AnX Robotica, the Plano, Texas-based developer of gastrointestinal imaging technologies, has highlighted United States Food and Drug Administration-documented clinical performance data showing that its NaviCam ProScan artificial intelligence-assisted reading software reduced mean small-bowel capsule endoscopy review time by 88.6% in a 133-patient prospective study. The October 11, 2026 announcement follows the FDA’s September 25 clearance of CapsoVision’s competing CapsoView and CapsoCloud AI-assisted reading software, whose separate reader study demonstrated a 24.7% reduction in mean review time.
The comparison draws attention to a potentially important operational benefit of artificial intelligence in capsule endoscopy, where clinicians must review extensive image sequences to identify abnormalities associated with suspected gastrointestinal bleeding. In the study supporting ProScan’s original regulatory authorisation, mean review time declined from 33 minutes 42 seconds with conventional reading to three minutes 50 seconds with AI assistance. CapsoVision’s more recent submission documented a reduction from 45.7 minutes to 34.4 minutes, although the studies used different patients, capsule systems, reader arrangements and evaluation methods.
The regulatory context is important because ProScan did not receive a new FDA clearance in October 2026. The FDA granted its De Novo classification request on December 12, 2023, establishing a regulated category for gastrointestinal capsule endoscopy analysis software. The recent development involves CapsoVision’s September 2026 510(k) clearance, which identified NaviCam ProScan as its primary predicate device when demonstrating substantial equivalence.
AnX Robotica’s announcement therefore revisits established performance evidence in response to a new competitive regulatory milestone. The figures highlight the potential for substantial time savings, but they do not demonstrate that ProScan is clinically superior to CapsoVision’s technology or that either platform will deliver identical results across all hospitals. The more consequential questions concern how reliably AI-assisted workflows identify significant lesions, how clinicians interact with the software and whether faster interpretation improves healthcare delivery without compromising diagnostic performance.
How did the ARTIC clinical study demonstrate faster ProScan-assisted capsule endoscopy interpretation?
The evidence supporting ProScan’s review-time claim comes from the ARTIC clinical study, a prospective multicentre investigation involving patients undergoing small-bowel capsule endoscopy for suspected bleeding. The study evaluated the NaviCam SB imaging system with and without ProScan assistance, allowing researchers to compare conventional physician interpretation with a workflow incorporating AI-selected images. The final analysis included 133 patients, providing a clinical dataset for assessing both interpretation time and the identification of potentially significant small-bowel lesions.
Conventional reading required a mean of 33 minutes 42 seconds, while AI-assisted review required three minutes 50 seconds. The difference represents almost 30 minutes of reading time saved per examination, with a statistically significant result reported in the FDA’s De Novo review. Expressed proportionally, the reduction was 88.6%, meaning that the AI-assisted workflow required approximately one-ninth of the conventional mean review time observed in that study.
The findings were subsequently published in The Lancet Digital Health in May 2024, providing peer-reviewed evidence beyond the original regulatory submission. The publication described a study conducted across 14 European centres, with participants undergoing capsule endoscopy for suspected small-bowel bleeding following negative upper and lower gastrointestinal examinations. The research evaluated the performance of physician-assisted AI interpretation rather than the ability of an autonomous algorithm to diagnose disease without clinical oversight.
This distinction is especially relevant to how the time-saving figure should be understood. The AI-assisted workflow allowed clinicians to focus their initial assessment on images selected by the algorithm, rather than systematically examining the full conventional sequence in the same manner. The observed efficiency therefore reflects a particular combination of software processing and physician interaction, not simply an acceleration of conventional image scrolling.
The measured review-time reduction is substantial within the study setting, but it should not be extrapolated directly into an equivalent reduction in overall procedure duration. Capsule ingestion, image acquisition, data transfer, video processing, clinical reporting and patient management involve additional steps that were not eliminated by the software. The operational benefit is therefore most directly associated with the physician’s image-review workload.
Did ProScan improve diagnostic yield while reducing the time needed to review capsule images?
