Colorectal Polyps Detection in Virtual Colonoscopy Using 3D Geometric Features and Deep Learning

Early diagnosis of colorectal polyps, before they turn into cancer, is one of the main keys for treatment. In this work, we propose a framework to help radiologists in identifying polyp candidates using virtual colonoscopy. In the proposed approach, first a colon is segmented from a CT scan, then 3D reconstruction, to generate a surface representation of the colon, is performed. From the reconstructed 3D colon, 2D images are generated using the virtual colonoscopy, Fly-In approach. To enhance polyp detection, we fuse these 2D images and the 3D colon representation by generating 3D geometric feature maps, e.g. depth and curvature maps. CNN-based models are trained and validated to detect polyps using the generated feature maps, which are combined in multi-channel images. These images are successfully used to train a CNN-based model that detects polyps with mAP ∼ 97.1%.Colorectal cancer, colon polyp, computerized tomography (CT), Detection, CNN, segmentation.