電腦輔助診斷系統應用於胰臟腫瘤辨識之研發

The Development of Computer-Aided Diagnosis System for Pancreatic Tumor Recognition

陳冠文(Guan-Wen Chen)

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(2015年)

https://doi.org/10.6840/CYCU.2015.00128

摘要

胰臟為人體消化道中之重要器官，致死率極高的胰臟癌不易早期偵測。此外，因胰臟炎而引發胰臟囊腫的病患，未來也較容易罹患胰臟癌。不同胰臟腫瘤之致死率、嚴重性、癒後效果皆不盡相同，進而會影響臨床醫師進行診斷與治療的效果。故本研究之目的為研發電腦輔助診斷系統於胰臟腫瘤辦識應用。
首先以中值濾波器與小波轉換法對原始CT影像進行影像前處理及增強；再利用GVF snake圈選胰臟腫瘤輪廓並進行紋理參數與特徵參數分析。而將分析結果先利用獨立T-test進行篩選後，找出具有鑑別能力的特徵參數並進行支持向量機的分類。且在對照系統分類結果與組織病理檢查之結果後，可對系統整體的診斷能力來進行測試與評估。研究中使用有施打顯影劑之CT影像68張(6張正常組織影像、10張惡性腫瘤影像、6張良性腫瘤影像、13張易癌變囊腫影像、33張發炎性偽囊腫影像)；未施打顯影之CT影像68張(6張正常組織影像、12張惡性腫瘤影像、3張良性腫瘤影像、16張易癌變囊腫影像、31張發炎性偽囊腫影像) 來進行系統之研發與評估。
由實驗結果顯示：1.有施打顯影劑之影像部分，正常組織影像與腫瘤影像之(第一階段)分類，效能為Sensitivity=1，Specificity=1，Accuracy=1，Kappa=1；實質腫瘤影像與囊狀腫瘤影像之(第二階段)分類，效能為 Sensitivity=0.87，Specificity=1，Accuracy=0.96，Kappa=0.91；實質腫瘤影像再細分為良性腫瘤影像與惡性腫瘤影像之(第三階段)分類，效能為Sensitivity=1，Specificity=1，Accuracy=1，Kappa=1；囊狀腫瘤影像再細分為易癌變囊腫影像與發炎性偽囊腫影像之(第三階段)分類，效能為Sensitivity=1，Specificity=0.94，Accuracy=0.95，Kappa=0.89。2.未施打顯影劑之影像部分，腫瘤影像與正常組織影像之(第一階段)分類，效能為Sensitivity=1，Specificity=1，Accuracy=1，Kappa=1；實質腫瘤影像與囊狀腫瘤影像之(第二階段)分類，效能為Sensitivity=0.71，Specificity=1，Accuracy=0.93，Kappa=0.79；實質腫瘤影像再細分為良性腫瘤影像與惡性腫瘤影像之(第三階段)分類，效能為Sensitivity=1，Specificity=1，Accuracy=1，Kappa=1；囊狀腫瘤影像再細分為易癌變囊腫影像與發炎性偽囊腫影像之(第三階段)分類，效能為Sensitivity=1，Specificity=0.94，Accuracy=0.95，Kappa=0.89。此外在有施打顯影劑的影像，以l_Average、c_Average、g_Entropy、c_Entropy、Lesion_Entropy、Area及Lesion_Mean，這7個參數適合作為不同胰臟腫瘤辨識之依據。而在未施打顯影劑的影像，以l_Average、g_Entropy、c_Entropy及Area，這4個參數也適合作為不同胰臟腫瘤分類之判讀，這些能協助醫師進行診斷之參考與提供參考意見。
本研究已初步研發出可增強CT影像應用於胰臟腫瘤辨識度，並且具有特徵參數分析之電腦輔助診斷系統，可有效降低病患遭誤診與進行侵入性檢查的機率。目前系統對有無顯影劑之影像的診斷效能相近，尚需要更多的影像資料進行評估，以確定病患施打顯影劑的必要性。

