assisted surgical and robotic surgical systems. Ordinary software used to highlight suspected colorectal polyps during colonoscopy is not separately identified among those healthcare entries.
However, the absence of an explicit CADe entry should not be interpreted as a definitive legal determination that every such product is low-risk.
The classification of an individual system requires consideration of its actual functions, intended purpose, implementation context and the applicable statutory criteria. Particular attention is needed where a system incorporates additional functions or influences clinical actions beyond visual detection assistance.
A system that highlights a suspected lesion should not automatically be equated with AI that performs or controls a surgical intervention.
It is also important to distinguish an organization's clinical risk assessment from the statutory classification of an AI system. A technology may warrant substantial local patient-safety controls even where its legal risk classification differs from that of another medical AI application.
Responsibilities relating to AI classification, documentation, notification, oversight and implementation should be assessed according to the applicable provisions and the respective roles of the provider and deploying organization.
Medical device regulation
Medical device requirements require a separate assessment.
Decree No. 98/2021/NĐ-CP on medical device management, as amended by subsequent regulations including Decrees No. 07/2023/NĐ-CP and No. 04/2025/NĐ-CP, provides an important regulatory framework for medical devices in Viet Nam [18].
Circular No. 24/2026/TT-BYT, effective July 1, 2026, further addresses risk determination and management measures for products classified as medical devices [19].
Whether a particular CADe application falls within medical device regulation, and which obligations apply, depends on its intended medical purpose, functions, regulatory status and the applicable provisions.
Healthcare organizations should therefore examine the product's documentation, intended use, relevant classification and applicable regulatory requirements before routine deployment.
Regulatory authorization obtained in another country may provide useful information, but it should not automatically be treated as sufficient evidence that the product satisfies all Vietnamese requirements.
Personal data protection
The Law on Personal Data Protection No. 91/2025/QH15 and Decree No. 356/2025/NĐ-CP took effect on January 1, 2026 [20, 21].
These provisions may be relevant when AI-assisted endoscopy involves the processing of identifiable patient information, endoscopic images or video associated with an identifiable individual.
Healthcare organizations should assess the lawful basis and purposes of processing, applicable consent requirements or exceptions, access arrangements, information security, retention and the respective responsibilities of the parties involved.
Additional attention may be required where patient data are processed by external technology providers, transferred across borders or proposed for secondary purposes such as algorithm training.
Whether information falls within a particular legal category, and which obligations apply, depends on the nature of the data and the actual processing arrangements.
A general statement that an AI system is used only to assist clinicians is not sufficient to establish compliance with personal data protection requirements.
For hospitals and endoscopy units, the practical implication is that legal assessment should be based on the specific product and deployment model, not simply on the general description of the technology as AI-assisted colonoscopy.
Making CADe work safely in Vietnamese endoscopy units
Responsible implementation begins with a clear clinical purpose. A hospital may be considering CADe because it wants to improve adenoma detection, reduce variation in examination quality or evaluate whether AI contributes to more consistent mucosal inspection.
These are reasonable objectives, but they should be defined before technology selection.
Clinical leaders and endoscopists need to participate in assessing the intended use, supporting evidence and expected effect on patient care. Biomedical engineering and information technology personnel may need to evaluate equipment compatibility, technical support, security and system reliability.
Quality and patient-safety professionals can contribute by helping the organization understand baseline performance, define meaningful outcome measures and establish arrangements for reviewing unexpected problems.
The proposed system should also be evaluated in relation to the resources required for its use. These may include equipment integration, staff training, maintenance, technical support, software updates and ongoing quality monitoring.
A hospital should not assume that the purchase price represents the complete cost of implementation.
Vendor arrangements deserve careful attention. The organization should understand what technical services are included, who is responsible for maintenance, how system problems are reported and how software changes will be communicated.
AI performance may be affected by software updates, modifications to imaging equipment or changes in system configuration. Consequently, clinical performance observed with one version should not automatically be assumed to remain unchanged after a significant modification.
The endoscopy unit should have a practical response when CADe becomes unavailable or produces unreliable output. Endoscopists must remain capable of continuing an appropriate examination using established techniques when AI assistance is not available or should not be relied upon.
