Skip to main content

ACcurate Reconstruction In PET: Fully 3D PET reconstruction with compressed scatter system matrix

Final Report Summary - ACRIPET (Accurate reconstruction in PET: Fully 3D PET reconstruction with compressed scatter system matrix)

In the fight against cancer, Positron Emission Tomography (PET) can provide 3D images of metabolic processes (i.e. of cells in action) in the body and contributes to diagnosis and treatment strategy. Two-dimensional PET scanners (2D PET) have been replaced by three dimensional PET machines combined with a CT (3D PET/CT). This led to complementary anatomical information (from the CT) and increased sensitivity (less noise in the images). However, the increased sensitivity comes along with an increased detection of scattered radiations that impair accurate quantitative measurements necessary for diagnosis and patient monitoring.

PET images are usually reconstructed iteratively from the signal detected by the scanner using a system matrix that models the imaging process. In clinical 3D PET/CT, this huge system matrix (order of Terabytes) is highly simplified to make the problem tractable. Based on previous proof-of-principle studies in 2D PET by the applicant, the ACRIPET project proposes to develop a novel and accurate 3D reconstruction method based on an accurate (unlike approximate) system matrix obtained through Monte Carlo modelling of all patient and detector related effects interfering with the imaging process. The method uses compression schemes to reduce Monte Carlo noise and allow storage of the system matrix and parallelised on multi node computer architecture to make accurate 3D PET reconstruction clinically feasible. By using the most modern computer resources to fully exploit the potential of 3D PET/CT scanners, PET image accuracy should be significantly increased, contributing to earlier detection and more precise characterisation of the disease and of its evolution.

The first part of the project aimed at accelerating the Monte Carlo simulation tool GATE. Since matrix simulations require the simulation of a very high number of particles, this part was crucial. In order to accelerate the calculation of attenuation within the patient a fast Siddon-like particle tracking algorithm was implemented within GATE. A detailed analysis of the GATE code further revealed that a large part of the execution time was needed for the simulation of the detection system and the signal processing chain. It was therefore decided to completely replace these parts in the simulation. This was achieved by creating a multidimensional B-spline to store the detection probability of a photon before it enters the detection system. This detection probability depends directly on the photon state when entering the detection system and can be evaluated very fast. The values of the B-spline can be fixed by a single conventional GATE simulation that needs to be run only once for a given PET scanner, prior to the matrix simulation.

The approach was implemented for the PET scanner Biograph 16 of Siemens Medical, but could be adapted to other scanners due to the generality of the B-spline method. The next step involved the usage of these stored detection probabilities during reconstruction. It was decided to combine the Siddon-like algorithm on-the-fly and to evaluate then the aforementioned B-spline that was calculated before reconstruction. In this way a memory efficient fast calculation of the system matrix elements during reconstruction was possible. First images were reconstructed.

The research is relevant for position emission tomography imaging. Position emission tomography imaging is used in cancer diagnosis and patient follow-up during and after therapy. Furthermore, this imaging modality is used in neurology and cardiology. Hence, the presented project aimed at improving and quantifying the information that is obtained by a PET scan and thus leading to improved diagnosis and better therapy outcome prediction. Furthermore, parts of the results of the work can be applied to small animal scanners and therefore improve pre-clinical imaging that is used to develop pharmaceuticals.