Neutrosophic set based clustering approach for segmenting abnormal regions in mammogram images

Abstract

Image segmentation is an important step in image processing application, especially for medical images. It is also a very important task in breast cancer detection. There are various stages for breast cancer detection. The first stage is extraction/segmentation of region of interest followed by detection and classification. In this paper, a new clustering procedure is proposed to extract/segment the region of interest/lesion in mammogram images using neutrosophic set (NS). A variety of image segmentation algorithms are in the literature, but accuracy is still a crucial problem. NS has an ability to handle indeterminant information thus reducing the uncertainty in the images. The image is initially converted to an NS domain, which is described using three membership sets: degree of truth (T), degree of indeterminacy (I), and degree of false (F). In our work, indeterminate degree is computed using a novel technique that uses Shannon entropy and standard deviation. Then, a neutrosophic similarity-based image is formed using neutrosophic similarity function, which is then clustered to detect lesion/tumor. In the clustering algorithm, three image features and also a second criterion function, which is an exponential entropy, are used. The proposed algorithm has been tested on different types of mammogram images along with a comparative study with existing methods both quantitatively and qualitatively. The experimental results demonstrate that the proposed method can segment the mammogram images more effectively and accurately.