| 摘要: | 核子醫學影像技術的臨床前小動物試驗必須使用比人類器官影像重建更高空間解析度的影像擷取設備，如此才能使體積小的動物器官重建影像具備與人類器官重建影像同等的精確度。本論文以單光子放射顯微鏡系統(Single Photon Emission Microscope, SPEM)作為投影影像的擷取設備，此系統為單光子放射電腦斷層掃描系統(SPECT)的高空間解析度特化分支，所用設備包括七針孔式準直儀、摻鉈碘化銫晶體CsI(Tl)、高質量光影像縮倍管以及電子增益電荷耦合元件(EMCCD)。
而在電腦斷層掃描技術的領域中，為了求得高品質的重建物體影像，需要有精確且高解析度的影像系統矩陣，即物空間與像空間的轉換關係矩陣。而本研究為了建立更加精確的SPEM系統之影像系統矩陣，改良了本實驗室舊有之影像系統矩陣的建立方法，內容包含系統幾何參數之擬合方法以及成像模型的建立方法。
過去本實驗室為了獲取SPEM的旋轉載物平台其旋轉軸與EMCCD之空間關係的幾何參數，而進行了幾何校正實驗，以及為了建立成像模型，而進行了格點掃描實驗。幾何校正實驗設計了三點源假體置於旋轉平台上，而旋轉平台是位於位移平台上，以每次旋轉一個固定角度來進行一次投影影像的擷取；格點掃描實驗則將單點源假體亦置於旋轉平台之上，以每次平移位移平台的一個固定格點寬度距離來進行一次投影影像的擷取。從這些點源圖形的投影影像之中，能夠獲得原物體空間之點響應函數，並擬合簡化為二維高斯函數。二維高斯參數分作光通量、投影點x及y座標位置、投影點橢圓的主軸方向以及投影點橢圓之長軸與短軸等六項參數。
本研究將過去所進行的前述兩項實驗之投影點的x座標及y座標位置一同用於擬合幾何參數，以取得假體與旋轉平台的空間轉換關係、假體與位移平台的空間轉換關係、旋轉軸的空間描述以及針孔的空間位置等幾何參數。之後透過分析前述所求得的系統之幾何參數，與格點掃描實驗投影影像之高斯參數間的關係，來構想出具有非圓對稱點響應函數之成像模型，並以非線性最小局部擬合的方式來求得成像模型係數，建立出通量模型、寬度模型以及主軸角度模型等成像模型。最終利用算出之幾何參數以及建立之成像模型，來建立各種格點間距的影像系統矩陣，並利用迭代式演算法中的序列子集之期望值最大化演算法來進行物體影像的重建，而影像重建的採樣方法考慮到取樣完整性，而分作圓軌重建以及螺旋軌重建兩種，透過分析重建物體圖像來判斷影像系統矩陣的精確度。最終圓軌以及螺旋軌取樣的三維物體圖像重建，與過去本實驗室所採用的舊方法之重建結果相比，新方法的重建結果皆得到了更加合理與清晰的物體圖像。
;Nuclear medicine imaging of preclinical small animal studies needs to use imagers with higher spatial resolution so the reconstructed images of small animal organs can have the same accuracy as that of human organ reconstructions. In this thesis, the single-photon emission microscope (SPEM) is used as the imaging device. This system is a high-spatial-resolution specialized branch of the single-photon emission computed tomography (SPECT). The SPEM system consists of a 7-pinhole collimator, a thallium-doped cesium iodide crystal [CsI(Tl)], an electrostatic demagnifying tube (DM tube) and an electron-multiplying charge-coupling device (EMCCD).
In the field of computed tomography, in order to obtain high-quality reconstructed object images, an accurate and high-resolution imaging system matrix (H matrix) is required. So, in order to obtain a more accurate imaging system matrix of SPEM, this study has improved the old method of creating imaging system matrix comes from our laboratory. The content includes the fitting method of system geometric parameters, which describe the spatial relationship between the system rotation axis and the EMCCD, and the imaging models for generating the H matrix.
In the past, we conducted the geometric calibration experiment and grid-scan experiment in order to obtain the geometric parameters and to create the imaging model, respectively. In the geometric calibration experiment, a three-point source phantom is placed on a rotary stage, which is located on the three-dimensional translation stages, and each time after the phantom is rotated by a preset angle we capture a projection image. In the grid-scan experiment, a single-point source is placed on the rotary stage, and each time after the source is translated by a preset grid spacing we capture a projection image. From the projection images of these point sources, the point response functions (PRFs) of the original object space can be obtained and fitted by two-dimensional Gaussian functions. Each 2D Gaussian function has six parameters: the amplitude as the luminous flux, the x and y coordinates of the projection centroid on the detector plane, the principal angle of the elliptical contours of the 2D Gaussian and the variances along the major and minor axes of the elliptical contours.
In this study, the x and y coordinates of the projection points from the two previous experiments were used together to fit the geometric parameters. We can obtain the spatial relationship between the phantom and the rotary stage, the spatial relationship between the phantom and the translation stages, the spatial relationship between the rotation axis and the EMCCD, and the spatial positions of pinholes. After that, an imaging model with non-circularly-symmetric PRFs was conceived by analyzing the relationship between the geometric parameters of the system and the Gaussian parameters of the projection images from the grid-scan experiment. The imaging model coefficients were obtained by nonlinear least-squares fitting, including the flux model, width models along the major and minor axes, and principal angle model.
Finally, the calculated geometric parameters and the created imaging model were used to generate the image system matrices of various grid spacing. The iterative algorithm Ordered-Subset Expectation Maximization is used to reconstruct the object images. The sampling method of image acquisition takes the sampling completeness into account, and is divided into circular and helical trajectories. The accuracy of the imaging system matrix is judged by analyzing the reconstructed object images. In the final reconstruction of three-dimensional object images sampled by the circular and helical trajectories, compared with the reconstruction results of the old method used in our laboratory for creating H matrices, the improved method have obtained more reasonable and clearer reconstructed images. |
