Gland segmentation leading to cancer detection For colorectal biopsy images (P-0338) (MFN 6803)

Abstract

The architectural appearance, morphology and structural formation of glands contains information that helps in the detection of adenomatous disease. To assist pathologists in the grading of cancer, the segmentation of the biopsy images obtained after colonoscopy is performed to extract the glandular information. Pathologist usually take days to come to the conclusion of whether the glands are having cancerous tissues or not through the manual process. Hence, we proposed an algorithm which can automate the segmentation as well as classification of histopathology images stained by hematoxylin and eosin of tumor tissues of colorectal cancer. The research conducted related to the colorectal cancer is scarce as compared with that of the breast and prostate cancer. The major problem arises because of the glandular variability and the heterogeneity of cells. The proposed technique utilizes the information related to the intensity and morphology form the segmentation of glands, along with Deep Convolutional Neural Network (CNN) which helps in evaluating the malignancy. Furthermore, Transfer learning technique is used to train AlexNet for classification. The data set taken to test the proposed implementation was taken from MCCAI GLAS challenge which contains couple of images of benign and malignant.