Integrating Artificial Intelligence into Endoscopic Decision-Making: ANN-Assisted CT for Common Bile Duct Stones

Integrating Artificial Intelligence into Endoscopic Decision-Making: ANN-Assisted CT for Common Bile Duct Stones | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Integrating Artificial Intelligence into Endoscopic Decision-Making: ANN-Assisted CT for Common Bile Duct Stones Hwehoon Chung, ChanWoo KWAK, Sang Deok Shin, Jae Guk Kim, Hyun Young Choi, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7919357/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Computed tomography (CT) is widely used in the initial evaluation of suspected common bile duct (CBD) stones, but limited sensitivity often necessitates additional endoscopic procedures. We developed and validated an artificial neural network (ANN) to enhance CT interpretation and assessed its potential to support endoscopic decision-making. Methods We An ANN model integrating UNETR for segmentation and ResNet50 for classification was trained to detect CBD stones on CT. Patients who underwent abdominal CT for suspected CBD stones between March 2018 and June 2023 at Hallym University Kangnam Sacred Heart Hospital were included. A retrospective derivation cohort (n = 830) was used for model training, and a prospective validation cohort (n = 225) for testing, with endoscopic retrograde cholangiopancreato- graphy (ERCP) serving as the reference standard. ANN performance was compared with that of expert radiologists and trainee radiologists with ANN assistance. Multivariate analysis evaluated clinical factors influencing diagnostic accuracy, and heatmap visualization assessed interpretability relevant to endoscopic decision-making. Results The ANN achieved diagnostic accuracy comparable to expert radiologists (93.3% vs. 93.8%). When assisting trainees, accuracy improved from 82.2% (AUC 0.82) to 91.1% (AUC 0.91), approaching expert performance (93.8%; AUC 0.94). Stone type and bile duct diameter > 10 mm significantly increased ANN detection rates. Heatmap visualization confirmed the plausibility of ANN predictions in both clearly identifiable lesions and indeterminate CT findings, improving interpretability for endoscopic decision-making. Conclusions The ANN achieved expert-level diagnostic accuracy for detecting CBD stones. By enhancing CT interpretation, it may optimize ERCP indications, reduce unnecessary invasive procedures, and improve training for less-experienced clinicians. Prospective validation and integration into multimodal endoscopic workflows are warranted. Health sciences/Diseases Health sciences/Gastroenterology Health sciences/Health care Health sciences/Medical research Common bile duct stones Computed tomography Artificial neural network Endoscopic retrograde cholangiopancreatography (ERCP) Endoscopic ultrasonography (EUS) Clinical decision support Figures Figure 1 Figure 2 Figure 3 Figure 4 Key summary What is already known about this subject? Computed tomography (CT) is widely used in the initial evaluation of suspected common bile duct (CBD) stones, but its diagnostic sensitivity is suboptimal, particularly for small or isoattenuating stones, necessitating additional endoscopic procedures. Artificial intelligence (AI) has shown promise in augmenting image interpretation across gastroenterology; however, validated CT-based AI algorithms for CBD stone detection remain limited. Current clinical guidelines recommend endoscopic ultrasound (EUS) or magnetic resonance cholangiopancreatography (MRCP) in cases of indeterminate CT findings and reserve endoscopic retrograde cholangiopancreatography (ERCP) primarily for therapeutic intervention. Demonstrating interpretability and reliability of AI algorithms (e.g., via heatmaps, predictive values, calibration) is essential for clinical translation and acceptance. What are the significant and/or new findings of this study? A CT-based artificial neural network (ANN) integrating UNETR-based segmentation with ResNet50 classification achieved diagnostic performance comparable to expert radiologists in a prospectively validated cohort. ANN assistance significantly improved the diagnostic accuracy of trainee radiologists, raising performance to near-expert levels. Heatmap-based visualizations substantiated the interpretability of ANN predictions in both unequivocal and indeterminate CT findings, supporting clinical plausibility. Complementary reliability analyses—including positive and negative predictive values and calibration—reinforce the potential integration of ANN-assisted CT into guideline-based EUS/ERCP decision pathways. Introduction Common bile duct (CBD) stones represent a clinically significant complication in patients with symptomatic cholelithiasis, with an estimated prevalence of 10–20% 1 . The presence of CBD stones can lead to potentially life-threatening sequelae, including acute suppurative cholangitis and biliary pancreatitis, both of which necessitate prompt diagnosis and timely endoscopic intervention 2 , 3 . Endoscopic retrograde cholangiopancreatography (ERCP) remains the standard of care for the definitive diagnosis and management of CBD stones; however, it is an invasive procedure associated with complications such as post-ERCP pancreatitis, bleeding, and perforation. Accordingly, optimizing preprocedural diagnostic accuracy is critical to ensure that only appropriate candidates undergo therapeutic ERCP, and that unnecessary interventions are avoided 3 . Although various imaging modalities are available, the sensitivity of conventional computed tomography (CT) for detecting CBD stones is suboptimal, particularly for radiolucent or isoattenuating stones ( Supplementary Fig. 1 ) 4 , 5 . While magnetic resonance cholangio- pancreatography (MRCP) and endoscopic ultrasound (EUS) can improve diagnostic yield, their routine use may be limited by availability, time constraints in emergency settings, and the requirement for experienced endoscopists and radiologists. Consequently, CT imaging remains the most commonly used initial modality in patients presenting with acute abdominal pain suspected to be of biliary origin, despite its diagnostic limitations for CBD stones. Recently, artificial intelligence (AI)—including artificial neural networks (ANNs)—has demonstrated considerable promise in augmenting image interpretation and improving diagnostic performance across various fields of medical imaging, including gastrointestinal diseases 6 – 8 . Prior studies have shown that computer-aided diagnosis systems can assist radiologists in detecting subtle lesions and differentiating between normal and pathological findings 9 . However, the application of ANN-based algorithms specifically for the automated detection of CBD stones on CT scans remains largely unexplored. Recent expert reviews have described the emerging role of AI for endoscopically diagnosing biliary tract diseases; yet, there is currently no validated CT-based computer-aided diagnosis system for CBD stone detection in clinical practice 10 . Moreover, the clinical factors that may enhance ANN performance in this context are not well understood. To address this unmet need, we developed an ANN algorithm integrating a UNETR-based segmentation network and a ResNet50-based classification network to automatically detect CBD stones on CT images 11 , 12 . We aimed to assess the diagnostic performance of the ANN relative to expert radiologists, and determine whether the ANN could improve diagnostic accuracy when used by trainee radiologists. Additionally, we sought to identify patient and imaging factors associated with improved diagnostic performance. We hypothesized that the application of an ANN to CT images would yield a diagnostic accuracy comparable to that of experienced radiologists, ultimately supporting more precise decision-making, reducing unnecessary invasive procedures, and enhancing training for less-experienced clinicians. Methods Data sources and search strategy This single-center study was conducted at Hallym University Kangnam Sacred Heart Hospital, Seoul, Korea. Patients who underwent abdominal CT due to clinically suspected CBD stones between March 2018 and June 2023 were included. The study cohort comprised a retrospectively collected derivation set and prospectively enrolled validation set. Patients with benign biliary strictures, biliary malignancies, or surgically altered anatomy were excluded. In total, 950 patients were initially screened. After applying the exclusion criteria, 830 and 225 patients were included in the derivation and validation cohorts, respectively. The gold standard for confirming the presence of CBD stones was ERCP. For clinical simulation, we assumed an ANN-guided triage pathway: patients with ANN-positive results would undergo immediate ERCP, whereas those with ANN-negative results would defer ERCP, with conditional secondary imaging (EUS or MRCP) in clinically high-risk cases. Unnecessary ERCPs were defined as procedures performed in patients without CBD stones. The study protocol was approved by the Institutional Review Board of Hallym University Kangnam Sacred Heart Hospital (IRB No. 2021-10-020). Given the retrospective design using existing CT and EMR data, the requirement for informed consent was waived by the by the same Institutional Review Board (IRB No. 2021-10-020). All methods were performed in accordance with the relevant guidelines and regulations (Declaration of Helsinki and institutional ethical standards). Image Acquisition, Preprocessing and Development of the Artificial Neural Network Model (Supplementary Fig. 2, also provided as the Graphical Abstract) Abdominal CT scans were obtained using standard institutional protocols for patients presenting with suspected biliary obstruction. The CT images were reviewed and annotated for the presence or absence of CBD stones by experienced radiologists. Each image was preprocessed to standardize the resolution, contrast, and region of interest before input into the deep learning model. The ANN was developed by combining a UNETR-based segmentation network and ResNet50-based classification network; both have demonstrated superior performance in medical image analysis tasks involving the gastrointestinal and hepatobiliary systems 7 . For the segmentation task, the UNETR architecture was used to delineate the biliary tree and potential stones within the CT images. The AdamW optimizer was applied with a learning rate of 0.0001 and DiceCE loss function. Training was performed for a maximum of 2,000 epochs, with early stopping to prevent overfitting. For the classification task, the ResNet50 backbone was employed to classify segmented regions as containing CBD stones or not. The Adam optimizer was used with a learning rate of 0.001 and a cross-entropy loss function. The classification network was trained for up to 200 epochs with early stopping. The training and validation datasets were split according to the derivation and validation cohorts. The deep learning framework and training processes were implemented using PyTorch (version ≥ 1.12)/TorchVision (version ≥ 0.13) and executed on dedicated GPU servers. Data augmentation techniques, including rotation and contrast adjustments, were applied to enhance the model robustness. Comparison with Radiologists The diagnostic performance of the developed ANN model was compared with that of experienced and trainee radiologists. In the prospective validation cohort, CT images were independently interpreted by expert radiologists, and trainee radiologists both with and without ANN assistance. Diagnostic performance metrics included sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and