Breast cancer is the malignancy that develops from the tissues in breast. Among the types of cancers, ductal carcinomas are those which develop in ducts, while those developing from lobules are known as lobular carcinomas and also mucinous carcinomas. All of these can be diagnosed by taking a biopsy from that particular lump. For reducing the human limitations in observing capacity and also to arrive at the most accurate diagnosis, assistance of computers in diagnosing is used. In this system, an accurate procedure for computerized diagnosis of malignant cells in breast on histopathology images is proposed. In the earlier methods the diagnosis was done in a larger scale where the spread of the malignancy in a group of cell is identified but using the proposed method the features of each nucleus of the cell is extracted so here we can identify how much each cell is affected and the intensity of the cancer can be effectively diagnosed. The nucleus contains the DNA of the cell. The size, shape, and color of the nucleus of a malignant cell are different from that of normal one. Mostly, nucleus of a malignant cell is more dark and large when compared to a normal cell as it often contains too much DNA. In our system, the variation in the color of the nucleus is considered. For extracting and analyzing the color we cannot use RGB plane due to its characteristics like chrominance, luminance and non-uniform traits. So, here RGB is converted to YCbCr color model and is segmented using watershed algorithm and k-means segmentation algorithm. A multi resolution features are taken out from these segments and analyzed. The features that have a combination of wave atom and Zernike moment were used for study as it is more effective and the system achieved an accuracy of 97% when used with random forest classifier.
Volume 11 | 04-Special Issue
Pages: 255-266