Abstract
Inspection and categorisation of polypoid lesions is important for establishing diagnoses and influencing clinical management of colorectal cancer. We present a study analysing the surface morphology of excised colorectal polyps imaged using optical projection tomography. The differential geometry of polyp surfaces, segmented using a level sets method, is explored in terms of local, multi-scale shape index and curvedness descriptors. A surface region of interest can be represented using histograms of these descriptors. Experiments are described that investigate the ability to categorise regions based on these histograms. Specifically, polyps were automatically assigned Kudo's pit pattern classes (a scheme used for endoscopic assessment). Five types of classifiers were compared using descriptors at multiple spatial scales. Support vector machines obtained rates broadly in line with intra-observer agreement on a data set of 28 tomographic polyp images.
Acknowledgements
We are grateful to the anonymous reviewers for their careful comments on an earlier version of this paper.
Disclosure statement
No potential conflict of interest was reported by the authors.