假像消除及影像切割應用於生物及醫學影像

Abstract

生物或醫學影像處理的計算機方法改進了診斷或研究工作的效能。在這篇論文裡，我們發展了一些計算機演算法，並且發展了解決問題的工具。這些問題包括:從CR影像裡的格狀假像(Grid Artifact)消除，電泳圖像(Gel Electrophoresis)的自動比對，斑點( ELISA Spot)影像的自動分析，及從血管造影影片(Cine Angiogram)中截取冠動脈樹(Coronary Arterial Tree)等。 數位影像更容易儲存及傳送，不過，他們也包含了假像(Artifact)，例如格狀假像和不均勻的照明造成的假像。這些假像是既有(Inheritance)的問題。格狀假像和網紋圖案是使用Grid所引起的。 CR的取像平板(Imaging Plate)在X射線曝光期間會使用Grid來除去不想要的散射(Scatter) 。當使用顯示器來顯示所取得的影像時，格狀假像或網紋圖案(Moiré)的假像就可能會出現。當影像被顯示在一台低解析度的電腦螢幕上時，將使得格狀假像或網紋圖案更加嚴重。因此當影像的判讀必須用普通的電腦螢幕來達成時，這將成為一個大問題。在這篇論文裡，我們詳細探討這些假像造成的原因。研究顯示這些假像的頻率可以從DICOM標籤(Tag)及GRID的規格等訊息直接計算獲得。因此這些假像也就可以從頻率領域來去除。因為被除去的頻率，與解剖結構無關，所以處理過的影像比未處理前更加清楚。此外不均勻照明也造成另一種假像，不均勻照明將造成在物體上的光線不均勻，這使得在影像處理時，做門檻(Threshold)運算會有問題。因此除去這些照明的不均勻變化是有必要的。影像上的不均勻照明變化，在頻率領域上，是由低頻部分所組成。因此我們設計了一個濾波器來消除這些變化。影像在除去格狀假像及不均勻照明後，將更容易切割(Segmentation)。此外我們也提出切割的方法來切割消除假像之後的影像，如電泳影像和斑點影像等。我們也顯示，對取像時受污染的影像做去除假像的前處理(Preprocessing)，有助於之後的影像處理工作。 消除CR影像的假像，使得影像更加清楚有助於醫療診斷。電泳圖像及斑點影像的自動比對分析，減少生物學家的苦工，增進他們的研究效能。冠狀動脈樹的萃取，提供了分析動脈疾病很有用訊息。

Computer methods for medical or biomedical images processing improve the performance of the diagnosis or research work. In this dissertation, we developed computer algorithms and implemented tools to solve specific problems. These problems include removing grid artifacts from Computed Radiograph (CR) images, comparing the lanes in Gel Electrophoresis images, analyzing the ELISA spots images, and extracting the coronary arterial tree from cine angiogram. Digital CR images are easier to store and transfer from one place to another. However, CR images contain grid artifacts and moiré pattern that are the inheritance problems due to the using of grids to remove scattering. In this dissertation, the causes of these artifacts are investigated in detail. We show that the frequencies of these artifacts are fixed and can be estimated from the DICOM tags and grid specification. The artifacts can then be removed in the frequency domain. Because the removed frequency does not relate to the anatomical structure, the resulting images are clearer than before. Variable illumination is another kind of artifacts. Variable illumination artifact occurs when the ELISA Spots images were taken. This artifact causes a problem that the intensity threshold cannot be applied. In this dissertation, we design a filter to eliminate such variation. A sequence of image processing techniques is than applied to segment the ELISA spots. A tool was implemented based on the algorithm. This work helps to save biologist efforts in analyzing the ELISA spots image. Grid artifacts also occur in Gel Electrophoresis image. Gel Electrophoresis is an important tool in biology research area. To identify the same lane pattern is the goal. In this dissertation, we remove the artifacts before segmentation process are applied. We then convert a lane into the position vector for comparison. The presented method could reach 97% accuracy and thus save research effort of the biologist. The last work we studied was to segment the coronary arterial tree from cine angiogram. The artifacts were the backgrounds such as ribs and lung texture. We proposed to eliminate the backgrounds in time direction. Matched filter and wavelet techniques were than applied to segment the arterial tree. GVF snake is than applied to calculate the width of the vessel. The segmented arterial tree is fairly complete and the calculated width is accurate.