Shape based computer aided diagnosis and automated navigation in virtual colonoscopy

This paper proposes a robust algorithm to detect colon polyps and cancerous lesions in virtual colonoscopy and present them to the user by automatically guiding the virtual camera. The detection algorithm uses Gaussian filters to construct the Hessian matrix, which represents the second order derivatives of a vector variate scalar valued function. Based on the sign and scale of the eigenvalues of the Hessian matrix, blob like lesions can be selected on a given scale. In the visualization stage the camera is moved along the colon centerline with its speed and viewing direction adopted to the results of detection. The camera path and the viewing direction are described by Kochanek-Bartels splines. The velocity along the path is also governed by a C2 continuous function. The resulting fly through is smooth and physically plausible, and it is guaranteed that the user can see all regions of interest and spends sufficient time looking at each of them.