ERCP Fluoroscopy Images With Expert Clinical Annotations

19,018 raw and 19,317 processed fluoroscopy images from 1,602 patient ERCP procedures, curated by gastroenterologists Alda Andrade et al. and published on figshare in April 2026. A subset of 5,519 images is annotated with expert labels for conditions like biliary lithiasis, leaks, and strictures.

Use Cases

• Train image classification models to detect 'biliary lithiasis' or 'malignant strictures' from annotated PNG images.
• Develop object detection algorithms using the 'processed' sub-images to localize anatomical landmarks or pathologies within raw fluoroscopy frames.
• Validate model generalizability across different imaging systems by leveraging data from Ziehm Vision RFD 3D and Philips PCR Eleva units.
• Perform multi-label classification on the 5,519 labeled images across categories including 'benign strictures', 'biliary leaks', and 'normal findings'.
• Use the accompanying CSV metadata file for patient-level analysis or to stratify image data by clinical annotation categories.

Strengths

• 5,519 images feature expert annotations by gastroenterologists with over 5 years of experience.
• Data originates from a high-volume center performing over 750 ERCP procedures annually, ensuring clinical relevance.
• Includes 19,018 raw and 19,317 processed images, providing flexibility for different computer vision tasks.

Limitations

• Annotations cover only 5,519 of the 19,317 processed images, limiting the fully labeled subset.
• Data is sourced from a single tertiary care center, which may introduce institutional bias.
• The dataset is retrospective, extracted from a PACS system, and may not reflect real-time procedural variability.

Provenance

Source

Retrospectively extracted from the SECTRA PACS system of a tertiary care high-volume ERCP center.

Collection Method

A multi-stage Python pipeline exported, anonymized, and converted DICOM files to PNG images, with manual annotation by experienced gastroenterologists.

Freshness

Last updated April 2026.