DDANet: Dual Decoder Attention Network for Automatic Polyp Segmentation

Colonoscopy is the gold standard for examination and detection of colorectal polyps. Localization delineation polyps can play a vital role in treatment (e.g., surgical planning) prognostic decision making. Polyp segmentation provide detailed boundary information clinical analysis. Convolutional neural networks have improved performance colonoscopy. However, usually possess various challenges, such as intra-and inter-class variation noise. While manual labeling polyp assessment requires time from experts prone to human error missed lesions), an automated, accurate, fast improve quality delineated lesion boundaries reduce rate. The Endotect challenge provides opportunity benchmark computer vision methods by training on publicly available Hyperkvasir testing separate unseen dataset. In this paper, we propose novel architecture called “DDANet” based dual decoder attention network. Our experiments demonstrate that model trained Kvasir-SEG dataset tested achieves dice coefficient 0.7874, mIoU 0.7010, recall 0.7987, precision 0.8577, demonstrating generalization ability our model.