Adaptive application of spatial filters on raw CT images

The most significant constituent for a realistic blood flow simulation is the precise extraction of the patient specific geometry from digital computer tomographic images. The reconstructed data sets are not always sufficient to extract the contours/surfaces of the large aortic structures to build the computational model directly. The raw images are processed with the help of low level filters to remove in-homogeneity and undesired noise to achieve a smoother contour of the desired structures close to the original image data-set. A spatial filter, based on patch size, is applied iteratively on an image. It will be shown that with large patch sizes, distortion increases and with smaller patch sizes, a smooth contour is not achieved. A new, adaptive approach based on the image gradient which provides intrinsic information of the underlying objects in an image is presented. This new approach helps to apply the filters locally in a controlled manner to achieve the desired smooth contour with minimum distortion.