overall accuracy. Statistical Analysis Continuous variables are presented as means with standard deviations, and categorical variables as frequencies and percentages. Diagnostic performance was evaluated using standard measures of sensitivity, specificity, PPV, NPV, and accuracy, with corresponding 95% confidence intervals (CIs). Multivariate logistic regression analysis was performed to identify clinical and imaging factors associated with improved diagnostic performance of the ANN algorithm. Variables with a P-value < 0.05 in the univariate analysis were included in the multivariate model. Statistical analyses were conducted using SPSS version 29.0 (IBM Corp., Armonk, NY, USA) or equivalent software. Data availability The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request. All relevant data will be made available without undue reservation, in accordance with Scientific Reports data sharing policies. Results Patient Enrollment and Baseline Characteristics In total, 950 patients underwent abdominal CT due to clinical suspicion of CBD stones during the study period. After excluding patients with benign biliary strictures (n = 41), malignant biliary obstruction (n = 60), and surgically altered anatomy (n = 19), 830 and 225 patients were included in the retrospective derivation and prospective validation cohorts, respectively ( Fig. 1 ) . In the derivation cohort, 277 (33.4%) patients were confirmed to have CBD stones by ERCP, while 553 (66.6%) patients were not. In the validation cohort, CBD stones were present in 113 patients (50.2%) and absent in 112 (49.8%), based on ERCP findings. The baseline demographic and biochemical characteristics of patients with and without CBD stones are presented in Table 1 . Overall, there were no significant differences regarding age, sex distribution, or serum liver enzyme levels between the two groups. The mean bile duct diameter was significantly larger in patients with confirmed CBD stones than those without (P < 0.05; Table 1 ) . Table 1 Baseline characteristics of patients stratified by ANN diagnostic accuracy Derivation Cohort (n = 830) Validation Cohort (n = 225) Sex (male), n (%) 398 (48.0%) 110 (48.9%) Age (years), mean ± SD 55.8 ± 17.8 59.7 ± 15.5 Serum total bilirubin (mg/dL), mean ± SD 3.5 ± 1.3 2.8 ± 1.7 Serum AST (IU/L), mean ± SD 95.7 ± 25.5 127.3 ± 18.2 Serum ALT (IU/L), mean ± SD 120.9 ± 17.1 108.3 ± 21.0 Serum GGT (IU/L), mean ± SD 242.8 ± 335.5 223.5 ± 219.7 Presence of CBD stones, n (%) 310 (37.3%) 113 (50.2%) Diameter of bile duct (mm), mean ± SD 9.2 ± 3.8 9.8 ± 4.9 Size of CBD stones (mm), mean ± SD 8.3 ± 5.9 7.9 ± 6.1 Serum AST (IU/L), mean ± SD 95.7 ± 25.5 127.3 ± 18.2 Number of CBD stones Single, n (%) 182 (21.9%) 53 (23.5%) Multiple, n (%) 128 (15.4%) 60 (26.7%) Type of CBD stones, n (%) Black pigment stone 86 (10.4%) 27 (12.0%) Brown pigment stone 162 (19.5%) 63 (28.0%) CBD, common bile duct; CT, computed tomography; GGT, γ-glutamyl transferase; MRCP, magnetic resonance cholangiopancreatography; SD, standard deviation. Diagnostic Performance of the ANN Algorithm The ANN algorithm demonstrated high diagnostic accuracy for detecting CBD stones in the prospective validation cohort ( Table 2 ) , aligning with previous reports where deep learning models achieved expert-level performance in gastrointestinal imaging 13 . The overall diagnostic accuracy of the ANN model was 93.3%, which was comparable to that of expert radiologists (93.8%). The ANN achieved a sensitivity of 92.9%, specificity of 93.7%, PPV of 92.9%, and NPV of 93.8%. By comparison, expert radiologists attained a sensitivity of 93.8%, specificity of 93.8%, PPV of 93.8%, and NPV of 93.8%. The ANN model demonstrated excellent diagnostic performance, with an AUC of 0.93 (95% CI: 0.89–0.96), which was comparable to that of expert radiologists (AUC: 0.94; 95% CI: 0.90–0.97) in detecting CBD stones. The ANN-assisted trainee radiologists achieved an AUC of 0.91 (95% CI: 0.87–0.95), which was comparable to that of expert radiologists (AUC: 0.94; 95% CI: 0.90–0.97; Fig. 2 ). Notably, trainee radiologists assisted by the ANN algorithm demonstrated a substantial improvement in diagnostic performance; the overall accuracy was 91.1%, approaching the level of experienced radiologists (93.8%). This finding suggests that the ANN may serve as a valuable decision support tool, particularly for less-experienced clinicians interpreting CT images for suspected CBD stones. Table 2 ANN Diagnostic performance of ANN model compared with expert and trainee radiologists ANN % (95% CI) Expert Radiologists % (95% CI) Trainee Radiolosist + ANN % (95% CI) Sensitivity 92.98% (86.64–96.92) 93.81% (87.65–97.47) 91.15% (84.33–95.67) Specificity 93.69% (87.44–97.43) 93.75% (87.55–97.45) 91.07% (84.19–95.64) Accuracy 93.33% (89.24–96.22) 93.78% (89.78–96.56) 91.11% (86.61–94.49) AUC 0.93 (0.89–0.96) 0.94 (0.90–0.97) 0.91 (0.87–0.95) NPV 92.86% (86.93–96.21) 93.75% (87.96–96.85) 91.07% (84.91–94.87) PPV 93.81% (88.07–96.88) 93.81% (88.07–96.88) 91.15 (85.04–94.91) Abbreviations: ANN, artificial neural network; NPV, negative predictive value; PPV, positive predictive value; AUC, area under the receiver operating characteristic curve. Multivariate Analysis of Predictive Factors To identify factors associated with improved diagnostic performance of the ANN algorithm, multivariate logistic regression analysis was performed ( Table 3 ) . Among the clinical and imaging variables, stone type (adjusted odds ratio [OR]: 2.57, 95% CI: 1.25–6.09) and a bile duct diameter > 10 mm (adjusted OR: 2.31, 95% CI: 1.05–5.98) were independently associated with a higher likelihood of correct detection by the ANN (both P < 0.05). Table 3 Factors associated with accurate diagnosis (multivariate logistic regression) ANN Radiologists P value OR (95% CI) P value OR (95% CI) Number of stones 0.040 3.01 (1.31–7.67) Diameter of bile duct > 10mm 0.045 2.31 (1.05–5.98) Type of stones 0.025 2.57 (1.25–6.09) 0.056 2.29 (1.02–6.09) Abbreviations: ANN, artificial neural network; CI, confidence interval. Similarly, for radiologists, the presence of multiple stones (adjusted OR: 3.01, 95% CI: 1.31–7.67) and certain stone characteristics (adjusted OR: 2.29, 95% CI: 1.02–6.09)) significantly increased diagnostic accuracy. These findings underscore the potential of specific anatomical features to improve the diagnostic performance of both AI-based and human interpretations. Interpretability of the ANN Model: Heatmap Visualization and Summary of Diagnostic Performance Representative heatmap overlays generated by the ANN model are shown in Fig. 3 . These visualizations demonstrate that the algorithm appropriately localized regions of high prediction probability within the biliary tree in patients with confirmed CBD stones. In addition, Fig. 4 depicts clinical contexts in which CT findings were indeterminate but diagnostic clarification was achieved with EUS. This underscores the complementary role of EUS in addressing the inherent limitations of CT, as demonstrated in previous studies 3 , 4 . Such observations highlight the potential for ANN models to provide comparable support by enhancing interpretability in diagnostically challenging situations. Such interpretability supports the clinical plausibility of the model’s output and may enhance trust in AI-assisted diagnosis. Examples include true-positive, false-positive, and true-negative cases, illustrating the strengths and potential limitations of the ANN approach. Overall, the ANN algorithm’s diagnostic performance metrics were robust and consistent across subgroups, providing evidence that the model can reliably identify CBD stones in CT images with accuracy comparable to that of expert radiologists. Furthermore, the use of the ANN by trainee radiologists substantially reduced the gap in diagnostic accuracy between trainees and experts, suggesting that the ANN may serve as an effective educational aid, as well as a practical clinical decision support tool. Discussion In this study, we developed and validated an ANN algorithm using CT images for the automated detection of CBD stones, demonstrating a diagnostic performance comparable to that of experienced radiologists. These findings corroborate the recent AGA clinical practice guidelines that highlight the growing importance of AI-assisted diagnostic tools in gastroenterology 14 . Similarly, the ESGE position statement emphasized that AI systems in gastrointestinal endoscopy should achieve performance comparable to experienced endoscopists to ensure clinical applicability 15 . Although these guidelines primarily address endoscopic applications, they underscore the broader movement toward AI integration across gastrointestinal imaging. Notably, the ANN algorithm achieved a high sensitivity and specificity, significantly improving the diagnostic accuracy of trainee radiologists. These findings suggest that deep learning-based decision support tools can enhance the detection of CBD stones and reduce unnecessary invasive procedures. In emergency or after-hours clinical settings, in which MRCP or EUS may not be readily accessible, an ANN-based triage strategy may facilitate the reduction of unnecessary ERCP procedures while maintaining prioritization of patients with confirmed stones. The incorporation of a safety-net approach based on established clinical predictors (e.g., bile duct diameter > 10 mm or systemic signs of infection) may further minimize the potential risk of false-negative interpretations. Accurate preprocedural diagnosis of CBD stones remains a clinical challenge, especially in emergency settings where CT imaging is often the first modality performed 1 , 2 . Previous studies have reported suboptimal sensitivity of standard CT for detecting CBD stones, particularly for small or isoattenuating stones 3 , 4 . Moreover, although alternative imaging modalities, such as MRCP and EUS, have demonstrated higher diagnostic yields, they may not be readily available in urgent situations and are subject to operator dependency 5 , 7 . Our results are consistent with earlier work highlighting the diagnostic gap for CBD stones, and extend prior studies by showing that AI algorithms can help bridge this gap, particularly in emergency settings where CT is the only modality available 7 , 13 , 16 . Recent advances in AI, particularly deep learning and transformer-based segmentation networks, have shown promise in improving image interpretation for various abdominal diseases 12 . In this context, a recent comprehensive review summarized the growing applications of AI in biliopancreatic endoscopy, particularly in EUS and ERCP, highlighting its potential to enhance lesion detection, procedural decision-making, and overall clinical outcomes 17 . For instance, Hou et al. reported that AI models could enhance the detection of biliary tract abnormalities on imaging modalities other than CT; however, robust evidence focused on CBD stones is lacking 7 . Our study builds on this evidence by using a UNETR-based architecture for three-dimensional segmentation in combination with a ResNet50 classifier, aligning with the latest developments in medical image processing 11 . One important aspect of our findings is that the ANN algorithm significantly narrowed the performance gap between trainee and expert radiologists. This aligns with previous research suggesting that AI tools can function as educational aids by augmenting the diagnostic confidence and consistency of less-experienced clinicians 9 . Moreover, the heatmap visualizations provided by the model offer insight into the algorithm’s decision-making process, which may increase interpretability and trust among