Automatic Colon Cancer Detection and Classi?cation
Saima Rathore · 2015
In the past two decades, automatic colon cancer detection has become an active research area. Traditionally, colon cancer is diagnosed using microscopic analysis of pathological tissue imagery. However, the process is subjective and leads to considerable inter/intra observer variation in diagnosis. Therefore, reliable computer-aided colon cancer diagnostic systems are in high demand. In this thesis, a computer-aided colon cancer diagnostic (CAD) system has been proposed that comprises three main phases. In the ?rst phase, an unsupervised colon biopsy image segmentation technique, which is based on a few novel extensions in traditional object oriented texture analysis based seg mentation technique, has been developed. The second phase deals with classi?cation of colon image and gene based datasets into normal and malignant classes. For the colon biopsy image based datasets, two classi?cation techniques based on hybridization of var ious features have been proposed. In these techniques, some traditional features such as morphological and texture, variants of traditional features, and some novel features which have especially been designed to capture the variation between normal and malignant colon tissues have been used. Similarly, for the gene expression based dataset, a novel technique that utilizes various feature selection strategies for solving the challenging problem of larger dimensionality of gene based datasets, and a weighted majority voting based ensemble of various SVM classi?ers for performance improvement has been proposed. In the third phase of this work, the structural variation in the shape of lumen among var ious colon cancer grades has been quanti?ed in terms of a few novel structural features. These features are used for the classi?cation of malignant colon biopsy images into vari ous cancer grades. Performance of the proposed diagnostic system has been validated on various datasets, and superior qualitative and quantitative performance has been observed compared to previously reported methods of colon cancer detection.