Preclinical Development of a 3D Bili-MRI Reconstruction Tool and an Artificial Intelligence Algorithm to Assist Endoscopists in Performing Endoscopic Retrograde Cholangiopancreatography (ERCP)

Overview

ERCP is a complex interventional endoscopic procedure based on 2D visualization of the biliary tract. 3D modeling and AI are promising solutions to improve navigation and safety during ERCP. The MAAGIE project was designed to create one of these innovative solutions and received ANR funding in 2024. It includes several Work Packages, among which the first WP involves creating a database from imaging data of the biliary tract already collected during clinical care. The main objective of our study will be to develop and validate a tool that enables 3-dimensional modeling of the biliary tract from preoperative MRCP and abdominal CT scan sequences, allowing better visualization and understanding of patients' biliary anatomy before and during ERCP.

ERCP is a complex interventional endoscopic procedure based on 2D visualization of the biliary tract. 3D modeling and AI are promising solutions to improve navigation and safety during ERCP. The MAAGIE project was designed to create one of these innovative solutions and received ANR funding in 2024. It includes several Work Packages, among which the first WP involves creating a database from imaging data of the biliary tract already collected during clinical care. The main objective of our study will be to develop and validate a tool that enables 3-dimensional modeling of the biliary tract from preoperative MRCP and abdominal CT scan sequences, allowing better visualization and understanding of patients' biliary anatomy before and during ERCP. This is a bicentric retrospective study based on imaging data collected during routine care. Patients were monitored as part of the usual care pathway for individuals undergoing ERCP in the endoscopy units of Saint Antoine Hospital or Henri Mondor Hospital for a biliary condition related to a biliary stricture (benign or malignant). The research does not alter patient management in any way. The first step will involve identifying all patients who underwent ERCP in the endoscopy units of Saint Antoine Hospital or Mondor Hospital for a biliary condition related to a biliary stricture (benign or malignant) and who had both an MRCP and an abdominal CT scan performed in the imaging departments of the same hospitals, from 2020 through the end of 2025. After verifying eligibility criteria, a specific information notice about this study will be sent to them. If no response is received within one month, the collected data may be reused for research purposes. The imaging data collected will support the development and validation of 3D reconstruction/registration software (3D modeling of the biliary tract using mathematical algorithms for medical imaging and artificial intelligence networks applied to the acquired images). Pseudonymized data will be centralized under the supervision of Prof. Camus on a secure AP-HP server, with a correspondence table created for follow-up, accessible only to Prof. Camus. At the end of the study, data will be securely transferred to the partners for analysis and tool development. All data transfers will be performed using secure, encrypted external storage devices according to local AP-HP IT department procedures. The data volume is determined based on image-processing capacity and technological development goals. A dataset of 100 patients provides sufficient anatomical variability to train and test AI models. No inferential statistical analysis is planned. Tool performance will be evaluated using image-quality metrics (DSC, IoU) and localization error indicators for the algorithms. Finally, exchanges with project partners will support the development and validation of the 3D reconstruction tool. The analysis will be conducted by the partner teams at LIP6 and ISIR.