Mammogram Breast Cancer Detection using Fast Watershed Segmentation

Abstract

Image Processing represents the main research area within engineering and computing specialization. It is promptly rising technologies today, and its applications found in various aspects of biomedical fields specifically in cancer disease. Breast cancer is taken into account the fatal one in all cancer types in step with recent statistics everywhere the globe. It’s the foremost common cancer in women and also the second reason of cancer death between females. In this paper, we recommend implementing a fast segmentation algorithm employing a watershed transformation. This extends the partitioning of the dividing waterline by allowing the mixing of advance information about image objects and therefore the traditional watershed algorithm. Before the watershed transformation can begin, the algorithm needs a way of representing the test image in terms of the amount of change around any particular pixel. We apply the Sobel operator to each pixel in the grayscale representation of the original image. The tumors detected are circular or semicircular in shapes according to the shape, and the brightness of the tumor will be darker as we moved far away from its center. The complement for this prior information can be taken as a local minimum that required beginning the watershed algorithm. So each tumor image can be represented as a lake with minimum value represented by the center in the complement tumor image. After employing the method, the detection of tumor percentage becomes more reliable. Such result indicates that the new technique has improved the performance of our computer aided diagnosis system for mammographic breast cancer detection effectively. The algorithm was implemented in MATLAB and tested under Windows system. The method was tested over several images from MIAS (Mammogram Image Analysis Society, UK), that provides a standardized classification for mammographic studies. In this paper, we recommend implementing a quick segmentation algorithm employing a watershed transformation. This extends the partitioning of the dividing waterline by allowing the mixing of advance information about image objects and therefore the traditional watershed algorithm.