Weakly Supervised Polyp Segmentation in Colonoscopy Images Using Deep Neural Networks

Colorectal cancer (CRC) stands as a leading cause of mortality globally. Colonoscopy- based preventive screening has proven effective in reducing its mortality rates. However, studies indicate that a certain percentage of polyps evade detection during colonoscopy due to various factors. This research aims to explore the DeepLabV3+ model, a neural network architecture capable of identifying and delineating polyps within colonoscopy images. Starting from the previous works, this study also aims to propose a novel variant of the model by exploring diverse approaches and techniques. The objective is to expand the model's capabilities and enhance its performance in polyp detection within colonoscopy images. This novel variant, called DeepLabV3+CRF, seeks to integrate the strengths of the DeepLabV3+ model with Conditional Random Fields (CRF), aiming to improve the precision of polyp delineation and enhance the overall performance. The utilization of the CRF within the framework allows for refining the boundaries of detected polyps by incorporating contextual information, thereby augmenting the accuracy and spatial coherence of the segmentation results. These solutions will be experimented with and tested on the Polyp-Box-Seg dataset, alongside other commonly used datasets within the polyp segmentation domain.