A Perspective on FAIR and Scalable Access to Large Image Data

Julia Thönnißen, Sarah Oliveira, Alexander Oberstrass, Jan-Oliver Kropp, Xiao Gui, Christian Schiffer, Timo Dickscheid

Published: 04 Aug 2025, Last Modified: 27 Feb 2026ZenodoEveryoneRevisionsCC BY-SA 4.0
Abstract: The rapid development of new imaging technologies across scientific domains–especially high-throughput technologies–results in a growing volume of image datasets in the Tera- to Petabyte scale. Efficient visualization and analysis of such massive image resources is critical but remains challenging due to the sheer size of the data, its continuous growth, and the limitations of conventional software tools to address these problems. Tools for visualization, annotation and analysis of large image data are confronted with the fundamental dilemma of balancing computational efficiency and memory requirements. Many tools are unable to process large datasets due to memory constraints, requiring workarounds like downsampling. On the other hand, solutions that can handle large data efficiently often rely on specialized or even proprietary file formats, limiting interoperability with other software. This reflects diverging requirements: storage favours compression for efficiency, analysis demands fast data access, and visualization requires tiled, multi-resolution representations. Lacking a unified approach for these conflicting needs, the operation of large and dynamically evolving image repositories in practice often requires undesirable data conversions and costly data duplication. In addressing these challenges, the bioimaging community increasingly adheres to the FAIR principles [1] through national and international initiatives [2], [3], [4]. For example, the Open Microscopy Environment (OME) fosters standards such as OME-TIFF [5] and its cloud-native successor OME-NGFF [6]; BioFormats [7] and OMERO [8] facilitate metadata-rich data handling across diverse platforms; and BrAinPI [9] provides web-based visualization of images via Neuroglancer [10]. These tools represent important developments towards more efficient and standardized use of bioimaging data. However, for very large and dynamically growing repositories, it is still not feasible to settle on a single standard for a subset of these tools, in particular in the light of very diverging needs for massively parallel processing on HPC systems. Therefore, converting data to a single target format is often not a practical solution. We propose a concept for a modular image delivery service which acts as a middleware between large image data resources and applications, serving image data from a cloud resource in multiple requested representations on demand. The service allows reading data stored in different...