Colorectal cancer is a prevalent and severe global health issue characterized by high mortality and incidence rates. The study emphasizes the UNet model, a pioneering deep learning architecture tailored for glandular structure categorization within colon histology images. Remarkably employed in medical imaging, UNet exhibits remarkable efficacy in abnormality detection, including tumor identification. Leveraging the Warwick-QU dataset, featuring expertly annotated Hematoxylin and Eosin (H&E) stained slides, utilize research trains the UNet model to accurately delineate glandular structures and malignant tissues in colon histology images. Performance assessment of the employed UNet-based model entails rigorous evaluation through established metrics such as the Dice coefficient, Intersection over Union (IOU), model efficacy, training epochs, and Dice coefficient-based loss function. After-effects highlight the substantial potential of deep learning, particularly the convolutional neural network (CNN)-based UNet model, in early colon cancer detection, diagnosis, and subsequent intervention. Timely intervention holds promise for impeding disease progression and enhancing patient prognosis. Also, it pioneers advancements in early colon cancer detection and diagnosis, underpinning future clinical management enhancements. The study’s significance lies in its contribution to accurate and efficient automated gland segmentation techniques, pivotal for precise diagnosis and effective intervention in colon cancer cases.