MIQR-CC Dataset

DescriptionEndoscopic Retrograde Cholangiopancreatography (ERCP) is a key procedure for diagnosing and treating biliary and pancreatic diseases. Artificial intelligence methods have shown promise in automating ERCP image interpretation; however, publicly available ERCP datasets remain scarce. This collection aims to address this limitation by providing a large, curated, and clinically annotated ERCP fluoroscopy image dataset to support research in automatic image analysis and diagnosis.
Dataset OverviewTotal patients: 1,602Raw images: 19,018 PNG imagesProcessed images: 19,317 PNG imagesLabeled images: 5,519 images with expert annotationsImage modality: Fluoroscopy during ERCP proceduresData format: PNG images + CSV metadata fileClinical ContentThe dataset includes all fluoroscopic images acquired throughout the ERCP workflow, encompassing pre-cannulation images used for anatomical orientation and procedural planning, cholangiographic images obtained during contrast injection and therapeutic maneuvers, as well as post-procedural images documenting the final outcome of the intervention.
Annotated images include clinically relevant categories:
Biliary lithiasisBiliary leaksBenign stricturesMalignant stricturesNormal findingsAnnotation ProcessAll labeled images were manually inspected and annotated by two gastroenterologists with more than 5 years of ERCP experience and over 400 procedures annually. Annotations were independently reviewed by a senior gastroenterologist with more than 20 years of experience to ensure diagnostic reliability.
Data AcquisitionImages were retrospectively extracted from the SECTRA PACS system of a tertiary care high-volume ERCP center performing over 750 procedures annually. Procedures were conducted by experienced gastroenterologists, and fluoroscopic acquisition was performed by trained radiology technicians. Exams were independently reviewed by a second gastroenterologist for consistency.
Fluoroscopic systems used include Ziehm Vision RFD 3D and Philips PCR Eleva units, introducing natural variability in image characteristics.
Data ProcessingA multi-stage Python-based pipeline was applied to:
Export DICOM filesAnonymize all patient-identifiable informationConvert DICOM to PNGPartition raw fluoroscopic frames into individual processed imagesTwo data folders are provided:
Raw: Original anonymized fluoroscopy framesProcessed: Cropped sub-images extracted from raw framesA metadata file (metadata.csv) links each processed image to its corresponding raw image and includes anonymized patient-level and image-level descriptors. Each row corresponds to one processed image.
Potential ApplicationsThis dataset supports research in:
Automatic classification of ERCP findingsAI-assisted diagnosis of biliary and pancreatic diseasesImage-based analysis of therapeutic endoscopyFully supervised, semi-supervised, and weakly supervised learning approachesEthics ApprovalThe study was approved by the Board of Directors of Unidade Local de Saúde do Alto Minho (ULSAM).
Approval ID: 243/CA-2025
All data were retrospectively collected and fully anonymized.