Context-aware mesh smoothing for biomedical applications

Smoothing algorithms allow to reduce artifacts from mesh generation, but often degrade accuracy. Thus, we present a method that identifies staircase artifacts which result from image inhomogeneities and binary segmentation in medical image data for subsequent removal by adaptive mesh smoothing. This paper makes the following specific contributions: caps, which are flat regions, resulting from segmentation or clipping at the endings of anatomical structures are detected and modified by smoothing; the effects of the adaptive smoothing method involving context information are quantitatively analyzed with respect to accuracy and their influence on blood flow simulations; the image stack orientation, which is relevant for this context-aware smoothing approach, is estimated automatically from the surface models. Thus, context-aware smoothing enables to adaptively smooth artifact areas, while nonartifact features can be preserved. The approach has been applied to CT neck datasets, as well as phantom data and the results are evaluated regarding smoothness and model accuracy. The accuracy of model orientation estimation and cap detection has been evaluated for clinical and phantom data. Finally, context-aware smoothing has been applied to CT angiography data for the simulation of blood flow. The simulation results are presented and prove the general suitability of context-aware smoothing.