The ARTIC results also provided evidence concerning the identification of clinically relevant small-bowel abnormalities. With conventional reading, significant lesions were identified in 62.4% of patients, compared with 73.7% using the AI-assisted workflow. The 11.3-percentage-point difference favoured AI-assisted reading, with the peer-reviewed publication reporting statistically significant superiority for the study’s primary per-patient assessment of lesions with moderate or high bleeding potential.
The study used the Saurin classification framework to distinguish lesions according to their potential relevance to gastrointestinal bleeding. P1 lesions represent findings with uncertain or intermediate bleeding potential, while P2 lesions are considered more strongly associated with bleeding. Combining these categories allowed researchers to evaluate a broader group of potentially important abnormalities, although performance for different lesion categories requires separate interpretation.
The results suggest that concentrating clinician attention on algorithm-selected images did not necessarily require sacrificing diagnostic yield to achieve faster review. Indeed, the physician-assisted AI approach identified more patients with relevant abnormalities than conventional reading in this study. However, the findings should not be converted into a general claim that ProScan detects all clinically important lesions or independently establishes the correct diagnosis in every patient.
The FDA’s original review provides an important qualification. An expert assessment identified relevant small-bowel pathology in 105 patients, while AI-assisted physicians identified abnormalities in 98 patients and conventional readers identified them in 83. The AI-assisted approach therefore missed seven patients with pathology identified by the expert assessment, compared with 22 missed through conventional interpretation.
Those figures demonstrate an improvement in the observed study performance while confirming that missed findings remained possible. The FDA also noted uncertainty surrounding the performance of ProScan independently of clinician input in the clinical evaluation. As a result, the established evidence concerns a combined AI-and-physician workflow rather than autonomous diagnostic accuracy.
Why is comparing ProScan’s 88.6% time saving with CapsoVision’s 24.7% reduction potentially misleading?
The central limitation of AnX Robotica’s October announcement is that the two reported efficiency figures originate from separate investigations. Although both studies examined capsule endoscopy review time with and without artificial intelligence, they did not use the same patients, capsule imaging hardware, physicians or study methodology. A numerical comparison can describe the published results, but it cannot establish comparative superiority in the absence of appropriately matched evidence.
ProScan’s supporting ARTIC investigation was prospective and involved 133 patients with suspected small-bowel bleeding. CapsoVision’s FDA submission included a retrospective, multi-reader, multi-case evaluation using 111 capsule endoscopy videos collected through its United States commercial clinical workflow. The CapsoVision reader study involved 15 board-certified gastroenterologists, with participants reviewing conventional and AI-assisted examinations in a randomised crossover arrangement incorporating a washout period.
The baseline review-time difference illustrates one challenge in comparing the findings. Conventional review required an average of approximately 33.7 minutes in the ProScan study, compared with 45.7 minutes in the CapsoVision evaluation. These starting points may reflect differences in imaging characteristics, image volume, physician behaviour, review procedures and patient-related factors.
The studies also differed in their patient populations. ProScan’s original clinical investigation focused on adults with suspected small-bowel bleeding, while CapsoVision’s evaluation included a broader age distribution, with participants as young as four years and cases with both bleeding-related and other examination indications. Those differences complicate attempts to interpret the percentage reductions as measurements of the same clinical task under equivalent conditions.
CapsoVision’s AI-assisted workflow reduced mean review time from 45.7 minutes to 34.4 minutes, saving approximately 11.3 minutes per examination. Although that proportional improvement is smaller than the reduction reported for ProScan, the CapsoVision study also demonstrated changes in diagnostic performance within its own controlled reader evaluation. The two datasets therefore support conclusions about each product’s performance relative to its respective conventional workflow, rather than an evidence-based ranking between the products.
A direct comparative assessment would ideally evaluate both technologies under a prespecified protocol using comparable cases, reader expertise, reference standards and consistent definitions of review time. Because the platforms operate with different capsule imaging systems, such a study would also need to account for hardware and workflow differences rather than attributing every observed result to the AI software alone.
What did CapsoVision’s FDA-cleared AI Highlights study demonstrate beyond reading efficiency?
