|
Quantitative analysis of the branching geometry of multiple branching-order vascular trees from 3D micro-CT data requires an efficient segmentation algorithm that leads to a consistent, accurate representation of the tree structure. To explore different segmentation techniques, we use isotropic micro-CT-images of intact rat coronary, pulmonary and hepatic opacified arterial trees with cubic voxel-side length of 5-20 micrometer. We implemented an active topology adaptive surface model for segmentation and compared the results from this algorithm with segmentations of the same image data using conventional segmentation methods. Because of the modulation transfer function of the micro-CT scanner, thresholding and region growing techniques usually underestimate small, or overestimate large, vessel diameters depending on the chosen grayscale thresholds. Furthermore, these approaches lack the robustness needed to overcome the effects of typical imaging artifacts, such as image noise at the vessel surfaces, which tend to propagate errors in the analysis of the tree due to its hierarchical nature. Our adaptable surface models include local gray- scale statistics, object boundary and object size information into the segmentation algorithm, thus leading to a higher stability and accuracy of the segmentation process. 5-20 micrometer. We implemented an active topology adaptive surface model for segmentation and compared the results from this algorithm with segmentations of the same image data using conventional segmentation methods. Because of the modulation transfer function of the micro-CT scanner, thresholding and region growing techniques usually underestimate small, or overestimate large, vessel diameters depending on the chosen grayscale thresholds. Furthermore, these approaches lack the robustness needed to overcome the e*ects of typical imaging artifacts, such as image noise at the vessel surfaces, which tend to propagate errors in the analysis of the tree due to its hierarchical nature. Our adaptable surface models include local gray-scale statistics, object boundary and object size information into the segmentation algorithm, thus leading to a higher stability and accuracy of the segmentation process.
|
