Document Type : Original Research
Authors
- Mona Mohammad-Asghari 1
- Taha Pishro Dabaghiyan 2
- Tahereh Mahmoudi 3, 4
- Ramin Niknam 5, 6
- Hossein Parsaei 3
- Mohammad Mehdi Movahedi 3
- Kamran Bagheri Lankarani 5
- Fardad Ejtehadi 6
- Seyed Ebrahim Fallahzadeh Abarghooei 6
- Seyed Alireza Taghavi 6
- Gholam Reza Sivandzadeh 6
- Saghar Alihosseini 6
- Seyed Ali Malek-Hosseini 7
1 Research Committee, Shiraz University of Medical Sciences, Shiraz, Iran
2 1. Student Research Committee, Shiraz University of Medical Sciences, Shiraz, Iran
3 Department of Medical Physics and Engineering, School of Medicine, Shiraz University of Medical Sciences, Shiraz, Iran
4 Nanomedicine and Nanobiology Research Center, Shiraz University of Medical Sciences, Shiraz, Iran
5 Health Policy Research Center, Shiraz University of Medical Sciences, Shiraz, Iran
6 Gastroenterohepatology Research Center, Shiraz University of Medical Sciences, Shiraz, Iran
7 Shiraz Transplant Center, Abu Ali Sina Hospital, Shiraz University of Medical Sciences, Shiraz, Iran
Abstract
Background: Extraction of Common Bile Duct (CBD) stones via Endoscopic Retrograde Cholangiopancreatography (ERCP) can be technically challenging, with difficulty influenced by anatomical and procedural factors. Accurate automated assessment of these features may facilitate objective difficulty scoring and procedural planning. This study presents an intelligent Expert System for Difficulty Scoring (ESDS) that integrates anatomical features automatically extracted from fluoroscopic images with demographic and procedural parameters to predict the technical difficulty scoring.
Objective: To develop and validate an automated expert system for objective prediction of the technical difficulty of CBD stone extraction during ERCP using imaging and clinical features.
Material and Methods: This retrospective study used 52 ERCP cases to develop the ESDS. Image preprocessing was applied to enhance contrast and preserve structural details. The CBD, stones, and duodenoscope were delineated, and imaging-based features were extracted using customized image processing algorithms. These included stone diameter, proximal and distal CBD diameters, distal CBD angulation, proximal and distal CBD arm lengths. Demographic and procedural variables were also incorporated. An if then rule-based framework integrated these features to generate difficulty scores.
Results: The automated feature extraction method was validated against expert annotations using Bland Altman analysis, demonstrating excellent agreement between automated and manual measurements. The average Intra-Class Correlation Coefficient (ICC) across the evaluated anatomical features was approximately 99.62%, with a Root Mean Square Error (RMSE) of 0.36.
Conclusion: By enabling objective difficulty scoring, it has the potential to improve ERCP quality, reduce procedure time, and lower risks associated with stone extraction.
Keywords
)