- 學位論文
電腦輔助診斷系統中特徵參數於臨床乳房超音波與乳房腫瘤X光影像之應用
The Complement Both Texture and Application of Computer-Aided Diagnosis System for Lesions in Clinical Breast Ultrasonographics and X-ray Mammographics
摘要
女性乳癌為台灣地區十大癌症發生率中位居第四位。國內醫療機構基本以超音波影像（Ultrasongraphics）與X光攝影（Mammographics）作為初期診斷設備，可能因醫師訓練程度不同及以目測判斷腫瘤影像而產生誤判。因此，藉由電腦輔助診斷整合系統擷取同一病灶之乳房超音波影像與乳房X光影像之特徵參數，所得適用值做為乳房影像診斷初步依據。 乳房超音波影像使用病案經證實之52例之臨床影像；乳房X光影像使用病案經證實之43例之臨床影像；其中同一病灶之兩種影像為36例。研究所運用為研究室先前開發之電腦輔助診斷整合系統，以GVF Snake圈選出腫塊邊界，並選定腫塊之特徵參數，進行統計分析(t-Test)分析，並以操作特性曲線（Receiver Operating Characteristic）以曲線下面積（AUC，Area Under Curve）評估特徵參數診斷效益。最後利用倒傳遞類神經網路分析，提高對良、惡性腫塊的系統鑑別度。 研究結果顯示，在十四項特徵參數中，g_Average及Lesion_Mean均為超音波影像及X光影像之重要特徵參數（P<0.05）。超音波中以Lesion_Mean及Lesion_StDev具較好之診斷效益，此外，ROI_Entropy、Lesion_Entropy、g_Average、c_Average之P值小於0.05，依此六參數測試結果顯示系統之準確度（Accuracy）= 0.85、敏感度（Sensitivity）= 0.75、有效性（Specificity）= 0.93及Kappa值= 0.69；X光影像中以l_Entropy、g_Entropy及c_Entropy具較好之診斷效益，此外，l_Average、g_Average、Area、Circularity、Lesion_Mean之P值小於0.05，依此八參數測試結果顯示系統之準確度= 0.77、敏感度= 0.75及有效性= 0.8、Kappa值= 0.53。相關係數部分，以惡性腫塊共19例，探討g_Average及Lesion_Mean之兩組共同特徵參數，所得結果呈現負相關。 研究發現利用同一病灶之乳房超音波影像及X光影像，分別所得之是用特徵參數，應用於良、惡性腫塊得到明顯區分，同時提供醫學影像臨床診斷之初步參考依據。此結果因應未來大量影像診斷時而需發展CAD系統亦有很大幫助。
並列摘要
Women's breast cancer occupies the fourth place for the location in the ten major cancers incidence of Taiwan. In Taiwan domestic medical organization initial stage diagnostic device of ultrasongraphics with mammographics. Because doctor train degree with judge tumor image produce to estimate perhaps. So, computer-aided diagnosis system cute of combining picks is perhaps because it last degree doctor with and with estimating not for tumor image and not producing not judging by accident, diagnose the preliminary basis for the breast image in suitable value of the income. The breast ultrasonic image uses the clinical images of 52 that the medical record is verified; the breast X-rays image uses the clinical images of 43 that the medical record is verified; among them the same focus two kinds’ images of 36. The research institute uses the computer that develops for the research room before to diagnose the system of combining auxiliary, enclose and elect the lump border with GVF snake. Select the characteristic parameter of the lump, carry on the statistical analysis (t-Test) Analyze, and in order to operate the characteristic curve of ROC (Receiver Operating Characteristic) and AUC (Area under Curve) with the curve assess the characteristic parameter and diagnose benefit. It is changing neural network analysis of transmitting. To utilize finally, improve to good, malignant system of lump distinguish degree. The result of study shows, in 14 characteristics parameter g_Average and Lesion_Mean (P<0.05) is very important of ultrasongraphics with mammographics. With Lesion in ultrasonic wave Lesion_Mean and Lesion_StDev has better diagnosis benefits, in addition, ROI_Entropy, Lesion_Entropy, g_Average and c_Average P value is smaller than 0.05, the accuracy of six parameter test result display systems that the accuracy is 0.85, sensitivity is 0.75, specificity is 0.93 and kappa value is 0.69; With in the image of the X-rays l_Entropy, g_Entropy and c_Entropy has better diagnosis benefits, in addition, l_Average, g_Average, Area, Circularity and Lesion_Mean P value of Mean is smaller than 0.05, the accuracy of eight parameter test result display systems that the accuracy is 0.77, sensitivity is 0.75, specificity is 0.8 and kappa value is 0.53. Discover that utilizes the breast ultrasonic images and X-ray images of the same cases, that differentiated the income is to use the characteristic parameter, apply to the good, malignant lump and is obviously distinguished, offer the preliminary reference basis of clinical diagnosis of the medical image at the same time. The future diagnosed that focus in three-dimensional image. So, the clinician technician deal with report time has not increased, it is unable to increase hands to help. In the face of this situation, use the computer aided diagnosis that can help the efficiency and benefit of clinical diagnosis effectively tentatively.