電腦斷層影像形態學自動分割應用於肝腫瘤之放射治療計劃
Automatic Morphological Segmentation of CT Images for Radiotherapy Treatment Planning of Hepatic Tumor
中文摘要
肝癌是目前台灣民眾因癌症死亡的首位，三度空間順形放射治療在肝癌的治療上 已經是一個可以接受的治療選擇，其中有一部分的功勞是因為目前三度空間電腦 計劃系統在醫學影像上的處理能力愈來愈好。在三度空間順形放射治療系統，為 取得腫瘤的形象及體積大小，要先將腫瘤的輪廓在電腦斷層掃描影像一張一張先 描下來，不但費時間也容易造成錯誤。有許多的研究是利用不同的數學方法來切 割影像，可以自動描出腫瘤的外形，本研究是利用數學形態學的方式來切割電腦 斷層的影像，目的在於使電腦自動描繪腫瘤外形，減少人力的負擔，以及錯誤的 產生。首先，在進行三度空間順形放射治療之前，先取得肝癌病人腹部 5mm 厚 的電斷層影像，總共收集 242 張符合 DICOM 3.0 標準的電腦斷層影像資料來分 析，之後利用 Matlab 軟體影像處理工具將電腦斷層掃描影像讀取並存入電腦，
切割影像的方法有好幾個步驟，包括利用填色，平均濾波器，膨脹，侵蝕計算影 像的梯度，以及分水嶺轉換等，利用 Hausdorff 距離，Quality 值數以及平均 Dice 相似值數來評估分水嶺轉換的效能。在這個研究中顯示出平均分別是 10.455 mm，0.7478 以及 0.8523。根據上述結論：分水嶺轉換是可行而且在臨床上的應 用是可被接受。在肝癌形態分割使用腫瘤的輪廓描繪，無論如何，還是需要更多 的研究以及修改計算的程式來改良影像自動切割的品質及能力。
英文摘要
Hepatocellular carcinoma (HCC) is the leading cause of cancer related death in Taiwan. Three-dimensional conformal radiotherapy (3DCRT) is gaining acceptance as an option in the treatment for HCC, partly due to improvement in medical 3D imaging capabilities of current radiotherapy treatment planning system (RTP).
Tumor contour delineation is necessary for volume visualization and implementation of 3DCRT. Manual delineation of the tumor contour is time consuming and error prone. Numerous studies have been conducted in an effort to automate tumor delineation by using the different methods of image segmentation with varied success. In this study, CT image segmentation using mathematical morphology was studied with the aim of automating the delineation process and reducing operator workload and error. First, abdominal CT image in 5 mm slice were obtained from HCC patients before undergoing 3DCRT, a total of 242 CT images were collected for analysis. The DICOM 3.0 compliant images were then read using matlab image processing toolbox, the image segmentation method consisted of the several steps
including averaging filter, flood fill operation, calculation of the image gradient by image dilation and erosion, and watershed transform. The Hausdorff distance, quality index, and Dice similarity index were used for validation of watershed segmentation. The study showed the mean Hausdorff distance, mean quality index and mean Dice index were 10.455 mm, 0.7478, and 0.8523 respectively. In conclusion, morphological segmentation using the watershed transform for HCC tumor contour delineation is feasible and clinically acceptable, however, more study and refinement of the algorithm is needed to improve the quality and robustness of automatic image segmentation.