end-users 5 . Multivariate analysis also revealed that certain clinical factors, such as stone type and bile duct diameter, were associated with improved diagnostic performance. This suggests that integrating patient-specific anatomical features into AI models may further enhance detection rates. Similar findings have been observed in other biliary tract imaging research, underscoring the importance of incorporating clinical context into AI-driven diagnostic pathways. The potential clinical implications of our findings are noteworthy. Early and accurate detection of CBD stones can facilitate timely ERCP and prevent complications such as acute cholangitis or pancreatitis, while minimizing unnecessary invasive interventions. As deep learning algorithms continue to evolve, integrating such models into routine radiological workflows may help standardize diagnostic performance, reduce interobserver variability, and optimize resource utilization. Despite these strengths, several limitations of our study should be acknowledged; first, this was a single-center study with a relatively homogeneous patient population, which may limit generalizability of our results to other institutions and imaging protocols. Second, the ANN algorithm was developed and validated using CT images wherein stones were visibly present or highly suspected; therefore, its performance in cases of negative or equivocal CT findings remains to be determined. Third, although heatmap visualizations provide some interpretability, the decision-making mechanisms of deep learning models can still function as a “black box,” necessitating further research into explainable AI. Fourth, external validation using multicenter datasets with diverse populations are required to confirm the robustness and clinical applicability of our model. In addition, this study did not directly evaluate the reduction in ERCP procedures or clinical outcomes such as complication rates. Although improved specificity and NPV suggest a potential role in avoiding unnecessary interventions, this hypothesis requires confirmation in prospective multicenter studies. Interpretability remains a critical issue in the clinical acceptance of AI models, and although saliency maps and gradient-based visualization methods have been suggested to provide transparent explanations of model decisions, further validation is required 16 . Future research should focus on expanding training datasets, incorporating multimodal imaging, and developing multitask AI frameworks that integrate diagnosis, treatment planning, and prognosis prediction. The rapid development of vision transformer models and hybrid architectures holds promise for further performance improvements and interpretability in medical imaging AI applications. Conclusion In this single-center study, we developed and validated an ANN algorithm based on CT imaging for the automated detection of CBD stones. Our study demonstrates that an ANN algorithm based on CT imaging can achieve diagnostic performance comparable to that of experienced radiologists for detecting CBD stones, while also supporting the diagnostic confidence of trainee radiologists. These findings suggest that AI-based decision support may serve as a valuable adjunct in the noninvasive assessment of CBD stones and in clinical decision-making. Moreover, ANN-assisted interpretation of CT scans may help optimize the use of ERCP in clinical practice, particularly in settings where expertise is limited, or immediate endoscopic access is unavailable. Further multicenter studies are warranted to externally validate the model and assess its generalizability across diverse clinical environments. Future research should specifically assess whether ANN-assisted CT interpretation can translate into reduced ERCP utilization and improved patient outcomes in real-world clinical practice. Declarations Conflict of Interest The authors have no conflicts of interest to declare. Funding This research was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No.NRF-2021R1G1A1095798) and Hallym University Research Fund 2022 (HURF-2022-55). Author Contribution Hwehoon Chung (Writing– original draft, review & editing: Equal)ChanWoo KWAK (Methodology: Equal; Validation: Equal; Visualization: Equal)Sang Deok Shin (Writing– original draft, review & editing: Equal)Jae Guk Kim (Conceptualization: Equal; Data curation: Supporting)Hyun Young Choi (Conceptualization: Equal; Data curation: Supporting)Wonhee Kim (Conceptualization: Equal; Data curation: Supporting)Ji Young Woo (Data curation: Equal; Validation: Equal; Visualization: Equal)Young Jun Kim (Data curation: Equal; Validation: Equal; Visualization: Equal)Jae Keun Park (Conceptualization: Lead; Data curation: Supporting; Methodology: Equal; Validation:Equal; Visualization: Equal; Writing, review & editing – original)Jong Kyun Lee (Conceptualization: Equal; Data curation: Supporting; Methodology: Equal;Validation:Equal; Visualization: Equal; Writing, review & editing – original) Data Availability The datasets generated and analyzed during the current study are available from the corresponding author, Dr. Jae Keun Park ( [email protected] ), on reasonable request. All relevant data will be made available without undue reservation, in accordance with the data sharing policies of Scientific Reports. References Williams, E. et al. Updated guideline on the management of common bile duct stones (CBDS). Gut 66 , 765–782 (2017). Crockett, S. D., Wani, S., Gardner, T. B., Falck-Ytter, Y. & Barkun, A. N. Aga section. Gastroenterology 154 , 1096–1101 (2018). Kim, K. M. et al. Role of endoscopic ultrasonography in patients with intermediate probability of choledocholithiasis but a negative CT scan. J. Clin. Gastroenterol. 47 , 449–456 (2013). Park, J. K. et al. Long term outcome of EUS-based strategy for suspected choledocholithiasis but negative CT finding. Scand. J. Gastroenterol. 53 , 1381–1387 (2018). Chen, Y. T. & Wang, M. S. MR cholangiography 3D biliary tree automatic reconstruction system. Comput. Med. Imaging Graph. 28 , 13–20 (2004). Kröner, P. T. et al. Artificial intelligence in gastroenterology: A state-of-the-art review. World J. Gastroenterol. 27 , 6794 (2021). Hou, J. U. et al. Efficacy of an artificial neural network algorithm based on thick-slab magnetic resonance cholangiopancreatography images for the automated diagnosis of common bile duct stones. J. Gastroenterol. Hepatol. 36 , 3532–3540 (2021). Miyajima, C. et al. The Hippo Signaling Pathway Manipulates Cellular Senescence. Cells 14 , 13 (2024). Logeswaran, R. A computer-aided multidisease diagnostic system using MRCP. J. Digit. Imaging . 21 , 235–242 (2008). Correia, F. P. & Lourenço, L. C. Artificial intelligence in the endoscopic approach of biliary tract diseases: A current review. Artif. Intell. Gastrointest. Endoscopy . 3 , 9–15 (2022). He, K., Zhang, X., Ren, S. & Sun, J. in Proceedings of the IEEE conference on computer vision and pattern recognition. 770–778. Hatamizadeh, A., Tang, N. V., Yang, Y., Myronenko, D. & Landman, A. B, et al. in IEEE. 574–584. Byrne, M. F. et al. Real-time differentiation of adenomatous and hyperplastic diminutive colorectal polyps during analysis of unaltered videos of standard colonoscopy using a deep learning model. Gut 68 , 94–100 (2019). Sultan, S. et al. AGA living clinical practice guideline on computer-aided detection–Assisted colonoscopy. Gastroenterology 168 , 691–700 (2025). Messmann, H. et al. Expected value of artificial intelligence in gastrointestinal endoscopy: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement. Endoscopy 54 , 1211–1231. 10.1055/a-1950-5694 (2022). Khalifa, A., Obeid, J. S., Erno, J. & Rockey, D. C. The role of artificial intelligence in hepatology research and practice. Curr. Opin. Gastroenterol. 39 , 175–180 (2023). Agudo Castillo, B. et al. Advancements in biliopancreatic endoscopy - A comprehensive review of artificial intelligence in EUS and ERCP. Rev. Esp. Enferm Dig. 116 , 613–622. 10.17235/reed.2024.10456/2024 (2024). Additional Declarations No competing interests reported. Supplementary Files SupplFig.1..tif Supplementary Figure 1. Finding demonstrating the diagnostic limitation of CT in detecting common bile duct stones, with subsequent identification by EUS and therapeutic confirmation by ERCP. This underscores the complementary role of EUS in CT-negative CBD stone. CT, computed tomography; CBD, common bile duct; EUS, endoscopic ultrasonography; ERCP, endoscopic retrograde cholangiopancreatography SupplFig.2..tif Supplementary Figure 2. Schematic architecture of the ANN model integrating a UNETR-based segmentation network and ResNet50-based classification network for CBD stone detection ANN, artificial neural network; CBD, common bile duct VisualAbstract.tif Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7919357","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":543579918,"identity":"a30c409e-99df-4fe5-a34a-75dee7c3dff8","order_by":0,"name":"Hwehoon Chung","email":"","orcid":"","institution":"Hallym University Kangnam Sacred Heart Hospital","correspondingAuthor":false,"prefix":"","firstName":"Hwehoon","middleName":"","lastName":"Chung","suffix":""},{"id":543579919,"identity":"8a028a73-5505-4ed1-a806-ed34d6fbbacf","order_by":1,"name":"ChanWoo KWAK","email":"","orcid":"","institution":"Hallym University Kangnam Sacred Heart Hospital","correspondingAuthor":false,"prefix":"","firstName":"ChanWoo","middleName":"","lastName":"KWAK","suffix":""},{"id":543579920,"identity":"eba76f70-d09a-49f0-b58e-dff60e4d603f","order_by":2,"name":"Sang Deok Shin","email":"","orcid":"","institution":"Hallym University Kangnam Sacred Heart Hospital","correspondingAuthor":false,"prefix":"","firstName":"Sang","middleName":"Deok","lastName":"Shin","suffix":""},{"id":543579921,"identity":"1964f732-b058-4a90-b521-daffba775d7e","order_by":3,"name":"Jae Guk Kim","email":"","orcid":"","institution":"Hallym University Kangnam Sacred Heart Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jae","middleName":"Guk","lastName":"Kim","suffix":""},{"id":543579922,"identity":"c5804483-7563-45bc-93dd-2920e1b7c0c7","order_by":4,"name":"Hyun Young Choi","email":"","orcid":"","institution":"Hallym University Kangnam Sacred Heart Hospital","correspondingAuthor":false,"prefix":"","firstName":"Hyun","middleName":"Young","lastName":"Choi","suffix":""},{"id":543579923,"identity":"8966d852-66f7-4e31-a1d8-63eda157b997","order_by":5,"name":"Wonhee Kim","email":"","orcid":"","institution":"Hallym University Kangnam Sacred Heart Hospital","correspondingAuthor":false,"prefix":"","firstName":"Wonhee","middleName":"","lastName":"Kim","suffix":""},{"id":543579924,"identity":"01cc238f-b868-4ed2-82ca-ca257e089176","order_by":6,"name":"Ji Young Woo","email":"","orcid":"","institution":"Hallym University Kangnam Sacred Heart Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ji","middleName":"Young","lastName":"Woo","suffix":""},{"id":543579925,"identity":"4c18857e-92e6-4dd6-a9cb-17b825eb6a35","order_by":7,"name":"Young Jun Kim","email":"","orcid":"","institution":"Hallym University Kangnam Sacred Heart Hospital","correspondingAuthor":false,"prefix":"","firstName":"Young","middleName":"Jun","lastName":"Kim","suffix":""},{"id":543579926,"identity":"a9787a38-3e2f-4149-b2b7-6591be91ea60","order_by":8,"name":"Jong Kyun Lee","email":"","orcid":"","institution":"Samsung Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Jong","middleName":"Kyun","lastName":"Lee","suffix":""},{"id":543579927,"identity":"c40ab627-f282-4440-865b-92554e36c8d3","order_by":9,"name":"Jae Keun