Abstract
While performing a colonoscopy, a doctor may need to be able to return to a previously viewed location within the colon. In particular, when regions of the colon are identified as unsurveyed (inadvertently not viewed), the colonoscopist must be able to find their way back to them. Returning to previously-seen locations in the colon requires that it work under the types of deformations the colon undergoes and recognizes the fact that the colon does not have significant photometric texture. In this work, we present a novel approach to this task. Salient geometric features known as haustral folds are used as markers to enable precise guidance back to a location, for example, to a given fold next to an unsurveyed region. We present two of the first methods, both medial-shape-based, for detecting individual haustral folds from colonoscopic video: CurvFold, which uses surface curvature derived from monocular depth estimation, and DeepFold, which uses state-of-the-art object detection to directly identify haustral folds. We demonstrate that CurvFold and DeepFold are able to detect haustral folds in video frames with mean chamfer distances to the ground truth of 29 and 9 pixels respectively.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ShapeMI_026.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=FbDtJzAphm
BibTex
@InProceedings{EhrSam_Medial_MICCAISAT2026,
author = { Ehrenstein, Samuel AND McGill, Sarah AND Rosenman, Julian AND Pizer, Stephen M.},
title = { { Medial Skeletons for Haustral Fold Detection in Colonoscopy } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17259},
month = {pending},
page = {pending}
}