3D medical model low-pass filtering based on non-uniform spectral synthesis

Abstract

3D medical anatomical models often suffer from terrace artifacts instead of accurate anatomical structures, which would bring disadvantages to medical diagnosis and treatment. However, even advanced smoothing techniques may fail to remove terrace artifacts, because they tend to recognize the terrace artifacts as features to reconstruct them, rather than get rid of them. To solve the problem, we propose a new 3D low-pass filter, which is designed based on non-uniform spectral synthesis. Firstly, we use z axis oriented feature detection to correctly capture the terrace artifacts. Then, using spectral method as main tool, execute non-uniform spectral synthesis to reconstruct the geometry of 3D medical models with the high frequency noises and terrace artifacts removed absolutely. We implement our technique on several 3D medical models, and we also verify the results comparing to state of the art mesh smoothing approaches. Experimental results demonstrate that our method can obtain highly smooth 3D medical models with geometry accuracy and volume-preservation.