Towards an open high-performance platform for fully-automated analysis of whole organ light-sheet fluorescence microscopy data

CONTACT [email protected] Research Group Systems Biology / Bioinformatics References: [1] Surname, N., Surname, N. N., and Surname, N. N. (2008). Nature Reviews Immunology Picture of you CONTACT [email protected] Research Group Applied Systems Biology References: [1] Anika Klingberg et al., “Fully Automated Evaluation of Total Glomerular Number and Capillary Tuft Size in Nephritic Kidneys Using Lightsheet Microscopy,” Journal of the American Society of Nephrology: JASN 28, no. 2 (February 2017): 452–59, https://doi.org/10.1681/ASN.2016020232. [2] David Twapokera Mzinza et al., “Application of Light Sheet Microscopy for Qualitative and Quantitative Analysis of Bronchus-Associated Lymphoid Tissue in Mice,” Cellular & Molecular Immunology, February 12, 2018, https://doi.org/10.1038/cmi.2017.150. Towards an open high-performance platform for fully-automated analysis of whole organ light-sheet fluorescence microscopy data Ruman Gerst1,2, Anna Medyukhina1, and Marc Thilo Figge1,2 1 Applied Systems Biology, Leibniz Institute for Natural Product Research and Infection Biology – Hans-Knöll-Institute, Jena, Germany 2 Faculty of Biological Sciences, Friedrich-Schiller-University Jena, Germany 5. Bronchioles detection module (In progress) Automated assessment of Bronchus-associated lymphoid tissue (BALT) to investigate lung infection requires to segment the bronchioles of the lung, visible as holes with strong borders in LSFM images. We are currently developing an approach to segment those highly irregular structures. Light-sheet fluorescence microscopy (LSFM) allows quantitative threedimensional analysis of whole organs. This includes the evaluation of structural changes such as a reduced number of glomeruli in kidneys [1] or the formation of bronchus-associated lymphoid tissue (BALT) caused by lung inflammation [2]. MISA++ Modular Image Stack Analysis for C++ Implemented in modern C++ Automated parallelization of workloads Reusable modules for easy extension and integration Memory-efficient Fast Flexible MISA++ Library Tissue detection Glomeruli detection Tissue detection MISA++ library External pipeline Tissue detection Standalone Build executable Parameter documentation MISA++ modules can be used in C++ code to create other modules or exported to a standalone executable that can be integrated into other pipelines or applications such as Fiji/ImageJ. 2. Framework for automated analysis 3. Tissue detection module A B C 2D Segmented tissue of Lung 🄰 and Kidney 🄱, 🄲 🄰 🄱 🄲 Future: Graph assisted 3D object detection Removal of false positive 2D tissue detection results Quantification Number of pixels, volume 2D segmentation Algorithms based on percentiles, superpixels or auto thresholding In comparison to the implementation published by Klingberg et al. [1] (Python), MISA++ calculates the same work up to 148 times faster if all images are available at the same time. If the analysis is done per kidney (PK), the calculation is still approximately 27 times faster. Parallelization using 30 threads. MISA++ MISA++ (PK) Python 0 1000 2000 3000 4000 5000 26 min 142 min 3907 min Runtime (min) Glomeruli are functional structures within the renal cortex that are damaged by diseases and toxins [1]. Tissue segmentation MISA++ module 2D glomeruli segmentation Klingberg et al. or Localized Otsu + Shape 3D object detection and quantification Number of glomeruli, volume, diameter Glomeruli (red) inside the tissue (green) 4. Glomeruli detection module 🄳 Detection of Bronchioles (red) inside the tissue (green) suffers from a high number of false positive holes and borders, making it difficult to segment true positive objects. To solve this issue, holes 🄴 are segmented independently from borders 🄵. Tissue segmentation MISA++ module 🄳 🄴 🄵 Border splitting Borders (colored) shared by multiple holes (black) are split between the holes. A graph is built, with holes and borders as edges. The edge weight is the number of neighboring pixels between holes/borders and to the background. The score of a hole (star) is the summarized flow to the background. The goal is to remove all B-B edges. Many common tasks for different data Memory intensive Long processing Repeated code Large datasets } 1500 350 300 300 2300350 1200 1950 H1 H2 H3B B B B Background 1200 1950 0 1. Analysis of whole organs