臺灣博碩士論文加值系統

　　在現今的放射診療學上，核磁共振造影（MRI）為目前應用最廣的造影術之一，核磁共振造影可以提供多方向的的解剖切面造影，使得疾病的診斷更加準確，以提高治療的正面效果。當一病人被安排做MRI影像後，對人體某一器官切面部位，會產生一系列的多頻譜影像。如果把一系列的切面影像疊在一起組合而成後，便形成人體一個立體的三維結構。醫師就藉由這個立體結構得到一些醫學診斷的資訊，如器官的形狀、位置與體積大小。
　　要從上述的這一系列的切面影像估算出三維的結構時，必須先從影像中把器官分割出來，然後才來作三維影像重構的工作。一般傳統器官分割的方法主要是以人工描繪的方式，但此法耗時且費力。因此本論文即在研究由多頻譜MRI影像中，把腹部的主要器官（如肝臟、脾臟、腎臟）分割出來。
　　在本研究中，我們先嘗試使用數種多頻譜特徵轉換各自MRI影像做特徵擷取，這些轉換分別是主成份分析、特徵影像濾波法、目標點影像法與比值濾波器。比較這四種特徵擷取的結果中，我們採用特徵影像濾波法（Eigenimage filtering）當成主要的特徵擷取。關於特徵影像濾波法連算中所須的兩個輸入因子－－想得到的特徵向量（Desired feature vector）與不想得到的特徵向量（Undesired feature vector），本論文且發展其相對應之方法自動選取方法。一組多頻譜影像經由特徵影像濾波法後將可得到一張具有影像增強效果的特徵影像；亦即在此張特徵影像中，想得到的器官會顯現較亮的灰階值，不想得到的器官會顯現較暗的灰階值，如此一來將有助於後續的器官分割。我們將此張特徵影像利用數學形態學（Morphology）與人體醫學的解剖知識配合著臨界值分割法去做影像分割，再從二位元影像透過輪廓擷取以得到器官的邊界。得到器官的輸廓後再輔以輸廓修改技術以得到更加真實的器官邊界。

　　In modern radiotherapy treatment planning, Magnetic Resonance Imaging (MRI) is one of the most widely used radiographic techniques. MRI provides three descriptions (coronal imaging, sagittal imaging and axial imaging ) of internal structures to help doctors make treatment of diseases accurately. After a patient undergoes a MRI scan, a sequence of multispectral image slices is generated. Each slice represents one reoss-section image of tjhe three-dimensional human body. Reconstruction this sequence of 2D images forms a 3D struchure. Doctors will obtain valuable reference in disease diagnosis from the 3D structure, such as contours, location and volumes of organs.
　Segmenting organs from MRI images is the first step of the reconstruction of 3D structure. The straditional approach to segment organs calls for manual object outlining by operators. This approach is not only time-consuming but also labor arduous. The goal of our work is to segment the abdominal organs--liver, spleen and kidney from MRI abdominal images.
　 In this paper, four different multispectral feature transformations-Parinciple component analysis (PCA), Eigenimage filter, Target point image and Ratio filter have been implemented and evaluated for feature extraction of MRI images. Among them, Eigenimage filter shows its effectiveness so as to be chosen as our feature transformation. The eigenimage filter needs two imput factors- desired feature vector and undesired feature vector. We alsoi propose a method to select these two feature vectors automatically. An image, called eigenimage, is obtained from four different spectral MRI images of the same cross-section of human body and enhanced by eigenimage filtering. In this eigenimage, the gray levels of the desired organare brighter and which of the undesired orugan are darker. This property is useful for further segmentation. The eigenimage is processed by morphological operation with the help of anatomic knowledge and thresholding technique to generate a segmented image. From the segmented image, rough contours of organs are obtained and then modified by smoothing and protrusion eliminating to make the contours more accurate. From the experimental results, it shows that our proposed system is effective and efficient .