| Summary: | 碩士 === 國立中正大學 === 資訊工程所 === 93 === Breast cancer is one of the most common cancers among women all over the world. The early detection of the cancer provides a better chance of proper treatment. In the last decade, ultrasound imaging plays an important role in the field of breast cancer diagnosis because of its convenience and non-invasive. However, the ultrasound probe does not fully cover the breast width, it is not a convenient screening tool when adjunct with screening mammography. In this paper, an image stitching technique is proposed to stitch multi-pass images into a full-view image. In addition, a fully automatic screening system for whole breast with absolutely no human intervention is proposed. In order to reduce the processing time, the subsampling method is firstly used to accomplish the whole screening work as soon as possible. Owing to US images always have strong speckle noise, some preprocessing techniques, such as anisotropic diffusion filter and stick operator , are applied to reduce speckle noise and enhance the edges in the US image. Then, a thresholding measure based on gray-level slicing is used to partition the US image into several regions, and the suspicious tumor regions are determined according to some pre-defined criteria. By examining the US test database consisted of 25 cases, almost all the tumors or cysts could successfully be detected when an average of two false positive for each case is produced. A free-response operative characteristics (FROC) curve is also used to evaluated the performance. The experimental results show that the proposal system gives quite satisfactory detection performance.
|
|---|
