A Novel Approach for Detection of Tubular Objects and Its Application to Medical ImageAnalysis

摘要

We present a novel approach for detection of tubular objects in medical images. Conventional tube detection / lineness filters make use of local derivatives at multiple scales using a linear scale space; however, using a linear scale space may result in an undesired diffusion of nearby structures into one another and this leads to problems such as detection of two tangenting tubes as one single tube. To avoid this problem, we propose to replace the multi-scale computation of the gradient vectors by the Gradient Vector Flow, because it allows an edge-preserving diffusion of gradient information. Applying Frangi's vesselness measure to the resulting vector field allows detection of centerlines from tubular objects, independent of the tubes size and contrast. Results and comparisons to related methods on synthetic and clinical datasets show a high robustness to image noise and to disturbances outside the tubular objects.