Abstract
The histopathological analysis of tissue specimens is essential for the diagnosis and grading of colon cancer. But the procedures are subjective and lead to noteworthy intra/inter observer difference in analysis as it mainly depends on the graphical assessment of histopathologists. So, a dependable computer-aided method, which can automatically classify malignant and normal colon samples, is required but automating this method is challenging due to the presence of outliers. In this paper, we presented a new efficient method for detecting colon cancer from biopsy samples in the presence of outliers and the method consists of four important steps. Initially, the colon biopsy images are segmented through pillar k-means algorithm to form a set of redundant candidate region in which the clusters are formed. Then in the second stage from the clustered region the parameter called Lumen Circularity (LUC) is calculated and based on that a tree structure is generated where the samples are classified as normal or malignant. After classifying the samples the outliers among them are removed by calculating the Mahalanobis distance between the samples. Finally, the score-based classification is performed to classify the differentiation of normal as well as malignant colon biopsy samples into poor, moderate or great. The proposed methodology is implemented in the MATLAB platform and the experimental results show that our proposed method achieves an accuracy of 98.75%.