Optimum real-time reconstruction of Gamma events for high resolution Anger camera with the use of GPGPU

Abstract

Aim of the HICAM project is to build a new, compact Gamma-ray imager with a submillimeter spatial resolution, based on the Anger Camera principle. The system is composed of a detection module with frontend ASICs, an acquisition board and a host PC, where real-time data processing and image reconstruction is implemented. The detector is based on an array of 100 Silicon Drift Detectors (SDDs) of 1 cm 2 each in a 10 × 10 cm 2 format, coupled to a single scintillator crystal. Position of the interaction with the crystal and energy of the incident radiation are obtained through the use of a MLE algorithm, that optimally exploits the information obtained from the detectors. Moreover it is possible to modify the algorithm in order to determine the depth of interaction of the Gamma photon inside the crystal. The MLE algorithm, on the other hand, requires a large amount of calculations per event. In order to process the events in real-time we have implemented the MLE algorithm on a GPGPU, obtaining a processing rate of 150000 events/second, considering a FOV of 512 × 512 × 10 points (calculation of the z coordinate is performed). In the paper we discuss the derivation of the algorithm, its performance for what concerns spatial resolution and distortion, and the speed of its implementation on the GPGPU.

A machine learning method for fast and accurate characterization of depth-of-interaction gamma cameras

“…Second, the model-based approach is orders of magnitude more computationally efficient than nearest neighbor methods (Maas et al, 2009;van Dam et al, 2011b,a), requiring a single vector comparison per location of interaction, instead of thousands. Efficient implementations of the modelbased MAP estimation algorithm achieve millions of events per second on commodity hardware (Pedemonte et al (2009), https://sourceforge.net/projects/oree/).…”

“…Various numerical simulation techniques for forward modeling of gamma cameras have been described in the literature, including analytical approaches such as the simple solid-angle model based on L'Huilier's formula (Pedemonte et al, 2009;Li et al, 2010), and methods based on Monte Carlo integration of the light transport equations (Boschini and Fiorini, 1999). However, analytical models are too simplistic to capture the complexity of real-world gamma cameras; whereas Monte Carlo simulations are complex to set up and yet may not fully capture the real characteristics of the imaging devices.…”

“…A variety of algorithms can be utilized to maximize (7), such as simple exhaustive search or fast adaptive grid search methods (Pedemonte et al 2009).…”

“…therefore the estimation consists in first finding the location of interaction by maximizing (A.4) and then computing the energy in closed-form using (A.3). A variety of algorithms can be utilized to maximize (A.4), such as simple exhaustive search or more computationally-efficient adaptive grid search methods (Pedemonte et al, 2009).…”

Experimental Assessment of PCA and DT Classification for Streamlined Position Reconstruction in Anger Cameras

“…If the logarithm is implemented as a look-up table, this constraint is worsened of a factor 2. Some solutions have been proposed in literature to reduce the overall number of operations, such as splitting the MLE search in two rounds [46], but the number of operations per event would still be as large as 1.5×10 5 .…”