Voting Neural Network (VNN) for Endoscopic Image Segmentation

Colorectal cancer (CRC) is one of the most commonly diagnosed cancers in the recent few years. Recent studies show that 80%–95% of the endoscopic abnormalities are found in the intestine which would be the leading cause of cancer-related deaths globally. However, the conventional method of screening and diagnosis in colonoscopy is highly dependent on the expert endoscopist’s skills and experience. To automate colonoscopy analysis, several computer vision based techniques have been proposed for polyps detection and segmentation. However, polyp segmentation is a challenging research problem due to polyp morphology variabilities and intrinsic image appearance. The main goal of this research is an early and accurate segmentation of polyps to accurately determine the polyp region. In this paper, a heterogeneous ensemble deep learning model is proposed that combines the knowledge of three state-of-the-art deep learning architectures including UNet, ResUNet, and ResUNet++. Experiments are conducted on two real world datasets: Kvasir-SEG, and CVC-Clinic [1] and compared with state-of-the-art techniques. The proposed approach resulted in an accuracy of 0.93 on test data.