A staged introduction, where appropriate and legally permissible, can allow the organization to assess actual clinical use before expanding deployment.
The purpose of this process is not to create unnecessary administrative barriers. It is to make sure that technology adoption is connected to measurable clinical value and existing patient-safety responsibilities.
Successful implementation should not be defined by the number of AI-equipped rooms, the number of procedures performed with AI activated or the number of alerts generated.
It should be judged by whether the technology contributes to better-quality examinations, supports appropriate clinical decisions and operates safely within the healthcare organization's actual environment.
From better detection to better-quality colonoscopy
Computer-aided detection has become an important development in gastrointestinal endoscopy. Clinical trials, evidence syntheses and newer real-world studies show that CADe can improve adenoma detection in a range of settings.
Yet the evidence also makes clear that the clinical value of additional detection requires careful interpretation.
Not every additional lesion carries the same risk. More polypectomies are not automatically better care, and higher ADR has not yet been shown to guarantee reductions in colorectal cancer incidence or mortality attributable to CADe.
International organizations have therefore adopted different positions, ranging from no recommendation to a weak recommendation against routine use, a position favoring assistance in defined screening and surveillance settings, and conditional adoption during evidence generation.
For Vietnamese healthcare organizations, these differences provide a reason to make considered, evidence-informed decisions rather than treat AI-assisted colonoscopy as either an essential upgrade for every facility or a technology that should be rejected.
The foundations of high-quality colonoscopy remain unchanged. Adequate bowel preparation, systematic mucosal inspection, appropriate endoscopic technique, professional judgment and justified clinical management are essential whether CADe is available or not.
Artificial intelligence may strengthen some of these activities, but it cannot replace them.
Hospitals and endoscopy units should evaluate the specific product, understand its limitations, prepare clinicians to use it appropriately and monitor the consequences of its introduction.
Above all, an alert should remain a reason for clinical assessment, not a substitute for it.
The value of AI-assisted colonoscopy should ultimately be judged not by how many additional polyps are detected, but by whether the technology contributes to high-quality examinations, appropriate clinical decisions and safer, better care for patients.
Professional Education Notice
This article is intended for professional education and knowledge exchange. It does not constitute individualized medical, endoscopic, surgical, pathological, legal, regulatory or patient-safety advice.
The opening clinical examples are illustrative and do not represent documented patient cases.
The organizational approaches described in this article are proposed considerations for healthcare quality, patient safety and responsible AI implementation. They are not mandatory Vietnamese Ministry of Health CADe protocols, clinical practice guidelines or independently validated assessment instruments.
This article focuses primarily on computer-aided detection during colonoscopy. It does not establish the clinical safety or effectiveness of computer-aided characterization, optical diagnosis or automated decisions concerning biopsy, polypectomy or surveillance.
International recommendations from AGA, BMJ, ESGE, NICE and other professional organizations reflect particular evidence assessments, populations and healthcare contexts. They should not automatically be interpreted as binding requirements for healthcare organizations in Viet Nam.
Evidence of increased adenoma detection should not be interpreted as direct evidence that CADe reduces colorectal cancer incidence, colorectal cancer mortality or post-colonoscopy colorectal cancer.
Findings from technical evaluations, selected image datasets and clinical studies conducted in particular settings should not be generalized to every AI product, patient population or endoscopy service without appropriate consideration of their limitations.
The regulatory status and applicable obligations of individual AI systems, medical device software, clinical research activities, healthcare services and personal data processing require separate assessment under the relevant Vietnamese legal framework.
References
Sultan S, Shung DL, Kolb JM, et al. AGA Living Clinical Practice Guideline on Computer-Aided Detection–Assisted Colonoscopy. Gastroenterology. 2025;168(4):691–700. DOI: 10.1053/j.gastro.2025.01.002.
Foroutan F, Vandvik PO, Helsingen LM, et al. Computer Aided Detection and Diagnosis of Polyps in Adult Patients Undergoing Colonoscopy: A Living Clinical Practice Guideline. BMJ. 2025;388. DOI: 10.1136/bmj-2024-082656.
