Streamlining the User Experiencefor Reproducible Image Analysis in Galaxy

Streamlining the User Experience for Reproducible Image Analysis in Galaxy Leonid Kostrykin, Beatriz Serrano-Solano, Riccardo Massei, Karl Rohr Biomedical Computer Vision Group (BMCV) BioQuant, IPMB, Heidelberg University 1 Leonid Kostrykin, BMCV Group, Heidelberg University Streamlining the User Experience (2024–2025) Toolset Interoperability Efficiency Training Reproducibility Robustness 2 Leonid Kostrykin, BMCV Group, Heidelberg University PlantSeg 2024–2025: Toolset Extensions and Major Updates Color Deconvolution Split Image Along Axis Background Removal Spot Detection Crop Image OMERO Tool Suite Image Processing & Data Management Tools Image Analysis Tools SuperDSM UltralyticsBioImage.IOCellpose (Wolny et al., Elife 2020) (Stringer et al., Nat Methods 2021) (Ouyang et al., bioRxiv 2022) (Kostrykin and Rohr, IEEE ISBI 2024) Anup Kumar Yi Sun 3 Leonid Kostrykin, BMCV Group, Heidelberg University PlantSeg 2024–2025: Toolset Extensions and Major Updates Color Deconvolution Split Image Along Axis Background Removal Spot Detection Crop Image OMERO Tool Suite Image Processing & Data Management Tools SuperDSM UltralyticsBioImage.IOCellpose (Wolny et al., Elife 2020) (Stringer et al., Nat Methods 2021) (Ouyang et al., bioRxiv 2022) (Kostrykin and Rohr, IEEE ISBI 2024) Anup Kumar Yi Sun Image Analysis Tools 4 Leonid Kostrykin, BMCV Group, Heidelberg University PlantSeg 2024–2025: Toolset Extensions and Major Updates Color Deconvolution Split Image Along Axis Background Removal Spot Detection Crop Image OMERO Tool Suite Image Processing & Data Management Tools Image Analysis Tools SuperDSM UltralyticsBioImage.IOCellpose (Wolny et al., Elife 2020) (Stringer et al., Nat Methods 2021) (Ouyang et al., bioRxiv 2022) (Kostrykin and Rohr, IEEE ISBI 2024) Anup Kumar Yi Sun 5 Leonid Kostrykin, BMCV Group, Heidelberg University OMERO Integration for Seamless Workflows OMERO: Open Microscopy Environment Remote Objects Massei et al., Sci Rep 2025 Image data and metadata management system: Download IDR/OMERO OMERO Image Import For example: 6 Leonid Kostrykin, BMCV Group, Heidelberg University Enhanced Interoperability Example Use Case: Quantification of Gel Bands on Western Blots Initial GeoJSON Support Image of a western blot (Degasperi et al., PLoS ONE 2014) Manual Annotation of the image in QuPath (Galaxy IT) 7 Leonid Kostrykin, BMCV Group, Heidelberg University Tool for quantification of image intensities Enhanced Interoperability Example Use Case: Quantification of Gel Bands on Western Blots Initial GeoJSON Support Image of a western blot (Degasperi et al., PLoS ONE 2014) Manual Annotation of the image in QuPath (Galaxy IT) Annotations as label map 8 Leonid Kostrykin, BMCV Group, Heidelberg University Tool for quantification of image intensities Enhanced Interoperability Example Use Case: Quantification of Gel Bands on Western Blots Initial GeoJSON Support Image of a western blot (Degasperi et al., PLoS ONE 2014) Manual Annotation of the image in QuPath (Galaxy IT) Tool for conversion of GeoJSON to label map Annotations as label map Pavankumar Videm 15 Leonid Kostrykin, BMCV Group, Heidelberg University Enhanced Reproducibility Cellpose Python module Runs inside Docker container, NumPy built against MKL Galaxy EU: Containers run on VMs, Random results (A, B, C, D, etc.) Bare metal: Always Result A Fixed with MKL_NUM_THREADS=1 Concurrency issue in MKL? …or: Containerization? Virtualization? Hardware? Cellpose Integration Original image Predicted outlines Predicted masks Predicted flow field Original image Predicted outlines Predicted masks Predicted flow field Original image Predicted outlines Predicted masks Predicted flow field Original image Predicted outlines Predicted masks Predicted flow field 16 Leonid Kostrykin, BMCV Group, Heidelberg University Robust Graph Pruning !"