CapsoVision’s September 2026 clearance covered the introduction of AI-assisted functions into its CapsoView and CapsoCloud software for the CapsoCam Plus capsule endoscopy system. The features include AI Highlights, which identifies suspected abnormalities, Suggested Landmarks, which assists with anatomical orientation, and Green Detection, which identifies obscured or uninterpretable images. Together, these tools are intended to support interpretation and streamline the review of capsule endoscopy examinations.
The FDA submission reported an AI-assisted patient-level sensitivity of 92.3%, compared with 71.0% during conventional reading. The difference was 21.3 percentage points, with the company reporting that the study met its relevant non-inferiority and superiority assessments. Diagnostic yield also increased from 63.3% to 87.6% across the study’s respective reading conditions.
These findings demonstrate that the CapsoVision software produced measurable diagnostic performance improvements within its own evaluation, despite recording a smaller proportional reduction in review time than ProScan’s earlier study. However, its diagnostic-yield and sensitivity figures cannot be directly compared with the percentages reported for ProScan because the studies used different designs and reference standards. In particular, diagnostic yield describes the proportion of examinations identifying relevant findings, whereas sensitivity evaluates the proportion of reference-positive cases correctly identified.
The CapsoVision filing also documents an important trade-off involving specificity. Patient-level specificity declined from 56.2% during conventional reading to 26.4% with AI assistance, indicating that the enhanced sensitivity was accompanied by a greater tendency to classify examinations as positive. This matters because a system that highlights more possible abnormalities can increase the number of findings requiring clinician assessment, even while improving the identification of relevant pathology.
The standalone algorithm evaluation provides further context. CapsoVision reported 100% patient-level sensitivity and 0% specificity for its AI Highlights function, consistent with software designed to flag potential findings for review rather than provide a definitive patient-level diagnosis. The result reinforces the importance of distinguishing the algorithm’s screening behaviour from the final diagnostic performance of clinicians using its output.
The FDA clearance therefore supports use of the software as an adjunctive review tool rather than as an autonomous replacement for gastroenterologists. Its value will depend on how effectively the combination of image prioritisation, anatomical assistance and physician interpretation operates in routine clinical practice.
What does the FDA’s De Novo and 510(k) pathway reveal about the two capsule endoscopy platforms?
NaviCam ProScan received its original De Novo classification from the FDA on December 12, 2023, under submission DEN230027. The regulatory applicant was Ankon Technologies Co., Ltd., while AnX Robotica markets and supports the technology in the United States. The decision established the Class II category for gastrointestinal capsule endoscopy analysis software devices, creating a regulatory framework against which subsequent products could be evaluated.
CapsoVision’s more recent 510(k) clearance, identified as K254230, was issued on September 25, 2026. Its submission identified NaviCam ProScan as the primary predicate device and Deep Capsule from DigestAID as a reference device. The FDA determined that CapsoView and CapsoCloud were substantially equivalent for their specified intended purposes, allowing the cleared software functions to be marketed in the United States.
The use of ProScan as a predicate demonstrates regulatory relevance, but it does not establish that the FDA considers ProScan clinically superior to the newer platform. A 510(k) substantial-equivalence determination concerns the applicable regulatory requirements, intended use and technological or performance considerations needed to establish an appropriate basis for clearance. It is not a comparative endorsement of one company’s commercial product over another.
The authorised patient populations also differ in an important respect. ProScan’s original De Novo indication concerns adult patients whose capsule endoscopy images were obtained for suspected small-bowel bleeding. The CapsoVision clearance covers adult and pediatric patients older than two years undergoing the specified examinations for suspected small-bowel bleeding.
The broader age coverage of CapsoVision’s software is a regulatory distinction, not proof that it delivers better diagnostic outcomes in every age group. Similarly, the longer clinical history of ProScan does not automatically establish stronger performance across all possible gastrointestinal indications. Both platforms remain subject to the limitations of their respective labelling, and neither has been cleared as a standalone diagnostic replacement for physician assessment.
