3D彩色乳房超音波的腫瘤診斷

Abstract

由於乳房3-D彩色都卜勒 (Doppler)超音波成像技術可以成功地偵測血管資訊，所以近年來廣泛地利用血管生長分布來幫助醫生診斷腫瘤，研究也發現血管分布在腫瘤生長、惡化與轉移過程中上扮演著重要的角色。一般而言，惡性腫瘤在生長期時需要較多且複雜的血管以供給足夠的養分。而過去相關的研究只利用血管點數的多寡作為診斷的依據，然而更可以利用血管資訊的形態特徵來分析血管的生長情形。在此篇論文中，將著重於分析以腫瘤邊界向內及向外各一段距離所構成頻帶空間中，其內部血管的型態特徵來診斷腫瘤的良惡性。我們提出電腦輔助診斷系統將使用乳房3-D彩色都卜勒超音波成像技術，此種影像資料可被解碼成腫瘤的灰階影像與包含血管資訊的血管影像。首先，將灰階影像用level set方法切割出腫瘤輪廓並利用切出的腫瘤輪廓算出對應的三維距離圖；接著再利用三維細線化演算法取得血管骨幹。利用三維距離圖以及血管骨幹後，可以計算出距腫瘤輪廓特定範圍的帶狀區域內其中的血管型態特徵。最後再利用二元邏輯回歸方法結合血管型態特徵來判斷乳房腫瘤的良惡性。實驗總共分析119個乳房腫瘤病例，其中包含66個良性腫瘤和53個惡性腫瘤。根據實驗結果，發現我們所提出的方法區能區別腫瘤良惡性達到準確性84.03%、敏感性84.91%、專一性83.33%以及Az值0.9104。

Since the Doppler ultrasound (US) is successfully applied for detecting the blood flow, the studies of tumor vascularity have played important roles to diagnose diseases of breast recently. The tumor vascularity is critical for growth, invasion, and metastasis. In general, malignant tumors need more complex blood vessels to obtain sufficient nutrients for growing. In the past, researches about vascularity just count the number of vascular pixel or voxel to analyze the malignancy of tumor. However, the morphology characteristic can be employed to provide more important diagnosis information. In this paper, we demonstrate a computer-aided diagnostic (CAD) system for three-dimensional (3-D) power Doppler breast US image that can quantify vascular morphology in a region within the fixed distance inside and outside the tumor. At first, the tumor is segmented by a 3-D level set method and the 3-D distance map could be computed based on the segmented tumor contour. After the skeleton of blood vessels is extracted by using a 3-D thinning algorithm, the morphological features can be calculated for the vessels in the band at the fixed distance to the tumor contour based on the 3-D distance map. Finally, the extracted vascular features are used for the binary logistic regression model to classify the malignancy of the tumor. In our experiments, 119 lesions containing 66 benign tumors and 53 malignant tumors, are used to test the accuracy of our proposed computer-aided system. From the experimental results, we could find that the proposed method has better performance than the conventional method. Moreover, the proposed method could achieve a high performance with the accuracy, sensitivity, specificity and Az value being 84.03% (100/119), 84.91% (45/53), and 83.33% (55/66), 0.9104 respectively.