Contour Detection for the Breast Tumor in Ultrasonic Images Using Watershed Segmentation

• Main Authors: Xun-Yao Lin, 林勳耀
• Other Authors: Yu-Lung Lo
• Format: Others
• Language: en_US
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/4wc3r9

碩士 === 朝陽科技大學 === 資訊管理系碩士班 === 92 === Automatic contour finding for breast tumor in medical ultrasound (US) im-aging may assist physicians without experience in making a correct diagnosis. The purpose of this thesis combined the advantages of texture analysis tech-niques and watershed segmentation to extract precise contour of breast tumor from ultrasound images. Watershed transformation is a very reliable, automatic and unsupervised segmentation technique. However, the ultrasonic images al-ways include lots of noise and irregular texture. It usually generates over-segmentation regions after watershed transformation. Thus this study uses the textural information in the ultrasonic images to solve the problem of over-segmentation. The experiment results show that the proposal can get simi-lar contours and regions-of-interest (ROI), which compares with an ultrasonic breast tumor contour manually sketched by an experienced physician. Automatic contouring segmentation can save much more time and get reliable results than that of manual sketch in computer-aided diagnosis (CAD) application.