Why do false negatives, clinician oversight and post-market validation remain important for capsule endoscopy AI?
The principal clinical risk associated with an AI-assisted capsule endoscopy workflow is that relevant pathology may be overlooked if the software fails to highlight an important image or if clinicians rely excessively on algorithm-generated selections. Capsule examinations can contain extensive sequences of images, and potentially significant lesions may appear only briefly or in visually challenging conditions. Although automated image prioritisation can improve efficiency, it also introduces dependence on the reliability of the algorithms used to select and classify frames.
The FDA’s ProScan De Novo assessment specifically identified the risk of overreliance on AI output and the possibility of false-negative or false-positive findings. Its labelling requires clinicians to undertake their own assessment and review the complete video when clinically appropriate. A negative assessment generated by ProScan alone does not rule out small-bowel disease, while a positive AI-identified image does not independently establish a diagnosis.
The regulator also identified uncertainty concerning the contribution of the software separately from physician interpretation in the original clinical study. The De Novo decision included post-market clinical validation requirements intended to provide additional information about lesion detection, anatomical location recognition and performance across relevant analysis levels. Those requirements reflect the distinction between demonstrating that an AI-assisted workflow performs adequately and fully characterising the independent capabilities of the algorithm.
The October 2026 announcement did not provide a new independent validation dataset or updated post-market performance results. Its principal evidence remains the original FDA-documented studies and the subsequently published ARTIC investigation. This does not invalidate the reported findings, but it limits the extent to which the announcement can establish additional clinical progress since the initial clearance.
For CapsoVision, performance across different demographic and clinical subgroups will also be important as the software enters wider use. Its regulatory submission includes testing across United States regions, patient ages and bowel preparation conditions, but future clinical experience may identify situations in which algorithm performance varies. The reliability of anatomical localisation and the handling of obscured images may be especially important when poor image quality complicates interpretation.
These considerations explain why independent validation, ongoing monitoring and appropriately qualified users remain central to the safe implementation of medical AI. Faster interpretation can improve the experience of reviewing capsule examinations, but the diagnostic responsibilities remain with the clinician.
Can AI-assisted capsule endoscopy improve hospital productivity and reduce the cost of gastrointestinal investigations?
Capsule endoscopy can require substantial specialist interpretation time because examinations generate large volumes of images that must be assessed for bleeding sources and other abnormalities. In gastroenterology services managing high examination volumes, reducing physician review time could increase the number of studies that can be interpreted within available staffing capacity. Such improvements may be particularly relevant to centres experiencing delays in image reporting or difficulties allocating specialist time.
The approximately 30-minute mean saving observed in ProScan’s ARTIC study illustrates the potential scale of the opportunity under similar review conditions. If replicated consistently, shorter interpretation sessions could allow clinicians to redirect time towards reporting, patient assessment or other clinical responsibilities. However, this remains an operational scenario rather than evidence that hospitals will automatically achieve equivalent productivity gains.
Actual savings will depend on the surrounding workflow. Capsule examinations still require patient preparation, equipment management, data retrieval, image processing, reporting and clinical follow-up. Software licensing, integration, clinician training, technical support and compatibility with existing hardware may introduce additional costs that must be considered alongside interpretation efficiency.
A hospital evaluating the technology would therefore need to calculate its total operational effect rather than relying exclusively on the headline percentage reduction. Relevant measures could include completed examinations per reader session, reporting turnaround time, software operating costs, time spent reviewing additional flagged images and the frequency of clinically important missed findings. A faster image review is valuable only if the quality and reliability of the resulting clinical interpretation are maintained.
The potential economic benefits also differ between facilities. High-volume capsule endoscopy centres may be better positioned to capture efficiencies through repeated use, while smaller providers may have fewer examinations over which to distribute software and implementation costs. Institutions using an existing capsule platform may also face additional purchasing or integration considerations when evaluating an AI system tied to a different manufacturer’s hardware.
