| Detection of high-energy photons in positron emission tomography (PET) is a basic, yet crucial stage to image reconstruction. The ability to precisely calculate the photon coordinates from the electric pulses invoked in the PMTs as the result of scintillation implies high detection resolution and allows better reconstruction accuracy.This work studies several approaches to photon detection using GEANT4 Developer Toolkit. These approaches are: Anger algorithms and Neural Network algorithms. The main purpose is to prove that it is possible to apply the Neural Network algorithms in real PET scanners and to demonstrate its advantage over existing detection algorithms.The obtained results prove high efficiency of photon detection schemes based on artificial neural networks and open several ways to improvement of modern emission tomography methods.This report consists of seven chapters:Chapter 1 mainly introduces the basic conception of PET and how to apply the neural network algorithm on PET scanner. Then we overview the previous research work and show the developing state of photon detection.Chapter 2 we introduce the problem of photon detection in PET and describe then physics process and math model. Then we describe the classical approach to photon detection in PET adopted in industry since then invention of the Anger camera in 1957. After it we introduce a new algorithm of photon detection that is artificial neural network.Chapter 3 mainly describes the basic principle of neural network. First we introduce the basic conception of it, including the basic element of neural network and three basic types of network. Then we introduce some important conceptions while designing neural network, such as training process etc. At last we introduce some applications based on neural network and explain why it can be used in photon detection in PET.Chapter 4 we perform a Monte Carlo simulation of high-energy (511KeV) gamma rays transport and interaction in a thick NaI crystal based on Geant4...
|
PDF Full Text Request