Leland Stanford Junior University

Padmavathi Sundaram

Sandy A Napel

David Seungwon Paik

Christopher Allen Beaulieu

Christopher Frederick Beaulieu

Afra J Zomorodlan

Improved surface feature recognition in CT images is provided by extracting a triangulated mesh representation of the surface of interest. Shape operators are computed at each vertex of the mesh from finite differences of vertex normals. The shape operators at each vertex are smoothed according to an iterative weighted averaging procedure. Principal curvatures at each vertex are computed from the smoothed shape operators. Vertices are marked as maxima and/or minima according to the signs of the principal curvatures. Vertices marked as having the same feature type are clustered together by adjacency on the mesh to provide candidate patches. Feature scores are computed for each candidate patch and the scores are provided as output to a user or for further processing.

Polyp detection using smoothed shape operators

A method of identifying polyps and in a medical image is provided. In a first step, a 3-dimensional model is made of the medical image that contains both polyps (if any were present in the original medical image) and folds. Next, a second 3-dimensional model of the medical image, which is a filtered version of the first model, is constructed in which folds are preserved, but polyps are minimized or eliminated. In a third step, any polyps that were contained in the medical image are identified by subtracting the second 3-dimensional model from the first 3-dimensional model. This subtraction results in a third 3-dimensional model, in which polyps are preserved but folds are minimized or eliminated. With the present inventive method, polyps may be easily and quickly identified without interference from folds.

Growing community of inventors

Palo Alto, CA, United States of America

Padmavathi Sundaram