Facilitating Endoscopy Video Recording for Scientific Purposes: Experiences from Three Endoscopy Centers

Aims In routine clinical practice, endoscopic examinations are usually documented with still images alongside the examiner’s written report. However, capturing the complete video stream of these examinations could generate substantial, high-quality datasets valuable for scientific research. Artificial intelligence (AI) development in endoscopy, in particular, requires extensive data to create robust models that have the potential to support endoscopists, enhance diagnostic accuracy, and improve patient safety. This abstract outlines the technical and logistical challenges of routine video recording and presents practical solutions from three large tertiary care endoscopy centers in Germany and the United States, aiming to encourage wider adoption and foster a rich, diverse data landscape for advancing AI in endoscopy.

Methods We analyzed and compared practices for implementing routine endoscopy video recording at three tertiary care centers: Dresden (Germany), Würzburg (Germany), and Boston (USA). The qualitative analysis focused on identifying and comparing solutions across two main challenge domains: technical and personnel.

Results Each center employed customized hardware setups to meet the unique requirements of its endoscopy unit. Consequently, all three centers had different recording devices installed that capture the video signal from the endoscopic video processor. All centers stored and archived video recordings on a hospital network drive. However, solutions for managing associated research data—such as examination reports, patient demographics, or biopsy results—varied, ranging from commercial cloud platforms (e.g., Microsoft Teams) to custom-built SQL databases. Although much of the recording process was automated, varying levels of manual intervention were still required. For instance, integration with electronic health records, privacy concerns, automating the initiation, and termination of video recording and the processing of raw video data remain technical challenges that currently still need human oversight. For personnel-related challenges, providing regular project updates helps maintain staff morale, and having dedicated research staff on-site to manage daily operations has proven essential, especially in the implementation phase.

Conclusions Despite differing requirements and setups, all three centers successfully implemented routine recording of endoscopy examinations for scientific purposes, contributing to expanding databases of high-quality endoscopy recordings. However, to develop reliable and generalizable AI models, more diverse datasets are needed, encompassing a wider range of endoscopists, procedures, endoscopes, and video processors. This will enable the creation of effective AI tools that support examiners and ultimately enhance patient safety. By presenting and comparing strategies from these three participating centers, we aim to encourage others to join in creating high-quality datasets for the scientific community.

 

Scroll to top

Conflicts of Interest

Authors do not have any conflict of interest to disclose.

Publikationsverlauf

Artikel online veröffentlicht:
27. März 2025