Paper: Learning colon centerline from optical colonoscopy, a new way to generate a map of the internal colon surface
Optical colonoscopy is known as a gold standard screening method in detecting and removing cancerous polyps. During this procedure, some polyps may be undetected due to their positions, not being covered by the camera, or missed by the surgeon.
In this paper, we introduce a novel ConvNet algorithm to map the internal colon surface to a 2D map (visibility map) which can be used to increase awareness of clinicians (particularly junior clinicians) about areas they might miss.
This was achieved by leveraging a colonoscopy simulator to generate a dataset consisting of colonoscopy video frames and their corresponding Colon Center Line (CCL) points in 3D camera coordinates.
A pair of video frames were used as input to a ConvNet, whereas the output was a point on the CCL and its direction vector. By knowing 3D centerline points for each image and modeling the colon as a cylinder we could unroll images to build a visibility map.
However, note that this model is just for visualization, in practice the actual colon need not be cylindrical nor even symmetric. We validated our results using both simulated and real colonoscopy frames.
Our results showed that using consecutive simulated frames to learn the colon centerline can be generalized to real colonoscopy video frames to generate a visibility map.
Fig 5: Projection of the cylinder onto a short sequence of real frames. The first row shows projection using CCL information estimated by our ConvNet and the second row presents results from [7]. When the camera moves from middle to left, the position of the projected cylinder using [7] changes with respect to the change of darkest region, but in our method, CCL remains consistent and keeps cylinder projection following the true centerline position.
Download the full paper here.
For more information, contact us.