ANA免疫螢光顯影影像分類之電腦輔助系統建構

Development of Computer-Aided System for Classifying ANA Immunofluorescence Images

目前免疫風濕科最常用來檢驗抗核抗體（antinuclear autoantibodies, ANA）類別的方法，乃是使用人類上 皮細胞癌（Human epithelioma type 2, HEp-2）之免疫螢光顯影細胞影像分析技術，因為如果給予 100 種不同 的抗核抗體，免疫螢光顯影的技術能辨識出超過 30 種不同的細胞核和細胞質樣式，所以利用此技術可偵測或辨識出是否存在有不同的自身抗體。臨床上辨識這些樣式，都是經由醫生手動檢查顯微鏡下的載玻片，再圈選出目標細胞核加以辨識，這個方法需要具有高度專業的技術人員來操作且耗時。本研究利用影像處理與資料探勘技術來分析抗核抗體免疫螢光顯影影像，並建構一個電腦輔助診斷系統，先採用Canny 邊界檢測方法將細胞分割開來，再使用共生矩陣所計算出來的紋理特徵，找出特徵關係，並輸入到類神經網路分類器上，利用合適之紋理特徵值加以訓練，最後將影像予以分類。實驗結果顯示，本研究 在細胞的辨識率上，錯誤率只有 15.62%，優於人工判斷的錯誤率 23.6%，Kappa 值帄均為 0.7455 > 0.4，準確度高，因此本電腦輔助診斷系統可以提昇醫生臨床診斷之效益。

HEp-2 cells are used for the identification of antinuclear autoantibodies (ANA). They allow for recognition of over 30 different nuclear and cytoplasmic patterns, which are given by upwards of 100 different autoantibodies. Therefore, indirect immunofluorescence with HEp-2 cells presents the major screening method for detection of autoantibodies in systemic autoimmune disease. However, this method requires highly specialized technicians and is also time consuming. This research applies image processing and data mining techniques to analyze the immunofluorescence images of HEp-2 cell. These techniques include the Canny edge detection method for segment a cell, a co-occurrence matrix for texture analysis, and a neural network for classification. The goal of this research is to find out relevant features and construct out a computer-aided diagnosis system to distinguish immunofluorescence images. Experimental results show that the error rate of this approach is 15.62% better than the one of human expert 23.6%. The Kappa value is about 0.75 which is greater than 0.4. This implies that this approach can help clinical doctors to diagnose diseases more efficiently with the computer-aided diagnosis system.