Automated visibility map of the internal colon surface from colonoscopy video

Abstract

Using a simple cylindrical model for the colon and reducing drift by registering band images allows for the generation of visibility maps. The current results also suggest that the provided feedback through the visibility map could enhance clinicians' awareness of uncovered areas, which in return could reduce the probability of missing polyps.

Learning colon centreline from optical colonoscopy, a new way to generate a map of the internal colon surface

“…3 Two sample visibility maps generated from two different colon segments generated by our method and [7]. On the left our map shows an uncovered area, this has not appeared on the map generated by Armin et al [7]. The difference between our method and [7] is also clear as the map generated by our method has fewer artefacts and shows colon folds due to its consistency in CCL detection Fig.…”

“…Band image and visibility map 2.2.1 Band image: By knowing intrinsic camera parameters, the above cylinder can be projected onto the colonoscopy video frame to unroll it and generate a radial strip called a band image ( Fig. 2c) [7]. The radius of the cylinder is set to be constant throughout.…”

“…A median filter was applied to the average motion flows to ensure a consistent motion. Some examples of a visibility map generated by our method and method explained in [7] are presented in Fig. 3 (first panel in each group).…”

“…The projection of the cylinder using our CCL ConvNet method is shown as the first row of Fig. 6 and the second row shows those generated by Armin et al [7] using the centre-of-dark region and camera R,t. In these sequences, camera translates to the left side of colon occluding part of the darkest region, though the centreline does not shift dramatically.…”

“…Such a map can provide useful information about uncovered areas during colonoscopy, map the position of any detected polyp, and be used as a reference to follow up with patients. Previous work taking a similar approach [7,9] generates a visibility map using the following steps: (i) estimate camera pose and infer 3D structure, (ii) fit a cylinder into the 3D structure to estimate an average radius, (iii) compute the centre-of-dark region for each frame, (iv) using camera parameters and the centre-of-dark region, project the cylinder onto endoscopy images and unroll the images (into band images), (v) stitch the band images to generate a visibility map. Since this method is based on camera pose estimation and 3D reconstruction, it can be computationally expensive and complex and can be unstable and fail for sparsely textured frames with the complex structure of the colon wall.…”