抗核抗體免疫螢光影像特徵萃取與分類之研究

Extraction and Classification of the Features of Anti-nuclear Antibody Immunofluorescence Images

謝宗佑(Tsung-Yu Hsieh) ； 張世旭(Shih-Hsu Chang) ； 謝祖怡(Tsu-Yi Hsieh)

《科學與工程技術期刊》 6卷2期 (2010/06) Pp. 51-59

https://doi.org/10.7117/JSET.201006.0051

摘要

在免疫疾病診斷上，抗核抗體(anti-nuclear antibodies, ANA)的分類可幫助病情的診斷，目前常用的檢驗方式是以間接免疫螢光法(indirect immunofluorescence）觀察螢光影像的抗核抗體的螢光形態。傳統上醫師以人眼觀察螢光影像進行分類，此作法傷眼及費時，因此本論文提出以電腦輔助對抗核抗體影像進行分類的方法。首先使用影像處理的技術先萃取影像特徵，再以特徵將影像分爲均質型(homogeneous)、周邊型(peripheral)、粗點型(coarse speckled)、與離散點型(discrete speckled)共四類型。特徵抽取法是從影像中找出細胞區域，從各細胞區域分別萃取出7 種特徵，特徵爲拉普拉斯運算值（水平與垂直的2 次微分、斜對角的2 次微分）的平均值與標準差、曲率的平均值與標準差、及細胞內灰階值的標準差。當各細胞區域的特徵向量取出後，找出具代表性的特徵向量代表該影像的特徵。輸入測試影像計算該特徵與各類訓練影像特徵集合的距離，以最小距離分類法對特徵分類。實驗結果證明所提出方法的四類型的分類正確率至少爲94%，系統整體分類正確率達98.75%。證明抗核抗體影像可以電腦輔助進行分類，所提出的分類方法簡單快速易於軟體實作。

關鍵字

抗核抗體 ； 螢光影像 ； 特徵萃取 ； 分類 ； 電腦輔助系統

並列摘要

Identification of anti-nuclear antibodies (ANA) is an important step in the diagnosis of autoimmune diseases. The most commonly used procedure is indirect immuno-fuorescence (IIF), in which the physician expends considerable time in reading the image patterns visually. To ameliorate this problem, a computer-aided classification method is proposed in this report. This system (1) uses image pre-processing techniques to obtain the regions of the nucleus, (2) extracts the feature information from the cell regions, (3) extracts the representative feature vectors from the features of the cell regions, (4) calculates the distance between the representative feature vectors in the test image and the training feature vectors, and (5) classifies the test image by a minimum distance classifier. In the experimental results, four patterns of anti-nuclear antibodies were used to evaluate the proposed method. These results indicated that the correct classification rate is above 94%, the average correct rate being 98.75%. This high classification rate shows that the proposed system is workable. Furthermore, this method is fast and easily implemented by software in a computer-aided system.

並列關鍵字

anti-nuclear antibodies ANA ； immuno-fluorescence images ； extraction of features ； classification ； computer-aided system