咽喉胃酸逆流影像分析診斷系統之設計與研發

Abstract

咽喉胃酸逆流在近年來被認為是許多耳鼻喉疾病的重要原因，目前標準診斷方法是屬於一種侵入性的檢查，且需要住院24小時，不利於門診施行。另外由於病人的症狀本來就有其主觀性，並且也容易因為檢查者的觀點不同，而產生偏差，依照醫生的經驗，對於沒有經過專精訓練的醫師，準確率可能不到五成，因此在臨床應用上仍然有其問題需要解決。本研究的目的就在於發展一個可以處理喉部影像的分析軟體，可以自動分離出診斷上的重要區域，並且收集大量正常無症狀對照組與病患組進行各方面的分析比較，希望以電腦客觀的方法找出更精準的診斷方式，以發展電腦輔助診斷的技術。 本論文的主要貢獻有四，第一，利用可變形模板(deformable template)自動分割出診斷上較為重要的影像區塊，解決目前以人工取樣不利於大量樣本研究的缺點。第二，利用數位影像處理的技巧，將各個部位上的特徵以數字表示，以利於尋找正確的數據以診斷病症。第三，提出一個以單張輸入影像估測深度資訊的方法，並以其推論杓狀軟骨區的腫脹程度，作為一個診斷上的依據。第四，完成一個咽喉胃酸逆流診斷系統，實驗結果顯示本系統在診斷的正確度上能夠達到不錯的成效。

Laryngopharyngeal reflux is known as an important reason for a lot of disease in recent years. The conventional way to diagnose Laryngopharyngeal reflux is not physical examination, so it is not convenient in the clinic. And the features of disease are different to doctor’s subjectivity. The accuracy of diagnosis is less than fifty percent for a doctor with not enough experience. So the diagnosis of Laryngopharyngeal reflux still has some problems. The purpose of this thesis is to design a system which can analyze the image of fauces and separate the image into different regions. This thesis collects a lot of images and finds a good way to diagnose Laryngopharyngeal reflux with computer. The contributions of this thesis are mainly summarized as follows: 1) this thesis designs an algorithm to separate the input image into different regions by deformable template; 2) this thesis designs a way to transform the features of diagnosis into numerical values in order to find a correct way to diagnose; 3) this thesis designs a way to estimate the data of depth from a single input image which can be used in analyzing the turgescence of arytenoids; 4) this thesis completes a diagnosis system for Laryngopharyngeal reflux. The experiment results show the system is satisfactory.