以膠囊內視鏡影像的色彩與紋理特徵應用於潰瘍之偵測

The Application of Color and Texture Feature in Wireless Capsule Endoscopic Image for the Detection of Ulcer

胡祐莊(Yu-Chuang Hu)

指導教授 ： 蘇振隆

中原大學/工學院/生物醫學工程研究所/碩士(2010年)

https://doi.org/10.6840/cycu201000979

摘要

膠囊內視鏡比起傳統內視鏡更容易觀察到小腸的不正常現象，例如阻塞、出血與潰瘍，但是膠囊內視鏡最大缺點，在於一位病人診斷影像高達五萬六千幅，一位熟練的專業醫師，需要耗費2∼3小時進行診斷，造成醫師診斷上很大的負擔，因此研發出膠囊內視鏡影像的自動辨識系統，以減少醫師進行診察所需觀看的影像數量，藉此改善醫師在診斷過程花費大量時間。
本研究發展膠囊內視鏡潰瘍影像的自動辨識系統，使用HSIC2四種彩色空間轉換，來突顯影像潰瘍上的區域，灰階直方圖分析、共生矩陣特徵，來做為潰瘍腸道病理病徵的紋理特徵參數。先利用t test進行分析，可找出35個具有鑑別力的特徵參數，其中HSIC2共同參數為強度、平均值、變異數、標準差、偏態係數、峰態係數；H非共同參數部分為能量、同質性；S非共同參數部分為能量、對比度；I非共同參數部分為能量、同質性、對比度、熵；C2非共同參數部分為能量、同質性、對比度，導入支持向量機進行訓練與分析，訓練與分析所使用潰瘍影像520張，正常影像3300張。並藉由簡易使用者介面操作，即可讀取出1位病人可疑潰瘍影像，進行觀看。
本系統於實際病例測試中，分批處理與擷取完整病人資料平均花費8分鐘；依照影片的編排次序讀取AVI 檔案與執行潰瘍的判斷法則，大約需要150分鐘；而其判斷法則的正確率為82.87％、敏感度為94.94％、有效性為81.20％，kappa值為0.48。
本研究已建立一套自動化潰瘍偵測系統，讓醫師除了傳統臨床上以肉眼判讀膠囊內視鏡影像來診斷潰瘍之外，提供醫師一個準確且可信賴的偵測系統，以增加臨床診斷時的參考依據，進而提高判讀膠囊內視鏡影像時的效率。

關鍵字

潰瘍 ； 色彩空間轉換 ； 支持向量機 ； 膠囊內視鏡

並列摘要

Capsule endoscopy compared with traditional colonoscopy is more easily observed irregularities in the small intestine, such as large yellow-green abnormal areas, bleeding and ulcer, the most disadvantage of wireless capsule endoscopy was that physicians will take 2-3 hours to diagnose 56,000 images for one patient so that it caused physicians heavy burden. Therefore, developing an automatic image recognition system of the capsule endoscope will reduce the physicians needed to view images of the number of checkup so that it can help physicians to save a lot of diagnostic time.
In this study, the development of capsule endoscopy ulcer automatic image recognition system, the HSIC2 four kinds of color space conversion were used to highlight area on the image ulcers, gray histogram analysis and co-occurrence matrix feature so that those could be as texture parameters of the pathological symptoms of intestinal ulcers. First, doing an independent t-test analysis, the 35 parameters with distinguishable features were identified into support vector machines to conduct training and analysis. Among of those of parameters, HSIC2 of Common parameters were intensity, mean, variance, standard, skew and kurtosis; Non-common parameters H were energy and homogeneity；Non-common parameters S were energy and contrast；Non-common parameters I were energy, entropy, contrast and homogeneity；Non-common parameters C2 were entropy, contrast and homogeneity. The support vector machine was introduced to do training and analysis which used 520 images of ulcers and 3300 the normal image. And it could be read from a suspicious ulcer patients by simple user interface.
A real case was used to test this system, the batch treating with picking up the complete patient information spent 8 minutes equally; however, it probably needs 150 minutes to read the AVI file according to the movie arrangement order and carry out ulcer's judgment principle. The result was that ulcer's judgment principle's accuracy was 82.87%, sensitivities was 94.94%, effectiveness was 81.20% and kappa value was 0.48.
This study has established a set of automated ulcer detection system. This system could offer physicians on clinical diagnose ulcer besides the tradition by the naked eye interpret capsule endoscope phantom, also provide physicians a accurate and trustworthy the detection system. Therefore, this system could be a clinical diagnosis reference to physicians, and then increase efficiency of interpret time of capsule endoscope phantom.

並列關鍵字

Color space ； Capsule endoscopy ； Support Vector Machine ； Ulcers

參考文獻

[4] 呂建鋒，”膠囊內視鏡發展現況”，光連：光電產業與技術情報， pp. 29-31，2007年

[8] 廖榮聖，”無線膠囊內視鏡影像資訊系統之設計”，中原大學醫學工程研究所碩士論文，中壢，2003年

[10] 張鳳玲，”一個可偵測多種異常膠囊內視鏡影像之整合辨識系統”，中原大學電子系碩士論文，中壢，2005年

[5] Neil I. Goldfarb, Amy Phillips, Mitchell Conn, Blair S. Lewis, David B. "Economic and Health Outcomes of Capsule Endoscopy: Opportunities for Improved Management of the Diagnostic Process for Obscure Gastrointestinal Bleeding", Disease Management , Vol. 5, No. 3, pp. 123-135, September 2002.

[9] Kodogiannis, V. S. "Computer-Aided Diagnosis in Clinical Endoscopy Using Neuro-Fuzzy Systems" , Fuzzy Systems, 2004. Proceedings. 2004 IEEE International Conference on, Vol. 3, pp. 1425-1429, July 2004.