Technical Safety And Diagnostic Accuracy Of AI-Assisted Digital Single Operator Cholangioscopy During Real-time Assessment Of Indeterminate Biliary Strictures: The ARTICO Multicentric, Randomized, Controlled, Crossover Diagnostic Accuracy Trial

Aims Even though DSOC offers a high-quality image for endoluminal identification of biliary strictures, an adequate sample acquisition via this technique remains challenging, especially in indeterminate biliary strictures (IBDS) [1]. The narrow cholangioscope diameter along with the small biopsy forceps lead to an insufficient amount of specimen obtained for pathological analysis [2]. Therefore, optimized visually targeted biopsy sampling can improve diagnostic accuracy given its high sensitivity and specificity [3]. The newly developed artificial intelligence (AI) algorithm for the identification of malignant biliary features plays an important role in the assessment of this pathology through a computer-based targeted biopsy. This study aims to evaluate the accuracy of an AI-Cholangioscopy software guided biopsies during real-time DSOC assessment.

Methods An independent, experimental, single-centre, randomized, controlled, diagnostic accuracy trial was conducted to assess the diagnostic accuracy of the AI-guided biopsies compared to AI-unassisted biopsies during real-time DSOC. Adult patients with an indication for DSOC were randomly assigned to either an AI-assisted DSOC (study group) or a standard AI-unassisted DSOC (control group). The primary outcome was to evaluate the accuracy of the AI algorithm in guiding biopsy sample collection during DSOC. An adequate sample was defined as a specimen containing biliary epithelium greater than 0.1mm in size. Diagnostic accuracy was assessed for both groups. A p-value<0.05 was considered statistically significant

Results A total of 65 patients were included in the study (36 AI-assisted vs 29 AI-unassisted), median age was 65 (54-76) and 49.2% were female. The most common DSOC indication was tumour suspicion (41.5%). The diagnostic accuracy of the AI-assisted compared to the AI-unassisted biopsy was 100% vs 90.1% sensitivity, 93.8% vs 71.4% specificity, 95.2% vs 90.9% PPV, 100% vs 71.4% NPV, and 97.2% vs 86.2%, with an area under the ROC curve (AUROC) of 0.98 vs 0.81 (p=.111). When comparing the correlation of follow-up vs AI-assisted or AI-unassisted biopsy, the AI obtained Kendall’s correlation (tau) of 0.94 vs tau 0.62 reached by the endoscopist (p<.001). The AI-assisted biopsy obtained an F1-score of 97.6% for predicting follow-up, with a Matthews correlation coefficient (MCC) of 94.5%, higher than the 90.9% and 62.3% reached by the endoscopist, respectively. Only one patient presented insufficient sample biopsy, coming from the AI-unassisted group.

Conclusions AI-assisted biopsy significantly enhances biopsy diagnostic accuracy compared to AI-unassisted biopsy. In addition to the assisted assessment of the biliary tract, AI integration improves the accuracy and reliability of the respective biopsies, reducing the risk of false negatives and insufficient samples.

 

MissingLinkText

Conflicts of Interest

Carlos Robles-Medranda is a key opinion leader and consultant for Pentax Medical, Steris, Medtronic, Motus, Micro-tech, G-Tech Medical Supply, CREO Medical, EndoSound, and mdconsgroup. All other authors have nothing to declare.

• References

• 1 Milluzzo SM, Landi R, Perri V, Familiari P, Boškoski I, Pafundi PC, Farina A, Ricci R, Spada C, Costamagna G, Tringali A.. Diagnostic accuracy and interobserver agreement of cholangioscopy for indeterminate biliary strictures: A single-center experience. Dig Liver Dis 2024; 56 (05): 847-852 Add to Citavi project by DOI. Epub 2023 Nov 27. PMID: 38016895
• 2 Kobayashi K, Kobara H, Masaki T.. Novel cholangioscopy-guided targeted biopsy for diagnosing gallbladder carcinoma. Dig Endosc 2021; 33 (04): e51-e53 Add to Citavi project by DOI. Epub 2021 Jan 28 PMID: 33506562
• 3 Inchingolo R, Acquafredda F, Posa A, Nunes TF, Spiliopoulos S, Panzera F, Praticò CA.. Endobiliary biopsy. World J Gastrointest Endosc 2022; 14 (05): 291-301 Add to Citavi project by DOI 35719901 Add to Citavi project by Pubmed ID; PMCID: PMC9157693

Publikationsverlauf

Artikel online veröffentlicht:
27. März 2025