Image processing: 3D-Polarized Light Imaging workflow

This project is a collaboration of scientists working in two HBP Subprojects:

  • Human Brain Organisation (SP2)
  • HPAC Platform (SP7)

Markus Axer (SP2) is leading from a scientific point of view, Oliver Bücker (SP7) is responsible for the technical part.


3D-Polarized Light Imaging (3D-PLI) is a technique used to extract the orientation of nerve fibres at micrometer scale. Two different setups at the Institute of Neuroscience and Medicine (INM-1), Forschungszentrum Jülich are being used for the imaging.

A human brain is cut into about 2500 sections that are imaged using 3D-PLI microscopy, which results in about 1.3 micron pixel size and 80,000 x 100,000 pixels per section. The polarisation filters used in the imaging process are rotated 18 times so that 18 different images per section are available. The mentioned resolution and the 18 images result in a memory consumption of about 750 GB/section for the raw data, which is then transferred to the storage of the HPC systems at Jülich Supercomputing Centre (JSC), Forschungszentrum Jülich. The supercomputer JURECA is used to run the complex workflow that extracts the fibre orientations (3D vector per pixel) from the raw data. This analysis is not based on staining on the images but it uses the birefringence of the myelin sheaths that shield the axons of nerve fibres. Most tools in the workflow are developed at Forschungszentrum Jülich in the Fibre Architecture group led by Markus Axer and in close collaboration with Oliver Bücker and the Simulation Laboratory Neuroscience (both JSC). To speed up the workflow, it has been implemented using the UNICORE workflow engine (supported by André Giesler, JSC), which allows analysing a section in the order of hours instead of weeks. The goal of this use case project is now to integrate the existing UNICORE workflow (see figure below) into the infrastructure of the HBP.

Main steps of UNICORE-based 3D-PLI workflow
Main steps of UNICORE-based 3D-PLI workflow

It is planned to have the access mechanisms ready before the end of the Ramp-up Phase. A first release of the UNICORE provenance tracking is planned for summer 2016; the Fibre Architecture group already started preparing workflow and tools for it. The viewer integration is planned for the beginning of SGA1; a locally running viewer is already available.

Problems to be solved

The following problems require special attention and need to be solved in co-design with the users for implementing the use case successfully.

  • The UNICORE workflow engine needs to be well integrated with the other parts of the HPAC Platform to enable a seamless usage inside the HBP environment.
  • Provenance tracking of the entire, complex, UNICORE-driven workflow is needed in order to enable reproducibility.
  • Establishing access mechanisms to the HPC storage for non-HPC users to enable sharing of results.