Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/98122
標題: 乳癌治療決策預測:CK5/14/E-cadherin/Ki-67 三重免疫組織化學染色乳組織切片之腫瘤細胞計數
Prediction of Treatment Strategy for Breast Cancer: Cell Counting of CK5/14/E-Cadherin/Ki-67 Triple Immunohistochemical Staining on Breast Cancer Tissue Section
作者: 詹貿翔
Mao-Hsiang Chan
關鍵字: Ki-67 proliferation index
triple immunohistochemical staining
inflammatory cells
stromal cells
image processing
Ki-67 proliferation index
triple immunohistochemical staining
inflammatory cells
stromal cells
image processing
摘要: 乳癌是世界癌症發生率的第二位,女性發生率的第一位,也是我國女性發生率第一位之癌症。乳癌除了會侵犯乳房,更可能發生轉移的現象,不過乳癌的診斷的方式很多,其中病理組織更是判斷如何治療的重要一環。但是傳統的評估方法是由病理醫師人工評估至少1000個細胞,計算其陽性乳癌細胞比例,方能得到可信的Ki-67增殖指數,是相當耗費病理醫師的時間與精力,而且由於每個醫生的標準不同,因此會產生結果上的差異。但在發展自動化系統上也面臨了無法辨識發炎細胞、間質細胞和正常細胞的問題。本篇使用了三重染色的乳癌組織切片影像,來解決發展自動化系統所遇到的發炎細胞、間質細胞和正常細胞無法區分的問題。先用E-cadherin去標記乳癌細胞的部分,接著用CK5/14染色,將肌皮細胞標計出來,就可以去除掉非侵襲性癌的區域,最後再用Ki-67染色,看乳癌細胞的活躍程度。接著也提出了自動計數經Ki-67染色後的陽性與陰性乳癌細胞個數之系統,此系統使用灰階值117為門檻值去除沒被E-cadherin染色之區塊,保留乳癌細胞的區域,之後透過形態學、run-length和watershed等方法,找出細胞核的部分,再透過形狀找出肌皮細胞,並依照包覆比例去除掉非侵襲性乳癌的區塊,最後設定門檻值辨識細胞核的顏色,即可計算出Ki-67陽性與陰性乳癌細胞的比例。該系統的計數準確度是70.3%,但算出來的Ki-67的增值指數跟醫師的相比,平均誤差值只有6.5%,除了特殊情況的第六張影像,經E-cadherin染色後,顏色比較紅,而在平均G值不趨近於0的影像上,Kigo診斷決策的結果跟病理醫師是一致的。
Breast cancer ranks second on the worldly incidence of the cancer and first on the women's incidence of the cancer. It's also the top one for women's incidence of the cancer in our country. Breast cancer not only would invade the breast, but also may transfer. However, there are many ways to diagnose the breast cancer, and the important one of them is by justifying the biopsy. It's the traditional way that the doctor needs to evaluate at least one thousand cells artificially, and count the proportion of the positive breast cancer cells so that get the reliable Ki-67 Value-added index. That way spends much time and energy of the doctor, and there are difference results for the difference standards by different doctors. It also faces problems that unable to differentiating the inflammatory cells, stromal cells and normal cells on developing the automatic system. We use three kinds of the stained breast cancer tissue section images to solve the problems unable to differentiating the inflammatory cells, stromal cells and normal cells on developing the automatic system. First, use E-cadherin to mark the part of the breast cancer, and then use CK5/14 to dye which make myoepithelial cells marked so that remove the non-invasive cancer area. Last, dye with Ki-67 to observe the activity of the breast cancer cells. Then, a system for automatically counting the number of positive and negative breast cancer cells dyed by Ki-67 was proposed. The system adopts grayscale 117 as threshold to remove the area was not dyed by E-cadherin, to reserve the area of breast cancer cells. We found nucleus by mathematical morphology, run-length and watershed, then found myoepithelial cells through shapes. And remove the non-invasive breast cancer area according to the ratio of the coating. Finally, setting the threshold value to recognize the color of nucleus in order to calculate the ratio of Ki-67 positive and negative breast cancer cells. The counting accuracy of the system is 70.3%. The calculated Ki-67 proliferation index has only 6.5% average inaccuracy comparing with that of doctors. In addition to the exceptional case of the sixth image, after E-cadherin dying, the color became redder. On images with an average green channel that does not approach zero, the outcome of Kigo's diagnostic decision is the same as that of the pathologist.
URI: http://hdl.handle.net/11455/98122
文章公開時間: 2021-08-30
Appears in Collections:基因體暨生物資訊學研究所

文件中的檔案:

取得全文請前往華藝線上圖書館



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.