中文摘要

醫學超音波乳癌影像診斷為接受度最高的診斷工具，主因為對人體不會產生不良的副作用，並且可及時得到檢驗的結果，因此超音波掃描儀成為臨床上醫生初次診斷乳癌的利器。然而，由於超音波腫瘤影像含有大量的斑點、 雜訊及組織紋理，容易因為欠缺超音波掃描儀操作訓練的醫師產生誤判的情形，導致醫療資源的濫用。因此近年本計畫主持人與數位學者及醫師合作完成數個乳房超音波腫瘤電腦輔助診斷系統，得到了卓越的診斷結果，但是此一精確的診斷結果是依賴有經驗的醫師手動擷取腫瘤輪廓進而判斷其良惡性，所以在本計畫中，使用小波轉換紋路分析技術自動對乳房腫瘤的輪廓做近似，以期降低上述電腦輔助診斷系統的應用門檻，進而廣泛應用並造福更多婦女同胞。
在本計畫中，我們將使用小波轉換來萃取出超音波腫瘤影像的紋路特徵，產生一腫瘤紋路特徵影像，再使用分水嶺轉換(Watershed transform)逼近真正的腫瘤輪廓，本計畫研究方向是採利用紋路特性來切割腫瘤超音波影像，此種方式可避免超音波影像中過多之組織紋路及雜訊影響腫瘤輪廓近似的精確性，我們將會充分利用紋路特性切割腫瘤，使得找到的輪廓近似經驗豐富的醫師所認定的輪廓。

關鍵詞：影像切割、小波轉換、分水嶺轉換、乳房超音波、紋路分析、腫瘤輪廓近似

Abstract

Due to the ultrasonic examination would not cause any side effect upon patients’ body and ultrasonic scanner had the advantage of timeliness. The ultrasonic instruments grow into essential for hospitals and ultrasound became the most acceptable examination for patients. The ultrasonic image is also an efficient instrument of the clinical physicians to diagnose the breast tumors at an earlier stage. However, speckles, noise, and tissue-related textures always consist in the digital ultrasonic image, physicians without clinical experience always made a miscarriage of diagnoses. We have proposed a number of computer aided diagnosis system for breast ultrasound and given remarkable results. In these systems, the region of interesting in the ultrasonic image is manual sketched by the experienced physicians. In this project, we will adopt wavelet textural analysis techniques to find the contour of breast tumors in the ultrasonic images automatically. This work may assist physicians without experience in making a correct diagnosis.
Most traditional segmentation techniques for ultrasonic images do not perform well because the images contain speckles, noise, and tissue-related textures. We will use the wavelet coefficients as the textural features to generate textural images. The watershed transform is performed to find tumor contour in the textural image. Because of the ultrasonic image full of variety, we will try to find the most similar contour with the physician for the breast ultrasonic image.

Keywords: Image segmentation, Wavelet transform, Watershed transform, Breast ultrasound, Texture analysis, Tumor contour approximation