Gastroenterology · Medical Literature
Artificial Intelligence Versus Conventional Scoring for Assessing Small Bowel Capsule Endoscopy Cleanliness in Crohn's Disease.
Abstract
Evaluation of cleanliness in capsule endoscopy is essential to validate the exam. However, reliable evaluation of bowel cleanliness remains challenging. The validated KODA score provides robust assessment but is time-consuming and impractical in routine practice. We aimed to evaluate the performance of an AI-based tool (AXAROlite, Augmented Endoscopy) compared with the KODA score for assessing small bowel cleanliness in CD patients. This was a post hoc analysis of a multicenter randomized controlled trial (NCT05117996) including 142 CD patients undergoing SBCE after either a standard PEG-based or simplified clear liquid preparation. All videos were evaluated for cleanliness using the KODA score by trained readers. The same videos were analyzed using the AI-based AXAROlite tool. Correlation, agreement, and diagnostic performance of AXAROlite were compared with the KODA score. Clinical factors influencing AI-assessed cleanliness were also investigated. AXAROlite and KODA showed strong correlation for whole small bowel (ρ = 0.61; p 2.25 (AUROC = 0.85, 95% CI 0.79-0.92), with an optimal threshold of 72% clean frames (sensitivity 86%, specificity 72%). Mean cleanliness decreased progressively from proximal to distal small bowel with both methods (p = 0.001). The AXROlite score was independently reduced in patients with active disease and a longer small bowel transit time. AXAROlite provides a rapid, fully automated, and accurate evaluation of small bowel cleanliness in CD, comparable to the validated KODA score. Its integration into routine workflows may streamline reporting and reduce inter-observer variability.
Clinical relevance
Does the AI-based AXAROlite tool provide accurate and clinically comparable assessment of small bowel cleanliness relative to the validated KODA score in patients with Crohn's disease undergoing small bowel capsule endoscopy (SBCE)? __SNAPSHOT__ {"studyDesign":"Post hoc analysis of a multicenter randomized controlled trial (NCT05117996); diagnostic accuracy and correlation study","setting":"Multicenter; France (institutional affiliation includes Nantes Université and collaborating centers); specific number of centers not reported in abstract","sampleSize":"n=142 Crohn's disease patients (drawn from the parent RCT; intervention/control subgroup sizes not reported in this post hoc analysis)","population":"Adult patients with established Crohn's disease undergoing small bowel capsule endoscopy (SBCE); specific inclusion/exclusion criteria beyond CD diagnosis and SBCE indication not reported in abstract","intervention":"AI-based automated cleanliness assessment using AXAROlite (Augmented Endoscopy), outputting percentage of clean frames per video segment","comparator":"KODA score assessed by trained human readers — the validated reference standard for small bowel capsule endoscopy cleanliness","primaryOutcome":"Correlation and agreement between AXAROlite and KODA scores; diagnostic accuracy of AXAROlite for detecting adequate cleanliness (defined as KODA > 2.25), measured by AUROC","followUp":"Not reported (single SBCE assessment per patient; no longitudinal follow-up described)","fundingSource":"Not reported in abstract; notable conflict of interest: lead author Romain Leenhardt is disclosed as co-founder and shareholder of Augmented Endoscopy, the company that developed AXAROlite"}
Key findings
AXAROlite and KODA scores demonstrated a statistically significant strong correlation across the whole small bowel (Spearman ρ = 0.61; p 2.25 — AXAROlite achieved an AUROC of 0.85 (95% CI: 0.79–0.92), indicating excellent discriminatory performance. The optimal operating threshold was identified at 72% clean frames, yielding a sensitivity of 86% and specificity of 72%. Cleanliness decreased progressively from the proximal to the distal small bowel with both assessment methods (p = 0.001). On multivariable analysis, AXAROlite-derived cleanliness scores were independently reduced in patients with active Crohn's disease and those with longer small bowel transit times. NNT/NNH are not applicable to this diagnostic accuracy study. No safety adverse events related to the AI tool were reported (device-based assessment only). Absolute agreement statistics (e.g., kappa, intraclass correlation coefficient) beyond the Spearman correlation are not reported in the abstract.
Practice implications
This study establishes that AXAROlite provides automated small bowel cleanliness scoring that correlates well with the validated KODA reference standard and achieves excellent AUROC performance (0.85) for identifying adequate preparation in a Crohn's disease population. The 86% sensitivity at the optimal threshold is clinically meaningful, suggesting few inadequately prepared studies would be missed; however, the 72% specificity implies a non-trivial false-positive rate (i.e., studies flagged as inadequately prepared that KODA would judge as adequate), which could prompt unnecessary repeat procedures. Importantly, this is a post hoc analysis — not a prospectively powered diagnostic study — so these findings are exploratory. The population is restricted to CD patients undergoing SBCE, limiting generalizability to non-CD indications (e.g., obscure GI bleeding, polyposis syndromes). The finding that active disease independently reduces AI-assessed cleanliness warrants attention: disease-related mucosal changes (ulceration, hemorrhage) may confound the AI's cleanliness algorithm, potentially misclassifying an adequately prepared bowel as poorly prepared in inflamed segments. Only the abstract was available for review; full methodology, inter-reader KODA reliability data, segment-specific AUROC values, and the AI's processing time versus KODA are not available for scrutiny. A significant conflict of interest exists, as a co-author is co-founder and shareholder of the company that developed AXAROlite, which necessitates independent replication before broad adoption. __LIMITATIONS__ Post hoc analysis of an RCT not originally designed to evaluate AI diagnostic accuracy; all diagnostic estimates should be considered hypothesis-generating.,Restricted to Crohn's disease patients only; performance in non-CD SBCE indications is unknown and cannot be assumed.,Significant financial conflict of interest: lead methodologist (Leenhardt) is co-founder and shareholder of Augmented Endoscopy, manufacturer of AXAROlite — independent external validation is essential.,KODA score used as reference standard was assessed by 'trained readers,' but inter-reader reliability metrics for the KODA scoring in this dataset are not reported in the abstract, making the reliability of the gold standard uncertain.,Specificity of 72% at the optimal threshold suggests a meaningful false-positive rate that could result in unnecessary repeat procedures in clinical practice.,Active Crohn's disease independently confounds AI-assessed cleanliness, introducing disease-related bias that may undermine the tool's reliability in the highest-acuity patients — those most likely to undergo SBCE.,Sample size (n=142) is modest for a diagnostic accuracy study; confidence intervals around sensitivity and specificity are not reported in the abstract.,No comparison of AXAROlite processing time versus KODA assessment time is quantified in the abstract, limiting evidence for the claimed efficiency benefit.,Absolute agreement metrics (e.g., kappa, Bland-Altman, ICC) beyond Spearman correlation are not reported in the abstract.,Only the abstract was available for this review; full-text analysis may reveal additional methodological details or limitations not captured here. __BOTTOM_LINE_CATEGORY__ promising_but_preliminary __BOTTOM_LINE__ AXAROlite demonstrates strong correlation with the KODA score (ρ=0.61) and excellent AUROC (0.85, 95% CI 0.79–0.92) for detecting adequate small bowel cleanliness in Crohn's disease, representing a promising fully automated alternative to time-consuming human scoring. However, given that this is a post hoc analysis with a significant author conflict of interest (developer co-authorship), a specificity of only 72%, and performance confounded by active disease, independent prospective validation across diverse SBCE indications is required before routine clinical deployment. Clinicians should view these results as hypothesis-generating rather than sufficient evidence to replace KODA-based human review in current practice.