Detection of Polyps via Shape and Appearance Modeling

Abstract

This paper describes a CAD system for the detection of colorectal polyps in CT. It is based on stochastic shape and appearance modeling of structures of the colon and rectum, in contrast to the data-driven approaches more commonly found in the literature it derives predictive stochastic models for the features used for classification. The method makes extensive use of medical domain knowledge in the design of the models and in the setting of their parameters. The proposed approach was successfully tested on challenging datasets acquired under a protocol with little colonic preparation; such protocol reduces patient discomfort and potentially improves compliance.