GPU-based Volume Visualization of Functional MRI Data


Friedemann Rößler[1], Eduardo Tejada[2]
  Background    
Functional Magnetic Resonance Imaging (fMRI) describes the use of MRI to measure hemodynamic signals related to neural activity in the brain or spinal cord. The resulting images show which parts of the brain are activated by different types of physical sensations or activities, such as sight, sound or movement of a finger. This technique is used on the one hand by neuro surgeons for diagnosis and pre-operative planning. On the other hand cognitive scientists use fMRI images to learn more about how the brain works. The "brain mapping" of fMRI is achieved by setting up an advanced MRI scanner in a special way so that the increased blood flow to the activated areas of the brain shows up on Functional MRI scans. More information about fMRI and the so called BOLD (Blood Oxygen Level Dependent) effect cab be found at Introduction to fMRI[3].

During a cognitive experiment with fMRI data a number of scans of a single or several probands are taken while they solve certain tasks. A statistical analysis is then performed on these scans, which provides information about the activated brain regions. One of the most frequently used software for this analysis is SPM[4] (Statistical Parametric Mapping), which was developed by the Wellcome Department of Imaging Neuroscience of the University College London. Besides the statistical data, SPM generates 2D visualizations where the activation of the brain is mapped to an anatomical MRI image of a standard brain like in the figure below.

fMRI Aktivierung
fMRI activation in the brain.

  Project    
In cooperation with the Laboratory of Human Spatial Reasoning[5] (HSRLab), leaded by Prof. Markus Knauff and located at the Max Planck Institute for Biological Cybernetics Tübingen, and at the Center for Cognitive Science at the University of Freiburg, we currently develop an application for hardware-accelerated volume visualization of fMRI data and especially the statistical results obtained from SPM. In this project we are following two objectives:
On the one hand we want to provide the cognitive scientists of the HSRLab with an interactive and easy-to-use tool for three-dimensional exploration of fMRI data and on the other hand produce high quality visualizations for education and publication.

One of the main functionalities of our tool is the integrated volume rendering of an anatomical MRI volume of a standard brain and the color coded activation of a statistical activation map. For this we developed a hardware accelerated slice-based volume renderer for the correctly overlaying of "arbitrary" (only limited by the texture memory of the GPU) number of volumes. This renderer firstly slices all the volumes in a view-aligned manner and then sorts these slices on a slice stack in a back-to-front order. The slices on this stack are then rendered one by one. If the volumes are to be rendered in different manners, for example with different transfer functions or different blending modes, the renderer automatically changes all these parameters slice by slice.

A drawback of the correctly overlayed rendering of the brain and the activation is that the activation, which is located inside the brain, is only visible if the brain is rendered highly transparent. But the more transparent the brain is rendered the less it's structures can be perceived. For this reason we are currently extending the tool to support different techniques like clipping and transparent isosurfaces to improve the integrated 3D visualization of the brain's anatomical structures and the physiological activation in a single view.

Our fMRI visualization tool is based on a general object oriented framework for volume visualization, which we are developing in parallel. This framework provides a platform which can be used for all kind of hardware-accelerated volume visualization applications, not only for fMRI. It allows the easy reuse of application independent modules, like the basic volume renderer, and replacement of general modules by specific ones, optimized for a certain task.

  Results    
The following pictures show the first visual results of our fMRI Volume Visualization Tool:

[6] [7]

[8]

The fMRI Visualization Application.

  Contact    
Friedemann Rößler[9]

      University of Stuttgart, Institute for Computer Science,
Visualization and Interactive Systems Group

http://www.vis.uni-stuttgart.de/eng/research/fields/current/fmrivis/index.html