Bretthauer M, Ahmed J, Antonelli G, et al. Use of Computer-Assisted Detection (CADe) Colonoscopy in Colorectal Cancer Screening and Surveillance: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement. Endoscopy. 2025;57(6):667–673. DOI: 10.1055/a-2543-0370.
National Institute for Health and Care Excellence. Artificial Intelligence (AI) Technologies to Help Detect or Characterise Colorectal Polyps. HealthTech Guidance HTG773. Published March 19, 2026.
Dominitz JA, Gawron AJ, McKee GB, Hoggatt KJ, Kaltenbach T. Impact of Availability of Computer-Aided Detection Devices on Adenoma Detection During Colonoscopy: A Cluster Randomized Study. Gastroenterology. Published online June 5, 2026. DOI: 10.1053/j.gastro.2026.05.018.
Anthony AE, Ademi A, Flanagan JS, Stopper PB, Rolandelli RH, Nemeth ZH. Real-Time Computer-Aided Detection During Colonoscopy—Effects on Advanced, Diminutive, and Non-Neoplastic Lesion Detection: A Systematic Review and Meta-Analysis of Randomized Trials. Digestive Diseases and Sciences. Published online October 7, 2026. DOI: 10.1007/s10620-026-10269-1.
Jacobson BC, Anderson JC, Burke CA, et al. Optimizing Bowel Preparation Quality for Colonoscopy: Consensus Recommendations by the US Multi-Society Task Force on Colorectal Cancer. Gastrointestinal Endoscopy. 2025;101(4):702–732. DOI: 10.1016/j.gie.2025.02.010.
Rex DK, Anderson JC, Butterly LF, et al. Quality Indicators for Colonoscopy. Gastrointestinal Endoscopy. 2024;100(3):352–381. DOI: 10.1016/j.gie.2024.04.2905.
Mori Y, Kopylov U, Sinonquel P, et al. Curriculum for Safe and Effective Use of Artificial Intelligence in Endoscopy: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement. Endoscopy. 2026;58(2):202–210. DOI: 10.1055/a-2742-4342.
Dao HV, Nguyen BP, Nguyen TT, et al. Application of Artificial Intelligence in Gastrointestinal Endoscopy in Vietnam: A Narrative Review. Therapeutic Advances in Gastrointestinal Endoscopy. 2024;17:26317745241306562. DOI: 10.1177/26317745241306562.
Lui TKL, Hang DV, Tsao SKK, et al. Computer-Assisted Detection Versus Conventional Colonoscopy for Proximal Colonic Lesions: A Multicenter, Randomized, Tandem-Colonoscopy Study. Gastrointestinal Endoscopy. 2023;97(2):325–334.e1. DOI: 10.1016/j.gie.2022.09.020.
Đào Việt Hằng, Nguyễn Thị Huyền Trang, Vũ Thị Thu Uyên, et al. Trải nghiệm và sự chấp nhận của các bác sĩ nội soi về việc triển khai hệ thống tích hợp thuật toán trí tuệ nhân tạo hỗ trợ trong nội soi tiêu hóa tại Việt Nam. Tạp chí Y học Việt Nam (Vietnam Medical Journal). 2024;542(3). DOI: 10.51298/vmj.v542i3.11223.
Ahmad OF, Mori Y, Bretthauer M, et al. The Legal and Ethical Framework for Artificial Intelligence in Gastrointestinal Endoscopy: A World Endoscopy Organization International Consensus Statement. Annals of Internal Medicine. 2026;179(2):270–275. DOI: 10.7326/ANNALS-25-03415.
National Assembly of Viet Nam. Law No. 15/2023/QH15 on Medical Examination and Treatment. Adopted January 9, 2023; effective January 1, 2024, as subsequently amended. Consolidated Document No. 26/VBHN-VPQH, issued February 26, 2026.
National Assembly of Viet Nam. Law No. 134/2025/QH15 on Artificial Intelligence. 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 No. 91/2025/QH15 on Personal Data Protection. 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.