乳房超音波影像之多解析度模糊理論腫瘤偵測

Abstract

乳癌是全世界女性最常見的癌症之一，早期發現對於乳癌的治療是非常重要的，目前在診斷方面有許多的技術，包括超音波、X-光...等等，其中以超音波對人體的傷害較小，所以超音波檢測對於乳癌的篩檢和診斷是一種常用的方法。近年來，由一些公司所開發的可攜式個人電腦之超音波成像系統能夠提供在整合系統環境上開發電腦輔助診斷或偵測的應用，在此論文中，使用由Terason t3000 公司所開發的全乳房超音波系統，使用者可進行全乳房掃瞄，並經由一時鐘式的區域儲存方式，使用者可快速取出任一部位的超音波影像進行診斷。另為協助醫師判讀，同時也提出了以模糊理論為概念對於全乳房超音波影像進行電腦輔助檢測是否有腫瘤存在。首先，由於連續超音波影像之間的差異性很不明顯，所以我們會先將數張的超音波影像疊合在一起後再做處理以節省檢測時間。接著，將疊合後的影像做多重解析度的取樣處理。由於超音波影像通常本身有雜訊，因此前處理部份使用區域模糊的方式來去除雜訊，並用S型濾波器來強化對比並提升邊界資訊。接著，全乳房超音波不同解析度的影像會分別套用模糊理論來篩檢是否有腫瘤存在。在用模糊理論做篩檢後，只有標示為腫瘤的區塊會再經由一些事先定義的腫瘤篩選條件來去除不符合腫瘤特徵的區塊。最後我們會分別針對不同解析度各自篩檢出的可疑區塊互相做比對來得到最後的檢測結果。另外，我們使用FROC curve 來評估我們系統的偵測效能。根據實驗結果，系統有91.8%的偵測敏感度而每測試案例產生1.67次的false positives。

Breast cancer is the most frequently diagnosed cancer in women all over the world. The early detection of the breast cancer provides a better chance of proper treatment. There are many technologies for diagnosis, such as ultrasound (US) and X-ray. In general, US is the efficient method used for breast cancer detection and diagnose presently and has less injury to human body. In recent years, the portable PC-based US imaging systems developed by some companies can provide an integrated computer environment for the computer-aided diagnosis and detection applications. In this paper, an automatic tumor detection system based on the fuzzy approach using the PC-based US system Terason t3000 (Terason Ultrasound, Burlington, MA, USA) is proposed. In order to easily retrieve the US images for any regions of the breast, a clock-based storing system is proposed to store the scanned US images. A computer-aided detection (CAD) system is also included to save the physicians’ time for a huge volume of scanned US images. The multi-resolution technique is also applied in the CAD system for detecting tumors with different sizes. First, because the differences between the successive US images are unobvious generally; several US images will be overlapped together for improving the image quality and reducing the detection time. Then, the overlapped images will be resampled to three kinds of image resolutions for improving the detection accuracy. Because of the noise in the US image, some preprocessing techniques, such as smoothing filter and sigmoid filter, are used to reduce noise and enhance the object boundary. Then, the fuzzy technique is applied to detect tumors in the images with different resolutions, respectively. After the detecting tumor in the multiple resolution images, the pre-defined criteria evaluation is applied to remove the unreasonable tumors in order to reduce the false positives. Finally, the relative relationships of suspicious regions in multiple images with different resolutions can be compared to each other for obtaining the final detection results. Moreover, the free-response operative characteristics (FROC) curve is used to evaluate the detection performance of the proposed system. According to the experimental results of 60 cases, the proposed system yields a 91.8% detection sensitivity at the 1.67 false positives per case.