UW IRL Research tools

• Research Tools : SimSET | FusionViewer | Multiviewer | Image registration | Computing

Simulation Tools

Our simulation effort is now entering its 17th year of support for development of our public domain simulation software package, SimSET, and applications of simulation to image quality issues in PET imaging - both human and animal systems. The simulation package handles both PET and SPECT imaging systems. Physical effects modeled include: photon transport in objects (e.g., models of humans and animals); scatter and attenuation in collimators and shields, energy deposition in detector systems; and modeling of coherent scatter, positron range, and annihilation photon non-collinearity. The SimSET documentation package is on the web (http://depts.washington.edu/~simset/html/simset_main.html ) and is provided to the academic community as public domain software (with over 166 users in more than 90 laboratories around the world).   An alternate approach is provided by our analytic wholebody PET simulator (ASIM), which   can accurately simulate the noise properties of sinograms produced by several commercial PET scanners. It does not simulate photon transport in objects to estimate scattered or random coincidences (e.g. SimSET) but instead estimates their effect on sinogram statistics. Attenuation, scatter, randoms, detector efficiencies, and isotope decay are used to calculate the Poisson random deviate for each sinogram element.   Independent realizations of noisy sinograms can be rapidly generated with this approach.   In addition, the SimSET package is now being used with ASIM, where SimSET provides the system characteristics while ASIM provides rapid simulations of many realizations. The synergistic combination of SimSET and ASIM provide us and our collaborators with a particularly powerful set of simulation tools.

The simulation efforts for small animal PET scanners has been critical to the development of the MiCES scanner described earlier in this report. Simulations studies have been conducted to determine the number of events needed for various detection and quantitation tasks. Such studies have also been used to determine the impact of various photon transport processes on final image quality, including positron range, collinearity, parallax, and detector scatter. These studies lead directly to the development of the OSEM_DB reconstruction algorithm for the MiCES scanner (described earlier in this report).

[top]

FusionViewer

FusionViewer is an open source medical image display package developed by Insightful and the Imaging Research Laboratory. It is designed to improve the physician's ability to interpret the results of combined positron emission tomography (PET) and computed tomography (CT) studies. This software is a display application for facilitating and improving visualization.

Multiviewer

Multiviewer is a image volume viewer written in the IDL language ( figure 17 ).  It runs on all platforms IDL currently runs on, including Windows, Linux, Compaq Alpha, and Mac OS X.  With the IDL virtual machine (available since version 6.0 of IDL), Multiviewer can be used without an IDL license.

Figure 17. Example of the multivewer image viewer and image registration control application that   provides general imaging viewing and processing for multiple images. The tool is also being used to view and process PET/CT image pairs from PET/CT combined scanners.

Features of Multiviewer include:

• Reading of volume image files in Interfile, MetaIO, AVS, and raw file formats, with specification of voxel dimensions
• Adjustment of level and width interactively through the colorbar
• Up to 6 independent image panels,
• Linked cursors and navigation
• Multiple resizing options
• Selection of color tables
• Image fusion via alpha blending, with adjustable alpha and color table selection for the fused image
• Image checkerboarding for assessment of image registration
• Image cropping, either rectangular or polygonal cylinder
• Linked elliptical, polygonal, or freeform ROIs with basic statistics
• Measurement tool
• Zoom tool for looking at corresponding parts of a single view across multiple image volumes
• Hooks for running external image registration routines

[top]

Image registration

Registration can be performed using a rigid body package (NEUROSTAT) written by Prof. Satoshi Minoshima in the Division of Nuclear Medicine or by a non-rigid body package developed in the IRL. Both implementations are based on mutual information algorithms. The rigid body tool can usually be run without operator intervention (the operator selects the two data sets and then runs a tool we have written that does the data export, registration, and import of the registered data back into the Advance database). The non-rigid body tool requires operator intervention to define the volumes to be registered, set templates around parts of the anatomy that are different (e.g., arms in/out when registering body CT to body PET), and other QA operations. These operator tasks are performed with tools we have written in IDL.

Figure 18. Registered CT (above) and FDG-PET (below) images aligned using non-rigid registration. Images were obtained from a patient with clinical T2N2a cancer of the left pyriform sinus. One of several ipsilateral FDG-avid zone III nodes is indicated by fiducials. Arrows show FDG-avid primary tumor extending superiorly into the supraglottic region, seen only as a small left-sided mucosal abnormality on CT. On the right, fused axial and coronal CT (grayscale) and FDG-PET (hot-metal scale) images of the same patient show tumor and FDG-avid zone alignment. These images are then used for treatment planning using intensity modulated radiation therapy (IMRT).

[top]

IRL Computing Cluster

For extended calculations taking more than a few hours, such as PET scanner simulations, we have a computing cluster of five Apple XServe machines each with dual 1 GHz G4 processors and 2 GB RAM. The servers run the Mac OS X Server operating system, based on Unix, and are connected on a private network via a Gigabit Ethernet switch to a PowerMac 1.25 GHz G4, which functions as a head node. The head node serves as the connection to the outside world and serves to the cluster a shared file system, which is stored on an Apple XServe RAID with 690 GB of disk capacity (expandable to 2.5 TB). The cluster uses the open-source Sun Grid Engine resource management system for job queuing and scheduling, load balancing, and job accounting. In addition, parallel programs written using the MPI (Message-Passing Interface) library specification can be scheduled on the cluster using either the MPICH or LAM environments.

Phillip Cheng and Ruth Schmitz setting up parallel computing cluster

[top]

<< To the previous research section on Clinical PET and PET/CT