Abstract
242545
Introduction: The design for the dedicated human brain SPECT system we call AdaptiSPECT-C includes 24 modular gamma cameras that create a hemisphere about a subject’s head, with one additional larger camera being placed at the apex. Imaging is accomplished via innovative smart apertures that allow real-time selection of pinhole sizes and degrees of multiplexing for each camera. Acquisition is accomplished with custom back-end electronics that implement 81-channel sigma-delta modulation A/D conversion. Each camera is a node in a 10-Gigabit ethernet network. The intended applications for the system are pharmacokinetic studies for drug discovery as well as improved accuracy and workflow for clinical diagnostic studies.
Methods: In order to develop the instrument’s control software before all of the hardware is completed, predict the performance of the system, and explore the merits of proposed adaptation strategies, we have developed a full-physics GPU-based Monte-Carlo simulation package. This software produces list-mode data that stream from TCP/IP ports exactly as if they were the system cameras.
There are two components in the simulation – an emission simulation that starts with a (potentially time-varying) distribution of radioisotope activity in a digital human phantom such as XCAT and ends with gamma-ray depositing energy in scintillation crystals. The second component involves sampling light-sensor signals from camera simulations to model the 81-channel data per camera actually produced by the system. The physics simulation software called UAMonteCarlo was designed to run entirely on GPUs; it relies on sets of 3D model data that represent the instrument, with associated material properties assigned to each object. The objects in the simulation are allowed to undergo transformations, which allows us to model the dynamic adaptive capabilities of AdaptiSPECT-C. The software simulates absorption and scattering for photons with energies up to 100GeV.
The camera simulation models the transport of scintillation photons from gamma-ray interaction locations to light sensors, including reflections, refraction, attenuation, and quantum detection efficiency. Mean detector response functions are precomputed for all light sensors (up to 81) on a 1-mm three-dimensional grid spanning the scintillation crystal volume. For each light sensor, the mean number of primary electrons associated with a gamma-ray event is used as the mean of a Poisson distribution from which a sample is drawn. The complete set of light sensor data are gathered into a packet that is buffered and available for read out.
Results: UAMonteCarlo simulation software was validated against Geant4 using a suite of tests and all results closely matched those of Geant4. For simple source objects (point sources, spheres, etc.), our simulation software can model upwards of 350 million gamma-rays per second or around 10 mCi in real time on a single computer. Using XCAT models with attenuation, we can achieve approximately 1.5 mCi of real-time modeling.
In addition to the simulation software, we developed acquisition software that connects to the simulation through an ethernet cable (just like the real system) and manages the camera connections, requesting data, position estimation, and sending commands to modify pinhole configurations. We found that the acquisition computer could readily handle the expected data rates even for the highest sensitivity pinhole settings.
Conclusions: We have developed a real-time, dynamic model of the AdaptiSPECT-C imaging system. This real-time model has enabled us to develop the acquisition software and ensure that the position/energy estimation and adaptations rules can be implemented on a single computer and can process this information for the expected photon rates. In addition, this software has enabled us to develop and test adaptation rules which is impossible to perform prior to system completion and difficult to do even with a completed system.
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.