以DCE-MRI影像之血管分析為基礎的乳癌化療效果評估

Abstract

在所有癌症之中，乳癌已經成為全世界女性最常罹患的癌症並為現代女性主要的死亡原因之一。而在乳癌的治療上，化療是一種被普遍且多方面使用的治療方法，不僅用在手術前以縮小腫瘤使手術順利施行 (前置性輔助治療) 和手術後的輔助治療，也被用來控制癌細胞的轉移，因此若能提早得知化療的效果將在乳癌治療上提供很大的幫助。而近年來的研究顯示，血管與腫瘤的生長、惡化與轉移皆習習相關，通常腫瘤需要更多的血管來提供足夠養分以生長與存活，因此目前腫瘤血管已在乳房病變的診斷以及化療治療成效評估的相關研究中被廣泛的使用。有鑑於核磁共振成像 (MRI) 與其他醫學成像技術相比能提供較佳的血管對比度影像，因此本篇論文採用核磁共振影像作為實驗影像，並從核磁共振影像中擷取出血管特徵來嘗試評估前置性輔助治療的成效和嘗試探討血管資訊是否能提早預測出最終的化療結果。本篇論文大致可分為三大部份，乳房的切割、乳房內血管的擷取與血管特徵的計算。首先會使用以模板為基礎的乳房切割演算法來將乳房部份從完整的胸部核磁共振影像中切割出來，同時為了避免皮膚組織被誤認為血管取出，在擷取乳房血管前會先將灰階亮度與血管類似的皮膚組織從切割出的乳房中排除。在本篇論文中所使用的是以 Hessian matrix 為基礎的血管擷取演算法來擷取血管，並對血管作細線化以取出血管的中心骨架進而分別計算左右兩邊乳房內血管的型態與曲度特徵，最後以乳房化療前與化療早期血管特徵的變化率作為化療治療效果的評估指標。在本篇論文之中，乳房內的血管特徵將用來測試是否能預測化療晚期即最終化療治療的結果，並與腫瘤大小的變化率結合做預測，最後再與僅使用腫瘤大小變化率的預測做診斷效能比較。實驗總共分析了31個乳癌病例，包含了14個化療治療無效與17個化療治療有效病例。使用包含有腫瘤的乳房該側血管特徵做預測的正確率、敏感性與專一性分別為77.42% (24/31)、82.35% (14/17)與71.43% (10/14)，使用腫瘤大小變化率做預測的正確率、敏感性與專一性分別為80.65% (25/31)、76.47% (13/17)與85.71% (12/14)，而結合含有腫瘤的乳房該側血管特徵與腫瘤大小變化率做預測的正確率、敏感性與專一性則分別達到了83.87% (26/31)、88.24% (15/17)與78.57% (11/14)。實驗結果顯示，腫瘤大小變化率提供了較佳的專一性，而血管特徵的變化率則提供較好的敏感性，至於結合血管特徵與腫瘤大小變化率的方法則可以提升並提供最佳的診斷效能，這表示血管特徵的確擁有評估化療成效與預測最終化療結果的潛力。

In recent years, breast cancer has become the most common female cancer in the world, and it is also one of main common cause of death in modern times. Chemotherapy is one of the common methods of treatment breast cancer. Not only used before surgery to shrink the tumor (neo-adjuvant therapy) and post-operative adjuvant therapy, it is also used to control the cancer metastasis. Therefore, the early prediction of chemotherapy response would provide a great help in the treatment evaluation of breast cancer. On the other hand, recent studies have shown that tumors usually need more vessels than normal cells to provide sufficient nutrients for growth and survival. It means that the growth of vessels is high correlation with the growth of tumors. In addition, the vessel information has been widely used in the diagnosis of breast lesions and therapy response evaluation of chemotherapy. Hence, the main purpose of this study is to extract the vessel features from the 3-D MRI to evaluate the response of neo-adjuvant chemotherapy and analyze whether the vessel information can predict the final treatment response or not. There are three main parts in this study, including breast segmentation, vessel extraction and vessel feature extraction. First, the template-base breast segmentation algorithm is used to segment the breast region from the 3-D MRI. In order to avoid the skin being mistaken extracted as vessel, the breast skin would be excluded before vessel extraction. The line filter based on Hessian matrix is used to extract the vessels inside the segmented breast region in this study. After the vessel extraction, the vessels skeletons are obtained from the vessels by thinning algorithm and then transformed into tree structure in order to quantify the vessels. Total 17 vessel features including 11 morphological and 6 tortuous features are used in this study. The change ratio of vessel features between pre-chemotherapy and early stage of chemo treatment is utilized as indicators to evaluate the treatment response. That is, the vessel features is used to predict the final treatment response (i.e. the response of late stage of chemo treatment). In addition, diagnostic performance of vessel features method are compared with the change ratio of tumor size method and combined the change ratio of vessel features and tumor size method. This experiment evaluates total 31 breast cancer cases, including 14 positive response and 17 negative response cases. The accuracy, sensitivity and specificity for the vessel features in breast harboring tumor are 77.42% (24/31), 82.35% (14/17) and 71.43% (10/14), respectively. The accuracy, sensitivity and specificity for the tumor size method are 80.65% (25/31), 76.47% (13/17) and 85.71% (12/14), respectively. The combined vessel features and tumor size can provide the best diagnostic performance with accuracy, sensitivity and specificity improved to 83.87% (26/31), 88.24% (15/ 17), and 78.57% (11/14), respectively. The experimental results show that the tumor size method could provide a better specificity and the vessel features method can provide a better sensitivity. Furthermore, the combined method could provide the best diagnostic performance. It means that the vessel features could improve the diagnostic performance and might provide the capability of predicting the final chemotherapy response.