單光子放射電腦斷層掃描(SPECT)藉由放射性核種所釋放出的光子進行影像採集，本研究使用GATE (Geant4 Application for Tomographic Emission)模擬核子醫學影像系統，建構Micro-SPECT系統。設計上先行制定系統放大率、針孔位置、針孔數量、針孔型態、孔徑開放角參數、閃爍晶石參數建構、視域範圍設計、熱桿假體三區圓柱直徑大小及桿長參數制定，及使用核種活度及能量，通過GATE蒙地卡羅方法(Monte Carlo method) 取得冠狀軸和橫軸與螺旋橫軸三種假體軸面投影影像，後續建立H系統矩陣透過GATE粗略格點掃描取得點射源在不同三維位置的投影影像，並取得個別射源二維高斯參數化成像特性，再由距離權重高斯內插法建立完整的影像系統矩陣，搭配GATE模擬三種假體軸面投影影像，影像重建使用序列子集期望值最大化演算法(Ordered Subset Expectation Maximization, OSEM)結合圖形處理器(Graphics Processing Unit) CUDA(Compute Unified Device Architecture)架構，可將單一指令送交多個執行緒同時進行處理，具平行化優勢可大幅降低運算時間，最後比較三種軸面三維重建影像活度分佈優劣呈現。;The single photon emission computed tomography (SPECT) acquires the photons emitted by radionuclides to form the images of radioactivity distribution. This research employs GATE (Geant4 Application for Tomographic Emission), a Monte Carlo simulation tool for nuclear medicine imaging systems, to construct a micro-SPECT system. The system design starts from setting the magnification, the number of pinholes, pinhole locations, pinhole opening geometry, scintillator dimensions, the field of view (FOV), the hot-rod diameters and lengths of the resolution phantom, and the radioactivity and energy of the radionuclide. The coronal and transaxial projections of the hot-rod phantom are then generated for circular and helical trajectories via GATE modeling. The imaging system matrix required for image reconstruction is generated in three steps. Firstly, a radioactive point source is scanned on a coarse 3D grid to obtain the point response functions (PRFs) in the FOV of the imaging system through GATE modeling. Secondly, each PRF is fitted by a 2D Gaussian function with six parameters. A complete imaging system matrix on a finer grid is then generated by distance-weighted Gaussian interpolation method. The image reconstruction utilizes the complete system imaging matrix to reconstruct three projection image sets of the resolution phantom, including circular coronal projections, circular transaxial projections and helical transaxial projections. The reconstruction algorithm is the Ordered Subset Expectation Maximization (OSEM) implemented using the Graphics Processing Unit (GPU) in the Compute Unified Device Architecture (CUDA), which enables parallel computing to greatly reduce the computation time. The slices of the reconstructed 3D images are compared in their radioactivity distributions and image quality.
