摘要 無線膠囊內視鏡上市後,有助於小腸疾病的診斷。而膠囊內視鏡診斷的最大缺點,在於一位病人之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。本研究目的在發展一套利用影像處理及輔助醫師診斷的系統,以加速醫師診斷的效率,並協助醫師診斷腸道疾病。 本研究透過撰寫*.grml格式,事先擷取出完整病人資料,並針對小腸出血點的特徵、腸道阻塞的特徵及腸壁白點的特徵,來預先偵測腸道內可疑疾病的部份。所利用影像處理的方法,包括YIQ、AC1C2、HSI色彩空間的轉換、直方圖分析與區域成長,分析腸道病理病徵的特徵參數。並以最佳之接受者操作特徵曲線 (ROC, Receiver Operating Characteristics),分析符合其條件下最大的正確率,選取出最佳的面積閥值來減少FP 數目。並藉由使用者介面,提供影片播放、縮圖瀏覽、時間軸、預先處理病徵區、歸納及診斷記錄等,來輔助專業醫師診斷。 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36分鐘;依照影片的編排次序讀取AVI檔案與執行病徵的判斷法則,大約需要2個小時;而其判斷法則的正確率為0.85、敏感度為0.945、有效性為0.667及信賴度為0.843。其中FN AVI影片段落,經由實際的病例發現,可由下一段落AVI影片中偵測出該病徵,因此對於同一個病人來說並不會造成診斷上病徵的漏失。而本研究所設計之系統在與原系統之RAPIDTM軟體比較之後,使用預先找出病理特徵之實際效應,能使得專業醫師之負擔不再沈重,而且自動化分析所增加技術人員之操作程序,也不會複雜。 整體而言,本研究以影像分析來輔助醫師診斷,確實能加速醫師診斷疾病並輔助醫師更加了解患者的情況,進而增加醫療的品質。未來有三個努力方向:(1)增加臨床病例之收集與增加其他腸道病徵之參數,提高系統之敏感度;(2) 配合多部電腦分析以加快執行判斷法則的速度及特徵的細節;(3) 把判斷法則模組化,以適應不同系統。並利用DICOM及HL7標準,將醫師的診療結果與院內PACS、HIS系統結合,達成醫療影像傳輸標準化與醫師經驗的交流與成長。
Abstract Wireless capsule endoscopy made a contribution to the diagnosis of the small intestines disease after appearing on market. The most disadvantage of wireless capsule endoscopy was taking 2-3 hours to diagnose 56000 images for one patient. The goal of this study was to develop a set of utilizing image processing and assisting doctor diagnosis system in order to improve the efficiency of doctor's diagnosis and help doctor diagnosis disease of the small intestines. This study could catch the complete data of the patient by writing the format of *.grml in advance. The suspicious disease on the small intestines capsule endoscopy images were detected in advance by using properties of (1) the Suspected Blood Indicator (SBI), (2) the chyme blocked, and (3) white point of the small intestines. Image process methods which included the YIQ color model transforming, the AC1C2 color model transforming, the HIS color model transforming, histogram analyzing, and region growing were used to analyze the feature parameters of the small intestines disease. Furthermore, the best Receiver Operating Characteristics (ROC) curve was supported to analyze the area threshold of the maximum accuracy for reducing the number of false positive (FP). Also, displaying in Graphical User Interface (GUI) which provided the results of pretreatment, such as the video broadcasting, the feature pictures browsing, the time tracking, the pretreatment marking, and the record concluding, etc. were used to help doctor diagnosis disease of the small intestines. After the testing of real case, following results are obtained. (1) For time consuming: it cost 36 minutes on average to segment and catch the complete data for one patient, and took about 2 hours to read a sequence of the AVI films and carry out the judgment rules of the small intestines disease; (2) For the evaluation of judgment rules: the accuracy was 0.85, the sensitivity was 0.945, the specificity was 0.667, and the Kappa was 0.843, respectively. According to the AVI films of false negative (FN) in actual cases, the suspicious disease through the next AVI films could be detected. Therefore, detecting the suspicious disease couldn’t be lost for the same patient in our system. Finally, we found a great effect of the pretreatment from the comparison of RAPIDTM software and our system. Our system could reduce the diagnosis time and easily use for technician. The accomplishment of our system could provide image processing to assist doctor diagnosis and get promoted the healthcare quality. In the future, there are three aims in our studies. One is to increase the clinical cases collections and feature parameters of other small intestines diseases so that it could raise the sensitivity of system. The second is to use parallel computing to speed up judgment rules. The third is to make judgment rules become a module to adapt to different system. Moreover, the implementation of DICOM and HL7 standards into this system which make easy to combine with PACS and HIS could exchange the experience of physicians and patient record through network.