全乳房自動超音波之腫瘤偵測

Abstract

由於全乳房自動超音波在掃描乳房時是全自動操作，能降低不同使用者在操作上造成的差異，而且操作者毋須在掃描時檢測腫瘤是否存在於超音波影像中，而是在掃描影像之後再檢查影像，可以有效地減少病人在檢查的時間，三維的全乳房自動超音波影像能準確地定位出腫瘤的位置以利於之後病人的追蹤，因此在近年來，全乳房自動超音波已經成為在檢測乳癌上的一種常用的工具。然而，三維全乳房自動超音波影像通常包含了數百張二維超音波影像，醫生診斷三維全乳房自動超音波影像是否有腫瘤存在會需要花費許多時間且容易因為檢查大量資料產生疲倦而造成誤判。為了能夠減少醫生在檢測全乳房自動超音波影像的時間，可以使用電腦輔助偵測系統幫助醫生篩選出在影像中疑似腫瘤的區域。然而，在大量資料中的全乳房自動超音波影像中實作出能夠有效偵測出疑似腫瘤區域的電腦輔助偵測系統是非常困難的，因此到目前為止，只有少數在全乳房自動超音波影像的電腦輔助偵測系統被實作並提出。在此論文中，將會介紹二種全乳房自動超音波影像的電腦輔助偵測系統。在第一篇研究中，使用模糊平均分群法(FCM)從全乳房自動超音波影像的三個正交視角分別偵測出可疑腫瘤區域，接著結合這三個正交視角所偵測的結果以得到更準確的偵測結果。在第二篇研究中，使用了海森分析(Hessian analysis)在全乳房自動超音波影像中偵測腫瘤。由於腫瘤在超音波影像中的灰階值通常都比周圍組織的灰階值暗，腫瘤可以被視為一種球體結構，此球體結構可以使用海森分析偵測出。這二個電腦輔助偵測系統的實驗結果均能夠在全乳房自動超音波影像對偵測腫瘤有著不錯的靈敏度。

Automated whole breast ultrasound (ABUS) has been a novel screening tool in recent years due to the operator-independent, time efficient, and reproducibility. However, a three-dimensional (3-D) ABUS image contains hundreds of two-dimensional (2-D) ultrasound (US) images and physicians should need a lot of time to diagnose hundreds of images in an ABUS image. It is time-consuming to review a 3-D ABUS image and the misdetection might occur due to physicians’ fatigues. In order to reduce the review time and the misdetection by physicians, computer-aided detection (CADe) systems have been proposed to assist physicians in interpretation of these images. However, only few studies for development of CADe systems on ABUS images have been reported because it is difficult to detect tumors from a large number of images. In the thesis, two CADe methods for ABUS images were proposed. In the first study, the tumor detection based on the fuzzy c-mean (FCM) technique was applied to thee orthogonal view respectively and then the detected results in the three views were combined to obtain the more accurate results. In the second study, the tumor detection based on Hessian analysis was proposed. Because lesions are usually darker than the surrounding tissue, the lesions could be regarded as the dark blob structures and could be detected by using the Hessian analysis. The results showed that the two CADe systems could provide the high sensitivity for the breast tumor detection on ABUS images.