關鍵字

支持向量機 ； 紋理特徵分析 ； 斷層掃描影像 ； 胰臟腫瘤 ； GVF snake輪廓圈選

並列摘要

The pancreas is the important organ in the digestive tract of human body. The pancreatic cancer has an extreme mortality, because it is hard to be detected in the early stage for physicians. Moreover, patients with the pancreatitis cysts easily suffer from the pancreatic cancer in the near future. The mortality, severity, and prognosis of different pancreatic tumors are not similar. However, the pathological differences of them usually lead to a dissimilar diagnosis and/or treatment for patients. The purpose of this study is to develop the computer-aided diagnosis (CAD) system for pancreatic tumor to recognize application.
First of all, the original CT image was preprocessed and enhanced by using median filter and wavelet transform. Secondly, GVF snake circle pancreatic tumors contours to analyze texture parameters and feature parameters for this image. After evaluating the results by the independent T-test, the effective features were selected and severed as inputs in the support vector machines (SVM). The diagnostic efficiency of this CAD system was evaluated after comparing the classified results of CT images with the pathological results of patients. Totally, 68 CT images (6 normal pancreas, 10 benign tumor, 6 malignant tumor, 13 pancreatic cystic neoplasm, 33 pancreatitis pseudo cyst) and 68 CT images (6 normal pancreas, 12 benign tumor, 3 malignant tumor, 16 pancreatic cystic neoplasm, 31 pancreatitis pseudo cyst) with/without injected contrast media were used to develop and evaluate this CAD system, respectively.
The performances of system show as following: 1. For CT images with injected contrast medium (enhanced), the system can distinguish between the tissue of normal and tumor (the first stage classification) with sensitivity=1, specificity=1, accuracy=0.873, and kappa=0.839; and the tumor of parenchymal and cystic (the second stage classification) with sensitivity=0.87, specificity=1, accuracy=0.96, and kappa=0.91; and parenchymal tumor subdivided into the tumor of benign and malignant (the third stage classification) with sensitivity=1, specificity=1, accuracy=1, kappa=1; and cystic tumor subdivided into the cystic neoplasm of pancreatic and Pancreatitis Pseudo-cyst (the third stage classification) with sensitivity=1, specificity=0.94, accuracy=0.95, kappa=0.89, respectively. 2. For regular CT images, the system can distinguish between the tissue of normal and tumor (the first stage classification) with sensitivity=1, specificity=1, accuracy=1, and kappa=1; and the tumor of parenchymal and cystic (the second stage classification) with sensitivity=0.71, specificity=1, accuracy=0.93, and kappa=0.79; and parenchymal tumor subdivided into the tumor of benign and malignant (the third stage classification) with sensitivity=1, specificity=1, accuracy=1, and kappa=1; and cystic tumor subdivided into the cystic neoplasm of pancreatic and Pancreatitis Pseudo-cyst (the third stage classification) with sensitivity=1, specificity=1, accuracy=1, and kappa=1, respectively. Moreover, for enhanced CT images, seven parameters (l_Average, c_Average, g_Entropy, c_Entropy, Lesion_Entropy, Area, and Lesion_Mean) suitable as the basis for recognition of pancreatic cancer. While for regular CT images, four parameters (l_Average, g_Entropy, c_Entropy, and Area) are also suitable as different interpretation of pancreatic tumor classification, which provide physicians as diagnosis reference.
This study has already developed a computer aided diagnosis system for detecting pancreatic cancer in the early stage. We hope this system could effectively decrease the probability of misdiagnosis and/or making an invasive diagnosis for patients. So far, there is no significant difference for the performance of system for regular and enhanced CT images. However, we still need more data to evaluate and confirm the necessity of injecting medical contrast medium into patients during the diagnosis of pancreatic tumors.

並列關鍵字

GVF Snake Contour ； Pancreatic tumors ； Support vector machines ； Texture feature analysis ； CT images

參考文獻

[6] 何家奭、蔡明宏，“慢性胰臟炎合併膽道阻塞－病例報告及文獻回顧”，童綜合醫學雜誌，2011, 5(2): 78-83。

[7] 廖漢文，“急性胰臟炎－臨床與影像診斷的深入探討”，當代醫學，2011, 455: 668-674。

[10] 張明志，“化療雖無法緩解，但能延長存活期及改善生活品質－胰臟癌”，健康世界，2008, 267: 51-55。

[11] 吳沅樺、林炳文、蘇五洲、陳海雯，“組織切片證實為胰臟癌的放射治療成果”，放射治療與腫瘤學，2004, 11(2): 81-89。

[16] 田郁文、李伯皇，“胰臟癌”，臺灣醫學，2006, 10(4): 493-499。