Refining biomedical images: behind the scenes of organizing and preprocessing data

Advances in image enhancement, transformation, co-registration, no-reference evaluation, and FAIR data management of biomedical images

Abstract Text:

Biomedical research often involves managing diverse, complex imaging data. Integrating data from multiple imaging techniques poses additional challenges. In this work, we present advances across the image processing pipeline, covering pre-processing, quality assessment, image co-registration, and data management.

Pre-processing is essential for improving data quality by reducing artifacts and enhancing analysis performance. We show that tailoring methods to specific data and tasks yields more robust results (Dehbozorgi, 2024). To further enhance image resolution, we developed a physics-informed super-resolution CNN (incSRCNN), which embeds noise models directly into its loss function.

Besides improving images, it is often necessary to quantify the quality of raw images to evaluate by the measurement itself. We introduce Multi-Marker Image Quality Assessment (MM-IQA), a no-reference, machine learning method using physics-based markers (Corbetta, 2025). We also explore how color transformations affect co-registration and propose a two-stage pipeline combining rigid and non-rigid alignment, supported by unsupervised metrics for evaluating data fusion benefits.

Finally, we developed LEO, a platform linking electronic lab notebooks with the OMERO image repository. LEO enables seamless data integration and supports the FAIR principles, making image data more accessible and reusable.

ACKNOWLEDGMENT

This work is supported by the BMBF, funding program Photonics Research Germany (LPI-BT3-IPHT, FKZ: 13N15708; LPI-BT3-FSU, FKZ: 13N15710; LPI-BT5-FSU, FKZ: 13N15719) and is integrated into the Leibniz Center for Photonics in Infection Research (LPI). The work is co-funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) for the project NFDI4Bioimage (501864659).