Enhanced Colorectal Cancer Detection and Localization Using Self-Attention Mechanisms in Deep Learning

Colorectal cancer belongs to the top/top-ranked malignancies worldwide, ranking among the leading causes of cancer deaths. Hence, improved methods for detection and localization are required. The study presented here describes state-of-the-art methods to improve colorectal cancer detection and localization with self-attention in a Convolutional Neural Network framework. A method that can take advantage of CNN in feature extractions and further combine them with self-attention mechanisms capturing long-range dependencies and contextual information across medical images makes ours very effective in highlighting subtle features and intricate patterns characteristic of CRC often overlooked by conventional methods. Such is the architecture of the proposed CNN, carefully designed to overcome specific challenges in colorectal tissue analysis: tissue heterogeneity and varying appearance of cancerous regions. This self-attention module, therefore, enhances the accuracy and robustness of cancer detection and localization. The proposed approach has been evaluated on an overall dataset of medical images of a colorectal nature and outperformed existing state-of-the-art methods.