Park","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4ElEQVRIiWNgGAWjYBACAyjNzNjeAOJakKKl5wCIK0G8FgYGiQQwSViLOfvZYxI/d9SyM898fnXDjwIJBv727gS8Wix78tIke88cZ2acnVN2swfoMIkzZzfgd9iBHDMJ3rZjIC1pN3iAWgwkcgloOf/GTPIvSMvMM2k3/xCl5UaOmTRvWw0z4wz2Y7eJs+XGG2Nr2bYDwEDOYbstYyDBQ9gv53MMb75tq0s2bD/+7OabPzZy/O29+LUAAQswLg4nGzbwgOOIh5ByEGD+wMBQZyfPwP6AGNWjYBSMglEwAgEAvtlI3UAQlawAAAAASUVORK5CYII=","orcid":"","institution":"Hallym University Kangnam Sacred Heart Hospital","correspondingAuthor":true,"prefix":"","firstName":"Jae","middleName":"Keun","lastName":"Park","suffix":""}],"badges":[],"createdAt":"2025-10-22 08:27:57","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7919357/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7919357/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":96555713,"identity":"6db0e3f7-5686-424c-aa14-32e411530447","added_by":"auto","created_at":"2025-11-23 11:39:49","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4385850,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/c466f9cd7ee88d2d0ac85874.docx"},{"id":96605014,"identity":"62a5fb84-5144-4eab-9e41-3b38919714de","added_by":"auto","created_at":"2025-11-24 09:17:23","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":11792,"visible":true,"origin":"","legend":"","description":"","filename":"d4e4d875d02a48afb37e63ec5b64ddd4.json","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/b9d7ff8c1346ce46bccbe7e8.json"},{"id":96555718,"identity":"ede7339f-2998-4a38-b4e1-d575a82f21ce","added_by":"auto","created_at":"2025-11-23 11:39:49","extension":"tif","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":111342,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.1..tif","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/bfb8de9390c7a27c4955971a.tif"},{"id":96605039,"identity":"b5292276-97a0-4acd-9745-a0504a2a361f","added_by":"auto","created_at":"2025-11-24 09:17:45","extension":"tif","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":253084,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.2..tif","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/a97af7e31525ffd013576c3d.tif"},{"id":96604862,"identity":"77107c4f-af9c-4c05-8708-a2b3a77cf42d","added_by":"auto","created_at":"2025-11-24 09:15:23","extension":"tif","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1973784,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.3..tif","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/986960a9e53e04fac8257330.tif"},{"id":96555727,"identity":"57085dfb-b8c7-4f38-b23d-79e2161cffd5","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"tif","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1482880,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.4..tif","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/dfd6e971627c118176276e1c.tif"},{"id":96604771,"identity":"845c32d8-2d81-4b88-9628-1a8423520ef9","added_by":"auto","created_at":"2025-11-24 09:14:52","extension":"tif","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":139840,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig.1..tif","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/2a35d6d20e9fda425d253716.tif"},{"id":96555723,"identity":"d7d4d755-0df9-48e6-8424-bc2510c3f8cf","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"tif","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":320510,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig.2..tif","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/468c39a9c4a212bf9c1c3da0.tif"},{"id":96555736,"identity":"a5c685f2-a5a3-4dbe-bf34-c5b2d8c65f15","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"docx","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":22307,"visible":true,"origin":"","legend":"","description":"","filename":"Table.docx","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/8f1eccf62bdfb16cd594b772.docx"},{"id":96555721,"identity":"43afd848-3737-4e85-8f0f-4a629451eb8e","added_by":"auto","created_at":"2025-11-23 11:39:49","extension":"tif","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":419246,"visible":true,"origin":"","legend":"","description":"","filename":"VisualAbstract.tif","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/50f9df62da2c6fbda98347a8.tif"},{"id":96555738,"identity":"6480cdca-f468-4c32-be21-e4bc67fb0ef5","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"xml","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":83863,"visible":true,"origin":"","legend":"","description":"","filename":"d4e4d875d02a48afb37e63ec5b64ddd41enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/78ebed90042a108928fc89de.xml"},{"id":96555728,"identity":"42c6c289-f212-4f1f-b1db-75f50bddaeee","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"jpeg","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":111342,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/4662490e52ec3b7b351f0b7b.jpeg"},{"id":96555729,"identity":"70ad4d23-2aa0-4657-b7d8-8b1be13cec45","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"jpeg","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":253084,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/682dae54fe74821d32d1af4b.jpeg"},{"id":96604819,"identity":"066ae78e-87ee-459f-bddd-94ae17a07206","added_by":"auto","created_at":"2025-11-24 09:15:10","extension":"jpeg","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1973784,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/6887051a75d34f91300fc95f.jpeg"},{"id":96555722,"identity":"50403364-11b8-48de-811b-e7c2de08990e","added_by":"auto","created_at":"2025-11-23 11:39:49","extension":"jpeg","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1482880,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/6a4504a1f2974982ceba048c.jpeg"},{"id":96605032,"identity":"a74f4312-ee6c-4b6e-831c-1b3b7d961a71","added_by":"auto","created_at":"2025-11-24 09:17:41","extension":"jpeg","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":139840,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/f895908786bac152b156e92d.jpeg"},{"id":96604671,"identity":"919b2fc5-8871-415d-aa2e-4cee3c657321","added_by":"auto","created_at":"2025-11-24 09:14:32","extension":"jpeg","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":320510,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/fcaead3b3dd15f2866ef1778.jpeg"},{"id":96555719,"identity":"30bf1d97-a3bb-42cc-a437-2f935f97db27","added_by":"auto","created_at":"2025-11-23 11:39:49","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":22327,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/800cbec6d38761bacbbbded7.png"},{"id":96555737,"identity":"522504ab-f196-4fac-b6f4-6deb86ccad6c","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32307,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/0dbc116edfba032bd2c464db.png"},{"id":96604461,"identity":"cc436f1e-790a-49e2-a274-78d5afdb090b","added_by":"auto","created_at":"2025-11-24 09:14:00","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":148325,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/4f1a0d97e2d188e0a45d42e4.png"},{"id":96555744,"identity":"29a378e1-ed1e-476e-be68-4daf9ab20f36","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":95619,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/c5a9afd3ee64de597b1adb57.png"},{"id":96604993,"identity":"e2e2324d-8236-40b4-aaf7-33c47f2cc79b","added_by":"auto","created_at":"2025-11-24 09:17:13","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":33274,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/3e2fca2dfb0e72a40f365136.png"},{"id":96555730,"identity":"b2b892bd-851b-40fa-89cc-72e22951a45e","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":41233,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/e512fb0b1c7a4265a8875b75.png"},{"id":96555742,"identity":"f1f2bd53-109d-4acc-99c5-5e51b997e89c","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"xml","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":80042,"visible":true,"origin":"","legend":"","description":"","filename":"d4e4d875d02a48afb37e63ec5b64ddd41structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/b185dc2d755989b01f2ef62d.xml"},{"id":96555735,"identity":"d183a697-d444-4ab0-8686-54648a90cd6b","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"html","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":91004,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/800ddcd95b9a40529d7bd565.html"},{"id":96555712,"identity":"36c27140-75c9-40dd-bcdf-da8484bf4c94","added_by":"auto","created_at":"2025-11-23 11:39:49","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":177921,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eStudy design and cohort flow\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFlowchart demonstrating patient inclusion, exclusion criteria, derivation and validation cohorts, number of CT images analyzed, and ANN evaluation pathway\u003cbr\u003e\nANN, artificial neural network; CT, computed tomography\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/e3f1b948f366c24462fd6cea.png"},{"id":96604850,"identity":"c7ea74ba-47f6-4b69-8099-f50fe977aa36","added_by":"auto","created_at":"2025-11-24 09:15:20","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":269864,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDiagnostic performance of the ANN model versus expert and trainee radiologists\u003c/strong\u003e Sensitivity, specificity, PPV, NPV, and overall accuracy of the ANN algorithm compared to expert and ANN-assisted trainee radiologists\u003c/p\u003e\n\u003cp\u003eANN, artificial neural network; PPV, positive predictive value; NPV, negative predictive value\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/f325e880c1c63ad5579d5b63.png"},{"id":96555731,"identity":"df635334-df90-4f0b-abb5-071d9cc256c7","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":739582,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eHeatmap visualizations of ANN model predictions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCase 1: (A) Contrast-enhanced CT scan showing a hyperattenuating CBD stone (arrow). (B) Heatmap generated by the ANN model correctly localizing the stone (C) ERCP confirming the presence of the CBD stone\u003c/p\u003e\n\u003cp\u003eCase 2: (A) Contrast-enhanced CT scan of a radiolucent CBD stone with subtle attenuation difference (arrow) (B) Heatmap showing weak activation at the stone site, corresponding to a false-negative ANN prediction (C) ERCP demonstrating the CBD stone missed by the ANN\u003c/p\u003e\n\u003cp\u003eANN, artificial neural network; CT, computed tomography; ERCP, endoscopic retrograde cholangiopancreatography; CBD, common bile duct\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/76e36b55455f7162440c29f9.png"},{"id":96555734,"identity":"1b3a60b1-fb8e-46bf-8385-2e6159ac9a13","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":662442,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentative multimodal assessment of CBD stone\u003cbr\u003e\n(A) CT image with indeterminate findings, in which the suspected stone was not clearly visualized. (B) EUS demonstrating a CBD stone despite a negative CT finding (arrow). (C) Heatmap output from the proposed ANN model applied to the same patient, highlighting the region of interest corresponding to the stone. (D) ERCP confirming and successfully extracting the CBD stone.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/f4c1c54cc0750ad13703d7b2.png"},{"id":97684682,"identity":"c1cfa86e-8253-493b-a116-3dd20dd68275","added_by":"auto","created_at":"2025-12-08 10:08:00","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3143385,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/cec6fd16-c233-41f5-9a8b-6bd146adc9a9.pdf"},{"id":96604888,"identity":"2ca46f2c-aacd-48a3-b88c-384a4ab9284d","added_by":"auto","created_at":"2025-11-24 09:15:39","extension":"tif","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":139840,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Figure 1.