高倍率顯微組織影像之細胞核特徵分析－－以甲狀腺組織為例

Cell Nuclei Image Feature Analysis System for Microscopic Image on High-magnification Scales–for Examples with Thyroid Pathologic Tissue

顯微鏡下的組織型態變化在臨床診斷上往往具有非常重要的特徵及意義，細胞核的影像則在高倍率下的顯微影像下才能詳細觀察，本研究針對高倍率顯微組織影像的細胞影像及型態特性，依據醫師經驗診斷法則，建立病理組織之高倍率顯微影像特徵分析系統。高倍率(1000X)的細胞型態往往與病變直接相關，尤其細胞核的特徵變化明顯代表癌症的發生。本研究以甲狀腺組織為例，針對以蘇木精-伊紅染色(H&Estain)的高倍率甲狀腺組織的細胞影像進行特徵量化，以適應性區域分割的影像處理技術將細胞核的影像由其他組織及背景分離出來，採用13 種影像及型態參數量化細胞核的特徵，運用統計分析將參數分佈加以歸類分析，利用交叉驗證法對於正常甲狀腺濾泡細胞核的判斷正確率為93.5%，而常見的乳突癌病變濾泡細胞核的辨識率則為87%，這樣的分析方法將可以適用於各類高倍率組織影像的細胞核特徵觀察與分析。

The morphological features and image features of tissue represent meaningful characteristics in clinical diagnoses and researches. The high-magnification(1000X) microscopic images tissue including cells are important materials for clinical observation for the screening of various cancer. Especially the details of nuclei represent the conspicuous pathological characteristics revealing the stages of evolution of cancer disease. Referring to the clinical empirical rules of physician, this study has developed an image-based cell nuclei classification system with the high-magnification microscopic image of cell nuclei of thyroid tissue with H&E stain. The adaptive region segmentation approach was applied to extract the image of nuclei from the background and other tissues. thirteen morphological and image features of nuclei were characterized and quantified. The statistical discriminant analysis method was then applied to classify the groups based on the distribution of features. From the results of our experiments, the accuracy of nuclei classification for normal follicular cells is 95.55% and the popular papillary carcinoma can be actually discriminated. We believe that this system has the feasibility to provide the information for assisting clinical diagnosis and study for thyroid disease in the nearly future.