NeoUNet : Towards Accurate Colon Polyp Segmentation and Neoplasm Detection

Automatic polyp segmentation has proven to be immensely helpful for endoscopy procedures, reducing the missing rate of adenoma detection endoscopists while increasing efficiency. However, classifying a as being neoplasm or not and segmenting it at pixel level is still challenging task doctors perform in limited time. In this work, we propose fine-grained formulation problem. Our aims only segment regions, but also identify those high risk malignancy with accuracy. We then present UNet-based neural network architecture called NeoUNet, along hybrid loss function solve Experiments show highly competitive results NeoUNet on our benchmark dataset compared existing models.