符合波士頓清腸指標之盲腸辨識系統基於自適應增強算法

Abstract

在此篇論文中，我們提出一個可以從各式各樣大腸鏡照片中，自動辨識任一照片是否拍攝到盲腸的系統，藉此減輕大腸鏡醫師檢視大量照片的負擔。 大腸直腸癌在國人近年來癌症的罹患人數及醫療支出費用都名列前茅，值得慶幸的是第0, 1 期的大腸癌經過治療後，五年存活率可達九成以上，但是早期大腸癌並無明顯症狀，必須透過定期做大腸鏡檢查來提早發現，便能有效地提高治療效果，並降低大腸癌致死的風險。 大腸鏡檢查的品質攸關著是否能確實早期發現大腸癌，因此除了病人定期到醫院篩檢，也需要確保大腸鏡檢查擁有一定的品質，經研究發現醫生若在每次檢查都能確實深入到盲腸(盲腸到達率高)，病人罹患大腸癌的機率相對較低，也就是說盲腸到達率是用來評估大腸鏡品質的重要指標。 目前醫院是以人工的方式，交換大腸鏡照片給不同醫師來評估大腸鏡品質，因此我們提出一套自動辨識盲腸的系統，幫助醫師檢視大量照片並計算盲腸到達率，此系統會先利用腸道與糞便的色彩差異評估大腸鏡照片清腸不潔的程度是否影響大腸鏡品質，再針對清腸較乾淨的照片判斷是否含有回盲瓣(Ileocecal Valve,ICV)、三叉紋路(Triradiate Fold)或闌尾口(Appendiceal Orifice)等盲腸的特徵，我們利用各種影像處理的演算法擷取這些盲腸特徵後，再利用機器學習的方式來辨識照片是否有拍攝到盲腸，最後我們在平均辨識準確率上達到94.0%，最高辨識準確率達到96.9%，未來便可利用這套系統判斷醫師是否有確實在大腸鏡檢查中進入盲腸，做為一個公正評估大腸鏡品質的第三方，同時減少人工檢視照片的負擔。

In this thesis, we proposed a system which can automatically recognize the cecum image from colonoscopy photos based on the variability of human intestinal. This system can assist doctors to check the colonoscopy photos and reduce the load on doctors. In recent years, the colorectal cancer is the top one cancer on incidence rate and medical expenses in Taiwan. Fortunately, early treatment of colorectal cancer in Tis and T1 can increase the survival rate of patient effectively. However, there is no symptom in the early stage of colorectal cancer. In order to detect the early stage of colorectal cancer, the colonoscopy examination regularly is very important. The colonoscopy quality is closely related to the detection of early cancer. There are some quality indicators for colonoscopy: Cecal Intubation Rate (CIR), Bowel Preparation (BP), Adenoma Detection Rate (ADR), and Withdrawal Time (WT). In this thesis, we focus on CIR and BP. In order to evaluate CIR, doctors need to view great amount of colonoscopy photos. Therefore we propose a cecum recognition system to help doctors to evaluate CIR automatically. The system will assess BP if so bad that we cannot get information and features in the image. Then, the system extracts features of cecum from the images with good BP by image processing, and we use machine learning algorithm to recognize cecum images. Our method achieves the average accuracy rate of 94.0% and the best accuracy rate of 96.9%.