以影像軟體分析乳癌Ki-67增生指數之正確性與臨床意義

Analysis of The Correctness and Clinical Significance of Ki-67 In Breast Cancer by Image Analysis Software

秦芷菁(Chih-Ching Chin)

指導教授 ： 洪志勳

義守大學/工學院/化學工程學系暨生物技術與化學工程研究所/博士(2023年)

摘要

背景：
Ki-67之數值常被用來做腫瘤惡性度分級的參考，但它的再現性不佳導致臨床應用受到侷限。而且，細胞之計數是一個非常耗費人力的工作。近年來，影像分析軟體逐漸被應用在免疫染色的分析上，但其分析結果是否有臨床意義則需要更多的實驗驗證。
方法：
在2018到2020期間，從高雄醫學大學中和紀念醫院的病理資料庫中，選取177個女性乳癌病患之臨床資料，且有Ki-67免疫染色之組織切片。請病理醫師重新閱片，並用掃片機將玻片轉為數位影像。首先選取30個樣本進行前導研究，同時以三種影像分析軟體ImageJ、 K-means和U-net來辯識Ki-67之陽性及陰性細胞計算出其增生指數(Proliferation index)，並以人工判讀為標準分別與之倆倆作差異性比較。最後，以最接近人工值的影像分析軟體分析全部樣本，並統計出臨床資料後，使用卡方檢定與羅吉斯迴歸統計進行分析比較。
結果：
根據前導試驗中，對30名患者使用三種影像軟體分析，發現U-net分析法獲得的 Ki-67 指數與病理學家以人工之計數最接近，相較於其他軟體誤差最小（平均誤差率 = 11.66）。因此，選擇U-net來分析大規模170 個有效樣本，發現高 Ki-67 指數相關於「發生腫瘤轉移」( x2 = 3.78，p = 0.05；勝算值 = 2.16，p = 0.05）和「雌激素受體陽性」（x2 = 6.49，p = 0.01；勝算值 = 0.38，p = 0.01）。
結論：
以U-net影像分析軟體所得到的Ki-67增生指數較ImageJ和K-means法準確，與人工判讀的數據相近。影像分析可有效率的得到具有臨床意義的Ki-67增生指數，其與雌激素受體陽性的比例與腫瘤轉移的程度有相關。

關鍵字

Ki-67增生指數 ； 影像分析 ； 乳癌 ； 免疫組織化學 ； 自動化細胞計數

並列摘要

Background：
Ki-67 analysis is frequently used for tumor grading, however, the application in clinical decision making is limited because of poor reproducibility. Furthermore, the counting process is laborious for pathologists. Recently, digital image analysis for immunohistochemistry was developed. The correlation between Ki-67 index obtained by image analysis and clinical data need to be verified.
Methods：
Between 2018 and 2020, we retrospectively identified 177 female patients with breast cancer with Ki-67 index and clinicopathologic data at the Kaohsiung Medical University Hospital. Ki-67 expression was scored by pathologist as the standard dataset. Slides were scanned and converted to virtual slides. Ki-67 index were counted using three image analysis software, ImageJ, K-means and U-net. First, the data from three software are compared with the manual counting dataset using pilot study (n = 30). In all patients, relationships between Ki-67 index obtained from the best software, and clinical data were evaluated using the Chi-square test and Logistic regression.
Results：
Based on the pilot study, Ki-67 index obtained using U-net image analysis was approximate to the pathologist’s scoring for 30 patients, had the smallest error than another tools (average error rate = 11.66). In the 170 effective samples, the high Ki-67 proliferation index group was associated with metastasis (x2 = 3.78, p = 0.05; odd ratio = 2.16, p = 0.05) and positivity of estrogen receptor (x2 = 6.49, p = 0.01; odd ratio = 0.38, p = 0.01).

Conclusions：
The Ki-67 proliferation index obtained using U-net image analysis had the smallest error compared to those scored by pathologists than ImageJ and K-means method. Digital image analysis can be effective for measuring Ki-67 values and they are associated with positivity of estrogen receptor and tumor metastasis.

並列關鍵字

Ki-67 proliferation index ； image analysis ； breast cancer ； immunohistochemistry ； automated cell counting

參考文獻

第一部分

[1] 福利部國民健康署-乳癌防治-檢自 https://www.hpa.gov.tw/Pages/Detail. aspx?nodeid=205&pid=1124 (Jan. 25, 2022)

[2] L.L. Pan, H.C. Yu, C.H. Lee, K.C. Hung, I.T. Tsai, C.K. Sun, “Impact of Staining Methods and Human Factors on Accuracy of Manual Reticulocyte Enumeration” Diagnostics (Basel), 2022;12(9).

[3] F.A. Dzulkifli, M.Y. Mashor, H. Jaafar, “An overview of recent counting methods for Ki67 IHC staining” J Biomed Clin Sci, 2019; (3)8.

[4] M. H. Jang, H. J. Kim, Y. R. Chung, Y. Lee, and S. Y. Park, “A Comparison of Ki-67 Counting Methods in Luminal Breast Cancer: The Average Method vs. the Hot Spot Method,” PLoS One, 2017;12(2): e0172031.