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFinding demonstrating the diagnostic limitation of CT in detecting common bile duct stones, with subsequent identification by EUS and therapeutic confirmation by ERCP. This underscores the complementary role of EUS in CT-negative CBD stone.\u003c/p\u003e\n\u003cp\u003eCT, computed tomography; CBD, common bile duct; EUS, endoscopic ultrasonography; ERCP, endoscopic retrograde cholangiopancreatography\u003c/p\u003e","description":"","filename":"SupplFig.1..tif","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/e5588724385956974e632e6d.tif"},{"id":96555715,"identity":"f3104da1-55e4-41f1-aff5-f198b1671776","added_by":"auto","created_at":"2025-11-23 11:39:49","extension":"tif","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":320510,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Figure 2.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSchematic architecture of the ANN model integrating a UNETR-based segmentation network and ResNet50-based classification network for CBD stone detection\u003c/p\u003e\n\u003cp\u003eANN, artificial neural network; CBD, common bile duct\u003c/p\u003e","description":"","filename":"SupplFig.2..tif","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/85482b6e6b0560b4f0e0985e.tif"},{"id":96555741,"identity":"2b1c38ff-f35e-42a4-9042-fc4172362bff","added_by":"auto","created_at":"2025-11-23 11:39:50","extension":"tif","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":419246,"visible":true,"origin":"","legend":"","description":"","filename":"VisualAbstract.tif","url":"https://assets-eu.researchsquare.com/files/rs-7919357/v1/c9a1efa9dc13bba6f6ffda77.tif"}],"financialInterests":"No competing interests reported.","formattedTitle":"Integrating Artificial Intelligence into Endoscopic Decision-Making: ANN-Assisted CT for Common Bile Duct Stones","fulltext":[{"header":"Key summary","content":"\u003cp\u003e\u003cstrong\u003eWhat is already known about this subject?\u003c/strong\u003e\u003c/p\u003e\n\u003cul start=\"50\"\u003e\n \u003cli\u003eComputed tomography (CT) is widely used in the initial evaluation of suspected common bile duct (CBD) stones, but its diagnostic sensitivity is suboptimal, particularly for small or isoattenuating stones, necessitating additional endoscopic procedures.\u003c/li\u003e\n \u003cli\u003eArtificial intelligence (AI) has shown promise in augmenting image interpretation across gastroenterology; however, validated CT-based AI algorithms for CBD stone detection remain limited.\u003c/li\u003e\n \u003cli\u003eCurrent clinical guidelines recommend endoscopic ultrasound (EUS) or magnetic resonance cholangiopancreatography (MRCP) in cases of indeterminate CT findings and reserve endoscopic retrograde cholangiopancreatography (ERCP) primarily for therapeutic intervention.\u003c/li\u003e\n \u003cli\u003eDemonstrating interpretability and reliability of AI algorithms (e.g., via heatmaps, predictive values, calibration) is essential for clinical translation and acceptance.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003eWhat are the significant and/or new findings of this study?\u003c/strong\u003e\u003c/p\u003e\n\u003cul start=\"50\"\u003e\n \u003cli\u003eA CT-based artificial neural network (ANN) integrating UNETR-based segmentation with ResNet50 classification achieved diagnostic performance comparable to expert radiologists in a prospectively validated cohort.\u003c/li\u003e\n \u003cli\u003eANN assistance significantly improved the diagnostic accuracy of trainee radiologists, raising performance to near-expert levels.\u003c/li\u003e\n \u003cli\u003eHeatmap-based visualizations substantiated the interpretability of ANN predictions in both unequivocal and indeterminate CT findings, supporting clinical plausibility.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cul\u003e\n \u003cli\u003eComplementary reliability analyses\u0026mdash;including positive and negative predictive values and calibration\u0026mdash;reinforce the potential integration of ANN-assisted CT into guideline-based EUS/ERCP decision pathways.\u003c/li\u003e\n\u003c/ul\u003e"},{"header":"Introduction","content":"\u003cp\u003eCommon bile duct (CBD) stones represent a clinically significant complication in patients with symptomatic cholelithiasis, with an estimated prevalence of 10\u0026ndash;20% \u003csup\u003e1\u003c/sup\u003e. The presence of CBD stones can lead to potentially life-threatening sequelae, including acute suppurative cholangitis and biliary pancreatitis, both of which necessitate prompt diagnosis and timely endoscopic intervention \u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e,\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. Endoscopic retrograde cholangiopancreatography (ERCP) remains the standard of care for the definitive diagnosis and management of CBD stones; however, it is an invasive procedure associated with complications such as post-ERCP pancreatitis, bleeding, and perforation. Accordingly, optimizing preprocedural diagnostic accuracy is critical to ensure that only appropriate candidates undergo therapeutic ERCP, and that unnecessary interventions are avoided \u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eAlthough various imaging modalities are available, the sensitivity of conventional computed tomography (CT) for detecting CBD stones is suboptimal, particularly for radiolucent or isoattenuating stones (\u003cb\u003eSupplementary Fig.\u0026nbsp;1\u003c/b\u003e) \u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e,\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. While magnetic resonance cholangio- pancreatography (MRCP) and endoscopic ultrasound (EUS) can improve diagnostic yield, their routine use may be limited by availability, time constraints in emergency settings, and the requirement for experienced endoscopists and radiologists. Consequently, CT imaging remains the most commonly used initial modality in patients presenting with acute abdominal pain suspected to be of biliary origin, despite its diagnostic limitations for CBD stones.\u003c/p\u003e\u003cp\u003eRecently, artificial intelligence (AI)\u0026mdash;including artificial neural networks (ANNs)\u0026mdash;has demonstrated considerable promise in augmenting image interpretation and improving diagnostic performance across various fields of medical imaging, including gastrointestinal diseases \u003csup\u003e\u003cspan additionalcitationids=\"CR7\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. Prior studies have shown that computer-aided diagnosis systems can assist radiologists in detecting subtle lesions and differentiating between normal and pathological findings \u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. However, the application of ANN-based algorithms specifically for the automated detection of CBD stones on CT scans remains largely unexplored. Recent expert reviews have described the emerging role of AI for endoscopically diagnosing biliary tract diseases; yet, there is currently no validated CT-based computer-aided diagnosis system for CBD stone detection in clinical practice \u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. Moreover, the clinical factors that may enhance ANN performance in this context are not well understood.\u003c/p\u003e\u003cp\u003eTo address this unmet need, we developed an ANN algorithm integrating a UNETR-based segmentation network and a ResNet50-based classification network to automatically detect CBD stones on CT images \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. We aimed to assess the diagnostic performance of the ANN relative to expert radiologists, and determine whether the ANN could improve diagnostic accuracy when used by trainee radiologists. Additionally, we sought to identify patient and imaging factors associated with improved diagnostic performance. We hypothesized that the application of an ANN to CT images would yield a diagnostic accuracy comparable to that of experienced radiologists, ultimately supporting more precise decision-making, reducing unnecessary invasive procedures, and enhancing training for less-experienced clinicians.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eData sources and search strategy\u003c/h2\u003e\u003cp\u003eThis single-center study was conducted at Hallym University Kangnam Sacred Heart Hospital, Seoul, Korea. Patients who underwent abdominal CT due to clinically suspected CBD stones between March 2018 and June 2023 were included. The study cohort comprised a retrospectively collected derivation set and prospectively enrolled validation set. Patients with benign biliary strictures, biliary malignancies, or surgically altered anatomy were excluded.\u003c/p\u003e\u003cp\u003eIn total, 950 patients were initially screened. After applying the exclusion criteria, 830 and 225 patients were included in the derivation and validation cohorts, respectively. The gold standard for confirming the presence of CBD stones was ERCP. For clinical simulation, we assumed an ANN-guided triage pathway: patients with ANN-positive results would undergo immediate ERCP, whereas those with ANN-negative results would defer ERCP, with conditional secondary imaging (EUS or MRCP) in clinically high-risk cases. Unnecessary ERCPs were defined as procedures performed in patients without CBD stones. The study protocol was approved by the Institutional Review Board of Hallym University Kangnam Sacred Heart Hospital (IRB No. 2021-10-020). Given the retrospective design using existing CT and EMR data, the requirement for informed consent was waived by the by the same Institutional Review Board (IRB No. 2021-10-020). All methods were performed in accordance with the relevant guidelines and regulations (Declaration of Helsinki and institutional ethical standards).\u003c/p\u003e\u003cp\u003e\u003cb\u003eImage Acquisition, Preprocessing and Development of the Artificial Neural Network Model (Supplementary Fig.\u0026nbsp;2, also provided as the Graphical Abstract)\u003c/b\u003e\u003c/p\u003e\u003cp\u003eAbdominal CT scans were obtained using standard institutional protocols for patients presenting with suspected biliary obstruction. The CT images were reviewed and annotated for the presence or absence of CBD stones by experienced radiologists. Each image was preprocessed to standardize the resolution, contrast, and region of interest before input into the deep learning model. The ANN was developed by combining a UNETR-based segmentation network and ResNet50-based classification network; both have demonstrated superior performance in medical image analysis tasks involving the gastrointestinal and hepatobiliary systems \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. For the segmentation task, the UNETR architecture was used to delineate the biliary tree and potential stones within the CT images. The AdamW optimizer was applied with a learning rate of 0.0001 and DiceCE loss function. Training was performed for a maximum of 2,000 epochs, with early stopping to prevent overfitting.\u003c/p\u003e\u003cp\u003eFor the classification task, the ResNet50 backbone was employed to classify segmented regions as containing CBD stones or not. The Adam optimizer was used with a learning rate of 0.001 and a cross-entropy loss function. The classification network was trained for up to 200 epochs with early stopping. The training and validation datasets were split according to the derivation and validation cohorts. The deep learning framework and training processes were implemented using PyTorch (version\u0026thinsp;\u0026ge;\u0026thinsp;1.12)/TorchVision (version\u0026thinsp;\u0026ge;\u0026thinsp;0.13) and executed on dedicated GPU servers. Data augmentation techniques, including rotation and contrast adjustments, were applied to enhance the model robustness.