所謂抗核抗體(ANA)就是血中的抗體對抗細胞核內的抗原，意即對抗自己細胞核內抗原的自體抗體。這些自體抗體的存在，與各種?同的免疫疾病息息相關，對於診斷?同的風濕性疾病具有非常重要的意義，ANA的檢查方法為一種間接免疫螢光法，此為?用一種培養細胞HEp-2的細胞株做為酵素基質，此細胞之細胞核大，?於觀看，是目前使用最廣的一種ANA檢驗方法，此方法迅速、簡單且敏感性高，幾乎取代過去所用?瘡細胞檢查法(LE cell)，所以?床上已成為診斷免疫風濕性疾病常?之檢查。關於近年來間接抗體螢光染色HEp-2細胞的研究主要都是著重於對這些細胞作分類的研究，因為要分別這些抗核抗體不同種類的樣式需要有受過專業的訓練或是有經驗的專家和醫生才能夠分辨的出來，而且這種專家或是醫生其實人數是少量的，所以目標是想發展一個能夠幫助醫生作診斷的系統，也就是能夠自動對這些抗核抗體先作切割，然後再對這些切割出來的抗核抗體作正確的分類的系統。本篇論文的切割方法是一個改良式的兩階層的分水嶺演算法，而這個方法總共對2305個抗核抗體細胞作實驗（包含456 diffuse patterns, 417 peripheral patterns, 719 coarse speckled patterns, 55 fine speckled patterns, 517discrete speckled patterns and 141 nucleolar patterns），而這些細胞是由44張間接抗體螢光染色影像上取得的，接下來再利用Learning Vector Quantization (LVQ) 和51個特徵對1036個切割出來的細胞做分類。 Rationale and Objectives: Indirect immunofluorescence (IIF) with HEp-2 cells has been used to detect antinuclear autoantibodies (ANA) for diagnosing systemic autoimmune diseases. An automatic inspection system for the ANA testing can be partitioned into HEp-2 cell detection, fluorescence pattern classification and computer aided diagnosis phases. The aim of this study is to develop an automatic segmentation scheme to sketch outlines of fluorescence cells for HEp-2 cell detection in the IIF images and fluorescence pattern classification.Materials and Methods: In the proposed a two-staged segmentation method, the similarity-based watershed algorithm with marker techniques was performed to obtain the contour of each fluorescence cell. This study evaluated 2305 autoantibody fluorescence patterns from 44 IIF images that can be divided into six pattern categories (including 456 diffuse patterns, 417 peripheral patterns, 719 coarse speckled patterns, 55 fine speckled patterns, 517discrete speckled patterns and 141 nucleolar patterns). And fluorescence pattern classification method utilized learning vector quantization (LVQ) and 51 features to classify. It total experiment on 1036 cells (782 training data and 254 testing data). Results: The sensitivity of the six patterns except for discrete speckled pattern was 86.8%, other patterns are among 94.7% to 100%. The total average sensitivity was 94.7%. In the classification simulation, the total average correct rate is 0.803. Conclusions: This study proposed an automatic segmentation method for detecting outlines of fluorescence cells in IIF images and then the proposed classification method is performed to identify the different fluorescence patterns. The segmentation and classification result is satisfying for clinical applications.
