自動化影像分析軟體開發應用於全血中循環腫瘤細胞之辨識與計數

Abstract

癌症末期是對生命造成威脅的，其癥結在於原發腫瘤(primary tumor)可透過其在血液中的循環腫瘤細胞(circulating tumor cell, CTC)進行轉移。而直至今日，血液中 CTC 的數目在轉移性的乳癌、直腸癌以及前列腺癌上，已由眾多研究經由 CellSearch 所開發的檢測系統上確立出其臨床價值，其中此檢測系統是目前唯一被美國食品藥品管理局 (Food and Drug Administration, FDA) 許可用於 CTC檢測並具備高度自動化的檢測儀器。然而這套系統卻因以 CTC 表面上皮特性 (epithelial expression) 為依據的分離技術，使得所分離出的 CTC 失去了代表整顆原發腫瘤的代表性，為此目前 CTC 純化技術透過非依表面特性的方式 (label-free method) 設計出可抓取所有表現性的 CTC，惟此舉使 CTC 計數更加費時費工。且 CTC 數目並不如腫瘤組織切片能提供治療方面等資訊給臨床醫師做參考，在此情況下，被認為是可替代腫瘤組織切片且相當具有潛力的 CTC 亞型研究被提出，然而 CTC 亞型辨識上如同 CTC 計數也是一項繁重的過程。 本論文開發一套可辨識 CTC 並計算 CTC 即其亞型數目的軟體，為了驗證此軟體確實可辨識並計數 CTC，其實驗是將人類肺癌細胞株 PC9 以及人類大腸癌細胞株 DLD1 加入在健康人血液樣本中，並經在純化等過程後，由顯微鏡拍成圖片，而後由軟體進行辨識。其結果顯示細胞回收率為 100% ± 9.3%，而自動辨識單顆 CTC 的準確率為98.4% 以及自動辨識團簇 CTC 的準確率為 72.7%。此外，人類乳癌細胞株 MCF7、 AU565 以及 MDA-MB-231 被選用來展示此軟體確實可用來辨識 CTC 亞型。

Late-stage cancer is still a life-threatening disease. The crux of which is circulating tumor cells (CTC) stemming from primary tumor metastasizing through the bloodstream. Until now enumeration of CTC has prognostic value for metastatic breast, colorectal and prostate cancers, established by numerous studies based on the highly automatic and only FDA-cleared CellSearch system. The system, however, lacks the representative of an integrated tumor since the technology isolates CTCs based on epithelial expression. The state-of-art technologies for CTC enrichment by label-free technique are designed to isolate all phenotypes of CTC. Unfortunately, robust enumeration of CTC is quite difficult in label-free approaches. And what’s more, CTC counting still could not provide actionable information for clinicians unlike information from tumor biopsy. In such case, studies on CTC subtype, which is considered as a promising substitutes for tumor biopsy, were proposed. Nonetheless, the process in the identification of CTC subtype is also cumbersome as in enumeration. In this thesis, a software is developed to identify potential CTC, and to calculate the number of CTCs and CTC subtypes. To validate the software, two cell lines, PC9 and DLD1, representing different characteristics spiked into whole blood from a healthy donor was imaged after enrichment. Results showed that the recovery rate was 100% ± 9.3% and the accuracy of automatic classification is 98.4% for CTCs and 72.7% for CTMs. Besides, breast cancer cell lines (MCF7, AU565, and MDA-MB-231) was used to demonstrate that the software is truly able to identify CTC subtypes.