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eComparison with Radiologists\u003c/h3\u003e\n\u003cp\u003eThe diagnostic performance of the developed ANN model was compared with that of experienced and trainee radiologists. In the prospective validation cohort, CT images were independently interpreted by expert radiologists, and trainee radiologists both with and without ANN assistance. Diagnostic performance metrics included sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and overall accuracy.\u003c/p\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003eStatistical Analysis\u003c/h2\u003e\u003cp\u003eContinuous variables are presented as means with standard deviations, and categorical variables as frequencies and percentages. Diagnostic performance was evaluated using standard measures of sensitivity, specificity, PPV, NPV, and accuracy, with corresponding 95% confidence intervals (CIs).\u003c/p\u003e\u003cp\u003eMultivariate logistic regression analysis was performed to identify clinical and imaging factors associated with improved diagnostic performance of the ANN algorithm. Variables with a P-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 in the univariate analysis were included in the multivariate model. Statistical analyses were conducted using SPSS version 29.0 (IBM Corp., Armonk, NY, USA) or equivalent software.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eData availability\u003c/h3\u003e\n\u003cp\u003eThe datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request. All relevant data will be made available without undue reservation, in accordance with \u003cem\u003eScientific Reports\u003c/em\u003e data sharing policies.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003ePatient Enrollment and Baseline Characteristics\u003c/h2\u003e\u003cp\u003eIn total, 950 patients underwent abdominal CT due to clinical suspicion of CBD stones during the study period. After excluding patients with benign biliary strictures (n\u0026thinsp;=\u0026thinsp;41), malignant biliary obstruction (n\u0026thinsp;=\u0026thinsp;60), and surgically altered anatomy (n\u0026thinsp;=\u0026thinsp;19), 830 and 225 patients were included in the retrospective derivation and prospective validation cohorts, respectively \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e. In the derivation cohort, 277 (33.4%) patients were confirmed to have CBD stones by ERCP, while 553 (66.6%) patients were not. In the validation cohort, CBD stones were present in 113 patients (50.2%) and absent in 112 (49.8%), based on ERCP findings.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe baseline demographic and biochemical characteristics of patients with and without CBD stones are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Overall, there were no significant differences regarding age, sex distribution, or serum liver enzyme levels between the two groups. The mean bile duct diameter was significantly larger in patients with confirmed CBD stones than those without (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05; Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eBaseline characteristics of patients stratified by ANN diagnostic accuracy\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDerivation Cohort (n\u0026thinsp;=\u0026thinsp;830)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eValidation Cohort (n\u0026thinsp;=\u0026thinsp;225)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSex (male), n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e398 (48.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e110 (48.9%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge (years), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e55.8\u0026thinsp;\u0026plusmn;\u0026thinsp;17.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e59.7\u0026thinsp;\u0026plusmn;\u0026thinsp;15.5\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSerum total bilirubin (mg/dL), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e3.5\u0026thinsp;\u0026plusmn;\u0026thinsp;1.3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2.8\u0026thinsp;\u0026plusmn;\u0026thinsp;1.7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSerum AST (IU/L), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e95.7\u0026thinsp;\u0026plusmn;\u0026thinsp;25.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e127.3\u0026thinsp;\u0026plusmn;\u0026thinsp;18.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSerum ALT (IU/L), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e120.9\u0026thinsp;\u0026plusmn;\u0026thinsp;17.1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e108.3\u0026thinsp;\u0026plusmn;\u0026thinsp;21.0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSerum GGT (IU/L), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e242.8\u0026thinsp;\u0026plusmn;\u0026thinsp;335.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e223.5\u0026thinsp;\u0026plusmn;\u0026thinsp;219.7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePresence of CBD stones, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e310 (37.3%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e113 (50.2%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDiameter of bile duct (mm), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e9.2\u0026thinsp;\u0026plusmn;\u0026thinsp;3.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e9.8\u0026thinsp;\u0026plusmn;\u0026thinsp;4.9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSize of CBD stones (mm), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e8.3\u0026thinsp;\u0026plusmn;\u0026thinsp;5.9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e7.9\u0026thinsp;\u0026plusmn;\u0026thinsp;6.1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSerum AST (IU/L), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e95.7\u0026thinsp;\u0026plusmn;\u0026thinsp;25.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e127.3\u0026thinsp;\u0026plusmn;\u0026thinsp;18.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNumber of CBD stones\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSingle, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e182 (21.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e53 (23.5%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMultiple, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e128 (15.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e60 (26.7%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eType of CBD stones, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBlack pigment stone\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e86 (10.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e27 (12.0%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBrown pigment stone\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e162 (19.5%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e63 (28.0%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"3\"\u003eCBD, common bile duct; CT, computed tomography; GGT, γ-glutamyl transferase; MRCP, magnetic resonance cholangiopancreatography; SD, standard deviation.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eDiagnostic Performance of the ANN Algorithm\u003c/h3\u003e\n\u003cp\u003eThe ANN algorithm demonstrated high diagnostic accuracy for detecting CBD stones in the prospective validation cohort \u003cb\u003e(\u003c/b\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e, aligning with previous reports where deep learning models achieved expert-level performance in gastrointestinal imaging \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. The overall diagnostic accuracy of the ANN model was 93.3%, which was comparable to that of expert radiologists (93.8%). The ANN achieved a sensitivity of 92.9%, specificity of 93.7%, PPV of 92.9%, and NPV of 93.8%. By comparison, expert radiologists attained a sensitivity of 93.8%, specificity of 93.8%, PPV of 93.8%, and NPV of 93.8%. The ANN model demonstrated excellent diagnostic performance, with an AUC of 0.93 (95% CI: 0.89\u0026ndash;0.96), which was comparable to that of expert radiologists (AUC: 0.94; 95% CI: 0.90\u0026ndash;0.97) in detecting CBD stones. The ANN-assisted trainee radiologists achieved an AUC of 0.91 (95% CI: 0.87\u0026ndash;0.95), which was comparable to that of expert radiologists (AUC: 0.94; 95% CI: 0.90\u0026ndash;0.97; Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Notably, trainee radiologists assisted by the ANN algorithm demonstrated a substantial improvement in diagnostic performance; the overall accuracy was 91.1%, approaching the level of experienced radiologists (93.8%). This finding suggests that the ANN may serve as a valuable decision support tool, particularly for less-experienced clinicians interpreting CT images for suspected CBD stones.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eANN Diagnostic performance of ANN model compared with expert and trainee radiologists\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eANN\u003c/p\u003e\u003cp\u003e% (95% CI)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExpert Radiologists\u003c/p\u003e\u003cp\u003e% (95% CI)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTrainee Radiolosist\u0026thinsp;+\u0026thinsp;ANN\u003c/p\u003e\u003cp\u003e% (95% CI)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSensitivity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e92.98% (86.64\u0026ndash;96.92)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e93.81% (87.65\u0026ndash;97.47)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e91.15% (84.33\u0026ndash;95.67)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSpecificity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e93.69% (87.44\u0026ndash;97.43)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e93.75% (87.55\u0026ndash;97.45)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e91.07% (84.19\u0026ndash;95.64)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e93.33% (89.24\u0026ndash;96.22)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e93.78% (89.78\u0026ndash;96.56)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e91.11% (86.61\u0026ndash;94.49)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAUC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.93 (0.89\u0026ndash;0.96)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.94 (0.90\u0026ndash;0.97)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.91 (0.87\u0026ndash;0.95)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNPV\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e92.86% (86.93\u0026ndash;96.21)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e93.75% (87.96\u0026ndash;96.85)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e91.07% (84.91\u0026ndash;94.87)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePPV\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e93.81% (88.07\u0026ndash;96.88)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e93.81% (88.07\u0026ndash;96.88)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e91.15 (85.04\u0026ndash;94.91)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"4\"\u003eAbbreviations: ANN, artificial neural network; NPV, negative predictive value; PPV, positive predictive value; AUC, area under the receiver operating characteristic curve.