=min !⊆#'($) %∈! **s.t.**."=./ Sparse, energy-minimal family !of image regions: Admissible image regions Set energy function with ! > # ! = {$!, $", $#, $$} $! $"$# $$ Clustered cell nuclei Adjacency "!"""#"$ "!" "#$ ""# "!"# '!"# = '!∪ '"∪ '# '!" = '!∪ '" … ""#$ # = %%, %&, %', %(, … , %%&'( "!"#$ #′ =%%, %&, %', %(, %&' !"′ ≤ ! "+ & ⋅ (#* − 2) 1) Remove any edge !, ! ∪ $ with %!! ∪ $ > %!! + %!($), 2) Retain only components which contain *, then: Algorithm: Up to 2,1xfaster than Kostrykin and Rohr, IEEE TPAMI 2023 Enhanced Efficiency SuperDSM: Deformable Shape Models for Segmentation of Cell Nuclei —2024 Update Kostrykin and Rohr, IEEE ISBI 2024 Deformable Shape Models !!",$=&""+("$ Deformable surface: )!",$= ℓ + !⊙ !!",$+ - ⋅ $# Set energy function: Convex energy of the deformable surface: ! : polynomial parameters #:deformation parameters $:image region %!:offset image intensities :logistic loss &:regularization of deformations ':sparsity parameter 2nd order polynomial basis functions Gaussian basis functions Image plane ("= 0 /$0 = inf %,')!",$+ 4 Convex optimization Now in Galaxy Original Images Segmentation results Ground truth 17 Leonid Kostrykin, BMCV Group, Heidelberg University Enhanced Robustness Extended image metadata, e.g.: Image width, height, depth, channels, frames, axes Enhanced TIFF support in Galaxy tools, e.g.: Treat single-channel images that have no channel information identically to images that have a single channel Galaxy tool for inference using BioImage.IO models: Support for 3-D images has been fixed Anup Kumar TIFF List of channels Channel 1 2-D array of pixel values TIFF 2-D array of pixel values vs. Example: 18 Leonid Kostrykin, BMCV Group, Heidelberg University Conclusion •11+ tools and tool suites added or received major updates (e.g., OMERO) •Enhanced interoperability (e.g., initial GeoJSON support) •Enhanced reproducibility (e.g., Cellpose integration) •Enhanced efficiency (e.g., SuperDSM) •Enhanced robustness (e.g., bugfixes, enhanced testing capabilities, metadata) 19 Leonid Kostrykin, BMCV Group, Heidelberg University Acknowledgements ●Galaxy Image Community and the Euro-BioImaging FAIR Image Data Workflows Expert Group Rafael Andrade Buono, Diana Chiang Jurado, Lucille Delisle, Melanie Föll, Anne Fouilloux, Daniel Franco-Barranco, Nadia Goué, Hans-Rudolf Hotz, Khaled Jum'ah, Matúš Kalaš, Isabel Kemmer, Anup Kumar, Sylvia Le Dévédec, Torec Luik, Aastha Mathur, Arrate Muñoz Barrutia, Pierre Osteil, Bugra Özdemir, Maarten Paul, Feriel Ramdhane, Yi Sun, Papawee Sutthirat, Dado Tokic, Vladimír Ulman, Ilva van Houwelingen, Pavankumar Videm, Rebecca Vorimo, Guillaume Witz, Tatiana Woller https://www.eurobioimaging.eu/expert-groups/fair-image-data-workflows-expert-group ●Thomas Wollmann (Heidelberg University, BMCV Group) for his pioneering work on Galaxy Image Analysis ●Björn Grüning (University of Freiburg) for technical advice with everything related to Galaxy HEIDELBERG CENTER FOR HUMAN BIOINFORMATICS (HD-HuB)