Assessment of Vascular Network Segmentation

We present an analysis framework to assess the quality and accuracy of vesselsegmentation algorithms for three dimensional images. We generate synthetic (insilico) vessel models which act as ground truth and are constructed to embodyvarying morphological features. These models are transformed into imagesconstructed under different levels of contrast, noise, and intensity. Todemonstrate the use of our framework, we implemented two segmentationalgorithms and compare the results to the ground truth model using severalmeasures to quantify the accuracy and quality of segmentation. Furthermore, wecollect metrics which describe the characteristics of the vessels it fails tosegment. Our approach is illustrated with several examples. Funded by NCIContract No.HHSN261200800001E.