Research- and Service Group
IT & Electronics Development Head: Dr. Stefan Vollmar
(IT & Elektronik-Entwicklung)
Visualization and Analysis of PET data
A software package for image analysis and visualization of volume data generated by medical tomography systems has been developed at our institute: VINCI – “Volume Imaging in Neurological Research, Co-Registration and ROIs included”. It has been developed for the analysis of neurological PET studies including fully automatic co-registration of image volumes and image fusion (multi-modality imaging) in addition to VOI and ROI evaluation.
VINCI is an ongoing project and current development is associated with group Cortical Networks. A „Lite-Version“ is available from the VINCI-Homepage .
VINCI (full screen, resolution: 1024 × 768 pixels) with image data processed for surgery planning. All relevant parts of the user interface (widgets) are scrollable, e.g. OrthoView and PlanesView, and can be resized and repositioned as required. VINCI projects can also take advantage of significantly larger screen sizes and have support for multi-display settings.
- Vergleich verschiedener Farbskalen für Falschfarbenbilder. (a) – (e) zeigen denselben transaxialen Schnitt durch eine PET-Aufnahme, negative Werte sind Rekonstruktions-Artefakte.
Figure left:
(a) a gray scale visualization of the full data range (γ=1.0) (b) gamma-corrected gray scale (γ=0.5) with data range adjusted to emphasize streak artefacts (c) linear gray scale visualization with data range suitable for diagnostic purposes, (d) visualization with „rainbow“ color scale (e) „hot metal“ color table. Further color tables are depicted in (f) to (j). Color scales with distinct levels like „GraySteps“ and „Bronson“ are suitable to visualize statistical properties of activation studies.
Quality Assurance (VHIST)
Methods and tools for analysis of image data are becoming increasingly complex. Thus, a detailed and complete documentation of all processing steps (e.g. reconstruction, quantification and statistical analysis) is desirable, even mandatory for good scientific practice: which software tools in which versions have been used with which arguments.
The VHIST-project provides tools for documentation:
- VHIST is a file format description which can be viewed with PDF browsers (e.g. Adobe Acrobat). A VHIST file can be processed with popular programming languages, no knowledge of the PDF specification is required.
- A VHIST file provides several means for validation (e.g. MD5 checksums) which allows to reliably extract embedded data for future reference.
- It can be used to generate a detailed processing log („history“).
- The reference implementation includes programs to append documentation on a particular workflow step to a VHIST file or to extract information.
- All software components aim at platform independence and comply with an OpenSource licence, increasing the project’s attractiveness for use outside of our institute.
Cluster Reconstruction
Reconstruction of PET data has been an active research topic for a number of years (Vollmar et al, HeinzelCluster 2002). A PET study with our HRRT system generates approx. 30 GB (gigabytes) of data (a single patient). Using a compute cluster, an image volume which still has size of about 50 MB (megabytes) is reconstructed from this.
One challenge posed by the HRRT system is to use iterative reconstruction methods - despite the large data sets - to better take into account the scanner’s imaging characteristics. These methods are very time consuming: even with an efficient parallelization of the reconstruction scheme, several hours of computing time on a 16 CPU compute cluster are required (equipment). In addition, a number of processing steps is required to turn the raw data into a quantified image volume: many parameters affect the image quality in subtle ways. The processing „history“ needs to be documented and reproducible (good scientific practice) (Quality Assurance).
Figure:
This rack contains the PET lab’s compute clusters: the ElvesCluster consists of eight nodes with two processors each. The configuration uses Gigabit Ethernet for communications. It has succeeded the older HeinzelCluster for 3D PET reconstruction. The HeinzelCluster is made up of 28 CPUs, four in each of the seven nodes (with Myrinet for fast networking).
Electronics Development
The members of the electronics laboratory advice and support the neurological research groups in imaging and electro-physiological techniques. They develop and assemble measuring components and laboratory systems. Their skills cover
- mechanics
- low level signal measurements (electro-physiological potentials)
- Aanalogue and digital techniques, controlling engineering (especially temperature controller)
- CAD techniques
- high frequency technology
The company Medres Medres arose from this group. It develops hardware for use in medical research.
VINCI –
“Volume Imaging in Neurological Research, Co-Registration and ROIs included”:
A software package for image visualization and analysis of PET- and MRI-images developed at the MPI
Quality Assurance:
The VHIST-Project
Technologies for medical research
Medres, a spin-off of the MPI for Neurological Research