Automated Mucous Glands Detection and Segmentation in Colon Histology Images Using Semantic Segmentation

Abstract

The most prevalent form of Colon Cancer is the Colorectal Adenocarcinoma which originates particularly in intestinal glandular structures. For its prognosis and plan of treatment, pathological tests are conducted where the morphology of intestinal glands including an architectural appearance as well as glandular formation are analyzed in samples collected during biopsies. But in modern pathology, to achieve good inter-observer along with intra-observer reproducibility of cancer grading still remains a key challenge owing to variations present in staining, sectioning and fixation procedures that also lead to artifacts introduction and variances in tissue sections and ultimately resulted in variances in gland appearances. Also, manual segmentation and classification of these glandular structures are time-consuming owing to large datasets from a single patient. Thus, in this paper semantic segmentation model for effectively detecting and segmenting the mucous glands from large size histopathological images i.e. 775 × 552 × 3 is proposed which keeps the original size at the input and is the main highlight of the model. The model is trained and tested using the dataset from GlaS@MICCAI 2015 Colon Gland Segmentation Challenge.