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eMultivariate Analysis of Predictive Factors\u003c/h3\u003e\n\u003cp\u003eTo identify factors associated with improved diagnostic performance of the ANN algorithm, multivariate logistic regression analysis was performed \u003cb\u003e(\u003c/b\u003eTable\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e. Among the clinical and imaging variables, stone type (adjusted odds ratio [OR]: 2.57, 95% CI: 1.25\u0026ndash;6.09) and a bile duct diameter\u0026thinsp;\u0026gt;\u0026thinsp;10 mm (adjusted OR: 2.31, 95% CI: 1.05\u0026ndash;5.98) were independently associated with a higher likelihood of correct detection by the ANN (both P\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eFactors associated with accurate diagnosis (multivariate logistic regression)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u003cp\u003eANN\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eRadiologists\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eP value\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eOR\u003c/p\u003e\u003cp\u003e(95% CI)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eP value\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eOR\u003c/p\u003e\u003cp\u003e(95% CI)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eNumber of stones\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.040\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e3.01\u003c/b\u003e\u003c/p\u003e\u003cp\u003e(1.31\u0026ndash;7.67)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eDiameter of bile\u003c/b\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003educt\u0026thinsp;\u0026gt;\u0026thinsp;10mm\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.045\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e2.31\u003c/b\u003e\u003c/p\u003e\u003cp\u003e(1.05\u0026ndash;5.98)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eType of stones\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.025\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e2.57\u003c/b\u003e\u003c/p\u003e\u003cp\u003e(1.25\u0026ndash;6.09)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.056\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e2.29\u003c/b\u003e\u003c/p\u003e\u003cp\u003e(1.02\u0026ndash;6.09)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"5\"\u003eAbbreviations: ANN, artificial neural network; CI, confidence interval.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eSimilarly, for radiologists, the presence of multiple stones (adjusted OR: 3.01, 95% CI: 1.31\u0026ndash;7.67) and certain stone characteristics (adjusted OR: 2.29, 95% CI: 1.02\u0026ndash;6.09)) significantly increased diagnostic accuracy. These findings underscore the potential of specific anatomical features to improve the diagnostic performance of both AI-based and human interpretations.\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eInterpretability of the ANN Model: Heatmap Visualization and Summary of Diagnostic Performance\u003c/h2\u003e\u003cp\u003eRepresentative heatmap overlays generated by the ANN model are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. These visualizations demonstrate that the algorithm appropriately localized regions of high prediction probability within the biliary tree in patients with confirmed CBD stones. In addition, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e depicts clinical contexts in which CT findings were indeterminate but diagnostic clarification was achieved with EUS. This underscores the complementary role of EUS in addressing the inherent limitations of CT, as demonstrated in previous studies \u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e,\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Such observations highlight the potential for ANN models to provide comparable support by enhancing interpretability in diagnostically challenging situations. Such interpretability supports the clinical plausibility of the model\u0026rsquo;s output and may enhance trust in AI-assisted diagnosis. Examples include true-positive, false-positive, and true-negative cases, illustrating the strengths and potential limitations of the ANN approach.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eOverall, the ANN algorithm\u0026rsquo;s diagnostic performance metrics were robust and consistent across subgroups, providing evidence that the model can reliably identify CBD stones in CT images with accuracy comparable to that of expert radiologists. Furthermore, the use of the ANN by trainee radiologists substantially reduced the gap in diagnostic accuracy between trainees and experts, suggesting that the ANN may serve as an effective educational aid, as well as a practical clinical decision support tool.\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, we developed and validated an ANN algorithm using CT images for the automated detection of CBD stones, demonstrating a diagnostic performance comparable to that of experienced radiologists. These findings corroborate the recent AGA clinical practice guidelines that highlight the growing importance of AI-assisted diagnostic tools in gastroenterology \u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. Similarly, the ESGE position statement emphasized that AI systems in gastrointestinal endoscopy should achieve performance comparable to experienced endoscopists to ensure clinical applicability \u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. Although these guidelines primarily address endoscopic applications, they underscore the broader movement toward AI integration across gastrointestinal imaging. Notably, the ANN algorithm achieved a high sensitivity and specificity, significantly improving the diagnostic accuracy of trainee radiologists. These findings suggest that deep learning-based decision support tools can enhance the detection of CBD stones and reduce unnecessary invasive procedures. In emergency or after-hours clinical settings, in which MRCP or EUS may not be readily accessible, an ANN-based triage strategy may facilitate the reduction of unnecessary ERCP procedures while maintaining prioritization of patients with confirmed stones. The incorporation of a safety-net approach based on established clinical predictors (e.g., bile duct diameter\u0026thinsp;\u0026gt;\u0026thinsp;10 mm or systemic signs of infection) may further minimize the potential risk of false-negative interpretations. Accurate preprocedural diagnosis of CBD stones remains a clinical challenge, especially in emergency settings where CT imaging is often the first modality performed \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Previous studies have reported suboptimal sensitivity of standard CT for detecting CBD stones, particularly for small or isoattenuating stones \u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e,\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Moreover, although alternative imaging modalities, such as MRCP and EUS, have demonstrated higher diagnostic yields, they may not be readily available in urgent situations and are subject to operator dependency \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e,\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Our results are consistent with earlier work highlighting the diagnostic gap for CBD stones, and extend prior studies by showing that AI algorithms can help bridge this gap, particularly in emergency settings where CT is the only modality available \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eRecent advances in AI, particularly deep learning and transformer-based segmentation networks, have shown promise in improving image interpretation for various abdominal diseases \u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. In this context, a recent comprehensive review summarized the growing applications of AI in biliopancreatic endoscopy, particularly in EUS and ERCP, highlighting its potential to enhance lesion detection, procedural decision-making, and overall clinical outcomes \u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. For instance, Hou et al. reported that AI models could enhance the detection of biliary tract abnormalities on imaging modalities other than CT; however, robust evidence focused on CBD stones is lacking \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Our study builds on this evidence by using a UNETR-based architecture for three-dimensional segmentation in combination with a ResNet50 classifier, aligning with the latest developments in medical image processing \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eOne important aspect of our findings is that the ANN algorithm significantly narrowed the performance gap between trainee and expert radiologists. This aligns with previous research suggesting that AI tools can function as educational aids by augmenting the diagnostic confidence and consistency of less-experienced clinicians \u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. Moreover, the heatmap visualizations provided by the model offer insight into the algorithm\u0026rsquo;s decision-making process, which may increase interpretability and trust among end-users \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. Multivariate analysis also revealed that certain clinical factors, such as stone type and bile duct diameter, were associated with improved diagnostic performance. This suggests that integrating patient-specific anatomical features into AI models may further enhance detection rates. Similar findings have been observed in other biliary tract imaging research, underscoring the importance of incorporating clinical context into AI-driven diagnostic pathways.\u003c/p\u003e\u003cp\u003eThe potential clinical implications of our findings are noteworthy. Early and accurate detection of CBD stones can facilitate timely ERCP and prevent complications such as acute cholangitis or pancreatitis, while minimizing unnecessary invasive interventions. As deep learning algorithms continue to evolve, integrating such models into routine radiological workflows may help standardize diagnostic performance, reduce interobserver variability, and optimize resource utilization.\u003c/p\u003e\u003cp\u003e Despite these strengths, several limitations of our study should be acknowledged; first, this was a single-center study with a relatively homogeneous patient population, which may limit generalizability of our results to other institutions and imaging protocols. Second, the ANN algorithm was developed and validated using CT images wherein stones were visibly present or highly suspected; therefore, its performance in cases of negative or equivocal CT findings remains to be determined. Third, although heatmap visualizations provide some interpretability, the decision-making mechanisms of deep learning models can still function as a \u0026ldquo;black box,\u0026rdquo; necessitating further research into explainable AI. Fourth, external validation using multicenter datasets with diverse populations are required to confirm the robustness and clinical applicability of our model. In addition, this study did not directly evaluate the reduction in ERCP procedures or clinical outcomes such as complication rates. Although improved specificity and NPV suggest a potential role in avoiding unnecessary interventions, this hypothesis requires confirmation in prospective multicenter studies.\u003c/p\u003e\u003cp\u003eInterpretability remains a critical issue in the clinical acceptance of AI models, and although saliency maps and gradient-based visualization methods have been suggested to provide transparent explanations of model decisions, further validation is required \u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. Future research should focus on expanding training datasets, incorporating multimodal imaging, and developing multitask AI frameworks that integrate diagnosis, treatment planning, and prognosis prediction. The rapid development of vision transformer models and hybrid architectures holds promise for further performance improvements and interpretability in medical imaging AI applications.