DilatedSegNet: A Deep Dilated Segmentation Network for Polyp Segmentation

Colorectal cancer (CRC) is the second leading cause of cancer-related death worldwide. Excision polyps during colonoscopy helps reduce mortality and morbidity for CRC. Powered by deep learning, computer-aided diagnosis (CAD) systems can detect regions in colon overlooked physicians colonoscopy. Lacking high accuracy real-time speed are essential obstacles to be overcome successful clinical integration such systems. While literature focused on improving accuracy, parameter often ignored. Toward this critical need, we intend develop a novel learning-based architecture, DilatedSegNet, perform polyp segmentation fly. DilatedSegNet an encoder-decoder network that uses pre-trained ResNet50 as encoder from which extract four levels feature maps. Each these maps passed through dilated convolution pooling (DCP) block. The outputs DCP blocks concatenated series decoder predicts mask. proposed method achieves operation 33.68 frames per with average dice coefficient (DSC) 0.90 mIoU 0.83. Additionally, also provide heatmap along qualitative results shows explanation location, increases trustworthiness method. publicly available Kvasir-SEG BKAI-IGH datasets suggest give feedback while retaining DSC, indicating potential using models real settings near future. GitHub link source code found here: https://github.com/nikhilroxtomar/DilatedSegNet .