3D 乳房彈性超音波之電腦腫瘤診斷

Abstract

乳癌為女性癌症中的主要的死因。然而，如果乳癌能早期被偵測到且給予適當的治療，則可降低乳癌的致死率。近年來許多研究顯示利用乳房彈性超音波做診斷，可獲得優於傳統 B-mode 超音波的診斷效果。彈性可表現出腫瘤的軟硬度，而腫瘤軟硬度的差異可用於診斷腫瘤。近來，三維彈性超音波不僅可取得傳統三維 B-mode 影像更可取得不同截切面的彈性影像。因此，此篇研究主要的目的是利用三維乳房彈性超音波進行電腦輔助診斷腫瘤。首先，由 B-mode 影像切割出的腫瘤區域擷取出紋理特徵、形狀特徵、以及建立最接近腫瘤的橢圓模型，取得腫瘤與此模型的異同點特徵，再從彈性影像中對應的腫瘤區域擷取彈性特徵。最後，再利用這些擷取出的特徵來診斷腫瘤的良惡性。在此實驗中採用了 159 個病理驗證過的腫瘤，包含 110 個良性病例以及 49 個惡性病例。經由實驗結果，結合紋理特徵、形狀特徵和彈性特徵會得到最佳的結果，可達到準確率 81% (129/159)，靈敏性 78% (38/49)，特異性 83% (91/110)，以及 ROC 曲線面積 0.8512。因此，使用三維乳房彈性超音波可以有效的診斷腫瘤良惡性。

Breast cancer is the leading cause of cancer death for women. The mortality rate of breast cancer can be greatly reduced if a proper treatment is adopted after an early detection. Recently, many studies have shown that adding elastography examination can improve the diagnostic performance comparing to using only conventional ultrasound. Elastography can estimate the tissue stiffness by calculating the tissue displacement under a certain force. The stiffness of benign and malignant tumors can be used to be features for classifying tumors. Therefore, this study proposed a computer-aided detection (CAD) system using 3-D B-mode ultrasound and elastographic breast images. The CAD system proposed in this study using morphology, texture, and elastography features extracted from the segmented B-mode tumor area. Combining these feature sets in a binary regression model generated the malignancy estimation model.The diagnostic performance was validated by using 110 benign and 49 malignant breast lesions. The performance of combinating gray-level co-occurrence matrix (GLCM), ranklet textures, shape and elastography features achieved an accuracy of 81% (129/159), a sensitivity of 76% (38/49), a specificity of 83% (80/110), and an Az value of 0.8512. Summarily, the combination of B-mode and elastography features is effective to the classification of breast tumor.