Neither the ProScan nor CapsoVision regulatory performance figures independently establish a reduction in hospital expenditure, downstream procedures or patient treatment costs. Those outcomes would require economic and implementation studies examining complete clinical pathways. Nevertheless, the reported reductions in interpretation time provide a reasonable foundation for assessing whether AI-assisted capsule reading can improve service capacity.
How will competition between ProScan and CapsoVision influence adoption of AI gastrointestinal imaging?
The September 2026 FDA clearance introduces another regulated AI-assisted option into the United States capsule endoscopy market. AnX Robotica’s response highlights how device manufacturers may increasingly compete through published clinical evidence, regulatory performance data and workflow efficiency rather than capsule imaging specifications alone. The availability of comparable regulatory documents also gives healthcare purchasers a more detailed basis for evaluating individual products.
ProScan can point to its original De Novo clearance, a prospective multicentre study and a peer-reviewed publication demonstrating substantial improvements in review efficiency and diagnostic yield. CapsoVision can point to a more recent reader evaluation, FDA clearance for its integrated software functions and compatibility with its CapsoCam Plus ecosystem. These are relevant product distinctions, but they do not amount to an independently established commercial or clinical ranking.
Integration may become an important factor in purchasing decisions because AI-assisted capsule interpretation is generally tied to specific imaging systems and software environments. Hospitals already using one manufacturer’s capsules may place considerable value on maintaining compatibility, familiar workflows and existing service arrangements. Switching to another platform could involve considerations extending beyond the performance of the AI algorithm itself.
CapsoVision is publicly traded on Nasdaq under the ticker CV, while AnX Robotica is privately held. For CapsoVision, the newly cleared AI functionality adds to the commercial proposition for its capsule endoscopy platform, although the approval does not establish immediate sales growth, improved margins or a particular level of market adoption. The October announcement from AnX Robotica likewise does not provide updated revenue, installed-base or market-share figures demonstrating the commercial effect of ProScan’s clinical performance.
The larger competitive question concerns which technologies can demonstrate reproducible improvements under routine conditions. Procurement teams are likely to evaluate not only review-time reductions and diagnostic accuracy, but also evidence quality, training requirements, reporting integration, technical reliability and the extent to which the software supports established clinical responsibilities.
What evidence will determine whether ProScan’s 88.6% time saving translates into routine clinical value?
The October 11 announcement reinforces the significance of an established finding: in a prospective clinical investigation, ProScan-assisted interpretation substantially reduced mean small-bowel capsule reading time while improving the diagnostic yield of clinically relevant lesions compared with conventional reading. The result is supported by FDA documentation and peer-reviewed research, making it more substantive than an unvalidated product-performance assertion. However, the figures describe performance within a defined study population and workflow, not a guaranteed result for every gastroenterology service.
The competing CapsoVision study demonstrates that another FDA-cleared system can also improve physician interpretation efficiency and diagnostic sensitivity, although its observed reduction in review time was smaller under different testing conditions. Both submissions show that AI-assisted capsule endoscopy has progressed beyond theoretical capability into clinically evaluated software applications. Neither submission establishes that one platform is superior to the other when evaluated under equivalent conditions.
Further evidence would be particularly useful in three areas: independent prospective validation under routine clinical conditions, reproducible diagnostic performance across varied patient populations and measurement of the effect on complete reporting workflows. Longer-term implementation studies could also clarify whether time savings persist after initial deployment and whether AI-assisted interpretation influences the speed at which clinically actionable findings reach treating physicians.
For AnX Robotica, the next meaningful milestone would be updated evidence demonstrating that ProScan’s performance remains reliable outside the original ARTIC study and across routine clinical environments. For CapsoVision, broader clinical experience will help determine whether the newly cleared software’s diagnostic and workflow benefits translate consistently into everyday use.
The emerging evidence supports a practical role for artificial intelligence in making small-bowel capsule endoscopy interpretation more efficient. The decisive measure of progress, however, will not be which manufacturer reports the largest percentage reduction from an independent study. It will be whether healthcare providers can consistently interpret examinations faster, identify clinically relevant abnormalities accurately and preserve physician oversight while integrating these tools into dependable patient-care workflows.
