偵側肺結核菌之影像系統的設計與研發

Abstract

目前傳統結核桿菌抗酸性痰液抹片，須在顯微鏡1000倍油鏡下，仔細察看每一視野是否有被染成紅色之桿菌，來斷定此痰液抹片其結核菌是否是陽性或其量為1+~4+，此方法最大缺點為很耗費眼力及必須很專心且很仔細依序察看每一個抹片有涵蓋之視野，也常因為傳染性病菌檢查而不願意參與此工作。工作必須很仔細且傷眼力、費時、有時不注意易有差錯，故為改良人為因素而發明自動快速偵測痰液抹片之結核桿菌方法以期能供臨床應用。本論文中使用電腦輔助診斷系統將利用電子式顯微鏡截取數位影像，經由影像處理的領域中，特徵擷取、影像輪廓偵測、影像分割和影像重建等數位影像分析技術，配合類神經網路(neural network)來分辨是否為結核桿菌，本論文的目的在於將桿菌之數量呈現出來並標出判斷為桿菌的位置，讓醫師可以更方便為病患作診斷及治療根據。本論文的主要貢獻有三，第一，解決目前以人工判斷結核菌數量的不便。第二，利用數位影像處理的方法，將圖上可疑目標物的特徵以數字表示，以用來判斷是否為結核菌。第三，完成一個偵測肺結核菌之影像系統，實驗結果顯示出本系統在判斷結核菌上有不錯的效果。

Currently, the phlegm smear examination of acid-fast bacilli (AFB) needs to be carefully examined under microscope and distinguished positive or negative. The biggest weakness of this method is the waste of eyesight taken on observing visual field of microscope carefully. A lot of people would not like to run the risk of contacting contagious germ to participate in the check work. The work must be taken carefully and harm human eyesight. It is also time-consuming and easy to lead mistakes. About these problems, design and investigation on the identification of tubercle bacilli image system is proposed in this thesis. The proposed system uses electronic microscope to capture digital images and apply feature extraction, image segment, image analysis and neural network to analyze tubercle bacilli. The purpose of the proposed system is to show the amount of tubercle bacilli and to find its position. The contributions of this thesis are summarized as follows: 1) the proposed system can take place of counting by human; 2) this thesis uses the image processing method to analyze tubercle bacilli; 3) the proposed system can detect tubercle bacilli automatically. The computer simulations have shown that the proposed system performs excellent.