不同取像角度之鼻咽部醫療影像之校準

Nasopharyngeal Medical Image Registration under Different Image Acquisition Angle

詹寶珠(Pau-Choo Chung) ； 張傳育(Chuan-Yu Chang) ； 林秀貞(Hsiu-Chen Lin) ； 朱唯勤(W. C. Chu) ； 邱南津(Nan-Tsing Chiu)

《Journal of Medical and Biological Engineering》 20卷2期 (2000/06) Pp. 51-61

摘要

對亞洲人而言，鼻咽癌是發生率非常高的一種疾病，目前對於其特徵之描述不夠具體化，常導致在患病末期才會被診斷出病患已罹患這種疾病。因此，醫師常需要整合多模組影像，以提供彼此互補的資訊。影像校準是在整合多模組影像資訊時的首要步驟之一，而所要進行校準的兩組影像，必須是由相同角度，所取得的相同部位之影像。但由於病患在做不同模組之醫療影像造影時，有時會取得不同角度的影像。因此在本論文中我們將提出一套完整的方法，可將兩組不同角度的影像先行轉換，使之成為相同角度的影像；接著再以相同角度的CT、MR 與SPECT影像為對象進行影像校準的動作。在本論文中的影像轉換部份，主要藉由模擬真實影像之造影情況，針對所要轉換成的影像中的圖素，計算出所有不同角度影像上，具備影響力的圖素，以達到影像轉換的目的。另外，本論文中並設計了一個影像校準的架構，可對CT、MR 與SPECT影像，其中任意兩組影像，作自動化的影像校準。首先利用影像之輪廓特徵，求出兩影像大致的比例（scaling）及位移量（translation），作初步影像校準。接著利用LVV運算子，擷取出影像之峰谷分佈特徵，並結合相關性技術，求出精確的旋轉角度（rotation）及位移修正量，以達成影像校準的目的。在實驗中，我們使用了成大醫院所提供的十二組病人的多張CT、MR與SPECT影像資料，以及高雄榮總所提供的phantom所攝得之二組CT影像資料，來測試本系統。最後的實驗結果，證實本論文提出的影像轉換方法可以得到良好的轉換結果，其平均灰階差異為6%。而CT與MR影像的平均校準誤差介於1.31～3.17個像素點（pixel）距離。

關鍵字

影像對準 ； 多模組影像 ； 影像轉換

並列摘要

Nasopharyngeal Carcinoma (NPC) occurs in high frequencies among Asian populations. Different from other cancers, the clinical records and medical studies only offer obscured nature of NPC. Nonspecific symptoms of NPC often leads to its late diagnosis. And, withal, if only one single modality medical image set is given to doctors, they can hardly detect NPC area. As a result, integrating multi-modality medical images is necessary in NPC diagnosis. Image registration is one of crucial steps to get the integrated information of two image data sets. For registration able to be conducted on two image data sets, they must be acquired from the same position with the same tile angle. However, image acquisition from different modalities is not guaranteed to be conducted under the same tile angle.In this thesis, a novel approach to register two image data acquired from different modalities under different acquisition tile angles is proposed. In our approach, the two image sets are first transformed to two image sets with the same tile angle. Then, the feature-ridges of the two same-title-angle image sets are extracted, and correlation technique is applied on the ridges to register the two image data sets.In order to evaluate our system, some image data sets were conducted in the experiments. These data sets include the CT, MR, and SPET images acquired from 12 patients, provided by National Cheng Kung University Hospital, And some CT images acquired from phantom provided by Kaohsiung Veteran General Hospital, and are used for evaluating image transformation results. The results show that the system performs reasonably well.