在具雜訊的正常血液抹片中進行白血球分類計數

Abstract

對醫學檢驗單位而言，血液抹片人工鏡檢是不可廢除的一項重要檢驗依據，但是人工鏡檢確實是一個耗時耗力的過程，除了血液抹片本身製作上的優劣之外，還得要考慮到不同醫檢人員因為疲勞或者是標準不同而有不一樣的結果，因此，以數位影像分析的技術來協助這項工作的進行，將能夠減少人力的消耗，讓血液抹片鏡檢更有效率。 本論文所研究的對象為具有雜訊的正常血液抹片，雜訊主要來自於製作不良的血液抹片，例如抹片太薄、抹片太厚、染色不均，以及細胞破裂等等因素，在同一個人的血液抹片影像裡面，可能會因為這些原因而在後續的分析中產生不一致的結果，因此，針對這類雜訊進行處理有其必要性，如此一來實驗結果才能具有讓人信服的依據。 本論文的貢獻有三，第一，透過二值化以及區域成長法，能將上述所定義的雜訊排除，進而找出本論文所要尋找的白血球細胞核區域；第二，在找到白血球細胞核區域之後，將可以完成多顆白血球影像定位，以利後續分析；第三，利用距離轉換(distance transform)以及平均值移動演算法(mean shift)，可以找出白血球細胞核分葉特徵，配合形狀特徵以及紋理特徵將能夠使判斷更為精確。

For the medical examination unit, the artificial blood smear examination is an important test for abolition, but the process is indeed a time-consuming examination process. In addition to making blood smear on their own merits, but also have to take into account the different medical laboratory personnel due to fatigue or different standards then have different results. Therefore, to digital image analysis technology , assisting in this work, will be able to reduce human consumption, making microscopic examination of blood smears more efficient. The object of study in this paper as having normal blood smear noise, noise mainly from the production of bad blood smears, for example, thin smear, thick smear, stain unevenly, and cell rupture, among other factors, blood smears in the same individual images which may be because of these reasons and in the subsequent analysis produced inconsistent results, therefore, were necessary for the processing of such noise, this way in order to have convincing results basis. There are three contributions of this paper, first, through the binarization and region growing method, able to rule out noise as defined above, and then find out in this paper to find the white blood cell nucleus area; second, find the white blood cells in the nucleus zone, more satellites will be able to complete the positioning of white blood cell imaging, to facilitate subsequent analysis; Third, the use of distance transform (distance transform) and moving average algorithm (mean shift), you can find leaf characteristics of white blood cell nuclei, with the characteristic shape and texture features will be able to make more accurate judgments.