利用三維超音波影像與電腦輔助系統診斷乳癌腫瘤良惡性與惡性等級

Abstract

本研究在於開發出透過三維超音波影像對乳房腫瘤進行良惡性分類，以及對乳癌腫瘤進行分級的計算機輔助診斷系統。對女性而言，乳癌是世界上造成最多死亡的癌症。為了改善病人的存活率，早期的乳癌診斷是必須的。過去研究中，在檢查乳癌裡，超音波成像已認為是一種能輔助乳房X光攝影的成像工具。所以，電腦輔助診斷系統是能被開發來協助醫生，透過超音波影像來進行乳房腫瘤的診斷。建立電腦輔助診斷系統的過程中，量化特徵是耗時的。所以，使用較少特徵的電腦輔助診斷系統是較好的。本研究中，首先從病人身上收集乳房腫瘤的三維超音波影像。然後，使用型態、紋理以及橢球近似特徵來描述乳房腫瘤的特性。之後，使用稀疏線性判別分析來區別出一組最小限度的特徵。研究結果顯示，使用最小限度的特徵以及全部特徵的這兩個系統，在統計學上的精確度是沒有差別的。本研究也探討乳癌腫瘤等級的區別。乳癌等級是一個對患者預後的標準指標。若能夠在早期的時候評估出乳癌等級，可使得醫生在患者的診斷上更有選擇性，且能適當地確定治療方式。本研究開發出協助醫生對乳癌腫瘤分級的電腦輔助診斷系統。研究過程中，從病人身上收集三維超音波的乳癌腫瘤影像。使用型態、紋理、橢球近似與後方回音的特徵來量化腫瘤的特性。使用結合基因演算法的支持向量機，開發出計算機輔助診斷系統來分類乳癌腫瘤等級。此電腦輔助診斷系統的精確度可達85.14%。此結果顯示，開發出來的電腦輔助診斷系統能有效分類低等級與高等級的乳癌腫瘤。

This work aims to develop computer-aided diagnosis (CAD) systems for (1) breast cancer screening and (2) tumor grade classification in ultrasound (US) images. Breast cancer is a leading cause of death for women worldwide. Early diagnosis of breast cancer is essential for improving the patient survival rate. US imaging has been reported as a noninvasive and effective tool for breast cancer screening as a supplement to mammography. CAD systems are often developed for screening breast cancer in US images. When developing CAD systems, the process of quantifying features is time consuming. It is preferable to use only crucial features in the CAD systems. In this work, three-dimensional (3D) US images of breast tumor were obtained at hospital. Tumor features, including morphological, textural, and ellipsoid-fitting features, were quantified to describe the characteristics of the tumors. Sparse linear discriminant analysis was then applies for identifying a minimal set of features to be used in tumor classification. The results indicated the accuracies of the systems developed using the minimal set and full set were statistically equivalent. This work also studied tumor grade discrimination in US images. Breast cancer grade is a standard indicator of prognosis in patients. Noninvasive and early assessment of tumor grade allows doctors to be more selective in their diagnosis and to appropriately determine the treatment to the patients. This work developed a CAD system for discriminating the breast cancer grades. In the process, the 3D US images of breast tumor were obtained. Morphological, textural, ellipsoid-fitting, and posterior acoustic features were quantified to describe the characteristics of the tumors. A support vector machine with genetic algorithm was then proposed to develop the CAD system that classifies breast tumor grades. The CAD system attained a classification accuracy of 85.14%. It is demonstrated that the developed CAD system can effectively classify between low and high grades of breast tumors.