Goethe Center for Scientific Computing (G-CSC)

Goethe University Frankfurt

Personal tools

NeuRA 2 - Home

NeuRA 2 - Home

NeuRA 2

 

 

The GPU-based implementation of the NEUron Reconstruction Algorithm

 

 

NeuRA 2 is the new GPU-based implementation of the Neuron Reconstruction Algorithm. Since 2-photon and confocal microscopy significantly improved within the last years, three dimensional microscope images of neuron cells with a resolution of 2048x2048x400 are available nowadays. Unfortunately the original serial implementation of NeuRA is not able to process such large data amounts. However, NeuRA 2 uses the massive parallel architecture of state-of-the-art graphic processing units (GPUs) and is able to distribute the data among multiple GPUs at single computers or computer clusters, allowing to process images of a size of up to several GBytes.

For extracting the surface morphology from objects contained in three dimensional images, NeuRA 2 provides following image processing operators:

  • Inertia based anisotropic diffusion filtering
  • Multiscale line enhancement filtering
  • Global segmentation
  • Statistical Otsu segmentation
  • Seeded region grow segmentation
  • Marching cubes mesh generator 

 

The object-oriented software design of NeuRA 2 strictly encapsulates data access, data processing and workflow control, allowing the fast processing of large images as well as the easy extension with other image processing operators.

NeuRA 2 automatically checks the available GPU devices and divides the input image in suitable overlapping subimages of a size, which can be processed on the single GPUs. Interpolation in the overlap regions, when reassembling the single subimages, guarantee continuous output images.

 

More information available under www.neura.org