本論文提出一套以紋理特徵為基礎之自動化超音波乳房腫瘤影像切割系統,主要應用影像處理技術於超音波乳房影像腫瘤輪廓切割上,以輔助醫師確認腫瘤輪廓位置,可作為超音波乳房腫瘤電腦輔助診斷系統之輸入樣本,提高乳房腫瘤良惡性辨識率,降低乳癌致死率。本研究中利用改良式螞蟻演算法對抗超音波影像複雜紋理與雜訊,並結合相關係數為紋理特徵,以最適路徑搜尋概念找出影像 上像素點間關係,採用映射與區域分群對於像素位置作分類,以切割出超音波乳房影像上腫瘤內外部區域,並以型態學方法進行腫瘤輪廓之修正。本論文所提出的超音波乳房影像腫瘤切割系統從實驗結果顯示,切割結果可被接受比例達94 %,可有效達到自動化切割準確性提升的目標,進一步做為電腦輔助診斷之參考依據。
In this study, we propose an automatic texture based segmentation method for breast tumors in ultrasound images. Image processing techniques are used to segment the contour of breast tumors to assist doctors for further differential diagnosis. The correlation coefficients were selected as the main texture feature to represent the characteristics breast tumor. The modified colony algorithm with optimum searching stratagem was then used to find the relationship among pixels. The inverse mapping and region classification were used to classify the position of pixels. Finally, the morphologic operations are used to modify the contour of the tumor. Experimental results show that the overall accepted rate is up to 94 % both on benign and malignant tumors of the ultrasound images. The proposed method can achieve the purpose of enhancing the accuracy of automatic segmentation effectively to assist further diagnosis.