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn this single-center study, we developed and validated an ANN algorithm based on CT imaging for the automated detection of CBD stones. Our study demonstrates that an ANN algorithm based on CT imaging can achieve diagnostic performance comparable to that of experienced radiologists for detecting CBD stones, while also supporting the diagnostic confidence of trainee radiologists. These findings suggest that AI-based decision support may serve as a valuable adjunct in the noninvasive assessment of CBD stones and in clinical decision-making. Moreover, ANN-assisted interpretation of CT scans may help optimize the use of ERCP in clinical practice, particularly in settings where expertise is limited, or immediate endoscopic access is unavailable. Further multicenter studies are warranted to externally validate the model and assess its generalizability across diverse clinical environments. Future research should specifically assess whether ANN-assisted CT interpretation can translate into reduced ERCP utilization and improved patient outcomes in real-world clinical practice.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eConflict of Interest\u003c/h2\u003e\n\u003cp\u003eThe authors have no conflicts of interest to declare.\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eThis research was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No.NRF-2021R1G1A1095798) and Hallym University Research Fund 2022 (HURF-2022-55).\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eHwehoon Chung (Writing\u0026ndash; original draft, review \u0026amp;amp; editing: Equal)ChanWoo KWAK (Methodology: Equal; Validation: Equal; Visualization: Equal)Sang Deok Shin (Writing\u0026ndash; original draft, review \u0026amp;amp; editing: Equal)Jae Guk Kim (Conceptualization: Equal; Data curation: Supporting)Hyun Young Choi (Conceptualization: Equal; Data curation: Supporting)Wonhee Kim (Conceptualization: Equal; Data curation: Supporting)Ji Young Woo (Data curation: Equal; Validation: Equal; Visualization: Equal)Young Jun Kim (Data curation: Equal; Validation: Equal; Visualization: Equal)Jae Keun Park (Conceptualization: Lead; Data curation: Supporting; Methodology: Equal; Validation:Equal; Visualization: Equal; Writing, review \u0026amp;amp; editing \u0026ndash; original)Jong Kyun Lee (Conceptualization: Equal; Data curation: Supporting; Methodology: Equal;Validation:Equal; Visualization: Equal; Writing, review \u0026amp;amp; editing \u0026ndash; original)\u003c/p\u003e\n\u003ch2\u003eData Availability\u003c/h2\u003e\n\u003cp\u003eThe datasets generated and analyzed during the current study are available from the corresponding author, Dr. Jae Keun Park ([email protected]), on reasonable request. All relevant data will be made available without undue reservation, in accordance with the data sharing policies of Scientific Reports.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eWilliams, E. et al. Updated guideline on the management of common bile duct stones (CBDS). \u003cem\u003eGut\u003c/em\u003e \u003cb\u003e66\u003c/b\u003e, 765\u0026ndash;782 (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCrockett, S. D., Wani, S., Gardner, T. B., Falck-Ytter, Y. \u0026amp; Barkun, A. N. Aga section. \u003cem\u003eGastroenterology\u003c/em\u003e \u003cb\u003e154\u003c/b\u003e, 1096\u0026ndash;1101 (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim, K. M. et al. Role of endoscopic ultrasonography in patients with intermediate probability of choledocholithiasis but a negative CT scan. \u003cem\u003eJ. Clin. Gastroenterol.\u003c/em\u003e \u003cb\u003e47\u003c/b\u003e, 449\u0026ndash;456 (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePark, J. K. et al. Long term outcome of EUS-based strategy for suspected choledocholithiasis but negative CT finding. \u003cem\u003eScand. J. Gastroenterol.\u003c/em\u003e \u003cb\u003e53\u003c/b\u003e, 1381\u0026ndash;1387 (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen, Y. T. \u0026amp; Wang, M. S. MR cholangiography 3D biliary tree automatic reconstruction system. \u003cem\u003eComput. Med. Imaging Graph.\u003c/em\u003e \u003cb\u003e28\u003c/b\u003e, 13\u0026ndash;20 (2004).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKr\u0026ouml;ner, P. T. et al. Artificial intelligence in gastroenterology: A state-of-the-art review. \u003cem\u003eWorld J. Gastroenterol.\u003c/em\u003e \u003cb\u003e27\u003c/b\u003e, 6794 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHou, J. U. et al. Efficacy of an artificial neural network algorithm based on thick-slab magnetic resonance cholangiopancreatography images for the automated diagnosis of common bile duct stones. \u003cem\u003eJ. Gastroenterol. Hepatol.\u003c/em\u003e \u003cb\u003e36\u003c/b\u003e, 3532\u0026ndash;3540 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMiyajima, C. et al. The Hippo Signaling Pathway Manipulates Cellular Senescence. \u003cem\u003eCells\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, 13 (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLogeswaran, R. A computer-aided multidisease diagnostic system using MRCP. \u003cem\u003eJ. Digit. Imaging\u003c/em\u003e. \u003cb\u003e21\u003c/b\u003e, 235\u0026ndash;242 (2008).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCorreia, F. P. \u0026amp; Louren\u0026ccedil;o, L. C. Artificial intelligence in the endoscopic approach of biliary tract diseases: A current review. \u003cem\u003eArtif. Intell. Gastrointest. Endoscopy\u003c/em\u003e. \u003cb\u003e3\u003c/b\u003e, 9\u0026ndash;15 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHe, K., Zhang, X., Ren, S. \u0026amp; Sun, J. in \u003cem\u003eProceedings of the IEEE conference on computer vision and pattern recognition.\u003c/em\u003e 770\u0026ndash;778.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHatamizadeh, A., Tang, N. V., Yang, Y., Myronenko, D. \u0026amp; Landman, A. B, et al. in \u003cem\u003eIEEE.\u003c/em\u003e 574\u0026ndash;584.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eByrne, M. F. et al. Real-time differentiation of adenomatous and hyperplastic diminutive colorectal polyps during analysis of unaltered videos of standard colonoscopy using a deep learning model. \u003cem\u003eGut\u003c/em\u003e \u003cb\u003e68\u003c/b\u003e, 94\u0026ndash;100 (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSultan, S. et al. AGA living clinical practice guideline on computer-aided detection\u0026ndash;Assisted colonoscopy. \u003cem\u003eGastroenterology\u003c/em\u003e \u003cb\u003e168\u003c/b\u003e, 691\u0026ndash;700 (2025).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMessmann, H. et al. Expected value of artificial intelligence in gastrointestinal endoscopy: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement. \u003cem\u003eEndoscopy\u003c/em\u003e \u003cb\u003e54\u003c/b\u003e, 1211\u0026ndash;1231. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1055/a-1950-5694\u003c/span\u003e\u003cspan address=\"10.1055/a-1950-5694\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKhalifa, A., Obeid, J. S., Erno, J. \u0026amp; Rockey, D. C. The role of artificial intelligence in hepatology research and practice. \u003cem\u003eCurr. Opin. Gastroenterol.\u003c/em\u003e \u003cb\u003e39\u003c/b\u003e, 175\u0026ndash;180 (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAgudo Castillo, B. et al. Advancements in biliopancreatic endoscopy - A comprehensive review of artificial intelligence in EUS and ERCP. \u003cem\u003eRev. Esp. Enferm Dig.\u003c/em\u003e \u003cb\u003e116\u003c/b\u003e, 613\u0026ndash;622. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.17235/reed.2024.10456/2024\u003c/span\u003e\u003cspan address=\"10.17235/reed.2024.10456/2024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Common bile duct stones, Computed tomography, Artificial neural network, Endoscopic retrograde cholangiopancreatography (ERCP), Endoscopic ultrasonography (EUS), Clinical decision support","lastPublishedDoi":"10.21203/rs.3.rs-7919357/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7919357/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003eComputed tomography (CT) is widely used in the initial evaluation of suspected common bile duct (CBD) stones, but limited sensitivity often necessitates additional endoscopic procedures. We developed and validated an artificial neural network (ANN) to enhance CT interpretation and assessed its potential to support endoscopic decision-making.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eWe An ANN model integrating UNETR for segmentation and ResNet50 for classification was trained to detect CBD stones on CT. Patients who underwent abdominal CT for suspected CBD stones between March 2018 and June 2023 at Hallym University Kangnam Sacred Heart Hospital were included. A retrospective derivation cohort (n\u0026thinsp;=\u0026thinsp;830) was used for model training, and a prospective validation cohort (n\u0026thinsp;=\u0026thinsp;225) for testing, with endoscopic retrograde cholangiopancreato- graphy (ERCP) serving as the reference standard. ANN performance was compared with that of expert radiologists and trainee radiologists with ANN assistance. Multivariate analysis evaluated clinical factors influencing diagnostic accuracy, and heatmap visualization assessed interpretability relevant to endoscopic decision-making.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eThe ANN achieved diagnostic accuracy comparable to expert radiologists (93.3% vs. 93.8%). When assisting trainees, accuracy improved from 82.2% (AUC 0.82) to 91.1% (AUC 0.91), approaching expert performance (93.8%; AUC 0.94). Stone type and bile duct diameter\u0026thinsp;\u0026gt;\u0026thinsp;10 mm significantly increased ANN detection rates. Heatmap visualization confirmed the plausibility of ANN predictions in both clearly identifiable lesions and indeterminate CT findings, improving interpretability for endoscopic decision-making.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e\u003cp\u003eThe ANN achieved expert-level diagnostic accuracy for detecting CBD stones. By enhancing CT interpretation, it may optimize ERCP indications, reduce unnecessary invasive procedures, and improve training for less-experienced clinicians. Prospective validation and integration into multimodal endoscopic workflows are warranted.\u003c/p\u003e","manuscriptTitle":"Integrating Artificial Intelligence into Endoscopic Decision-Making: ANN-Assisted CT for Common Bile Duct Stones","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-23 11:39:44","doi":"10.21203/rs.3.rs-7919357/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"50fb265e-7b2d-4054-b5ff-fa603e934eb8","owner":[],"postedDate":"November 23rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":57830379,"name":"Health sciences/Diseases"},{"id":57830380,"name":"Health sciences/Gastroenterology"},{"id":57830381,"name":"Health sciences/Health care"},{"id":57830382,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2025-12-08T10:06:18+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-23 11:39:44","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7919357","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7919357","identity":"rs-7919357","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}