Principal Flow for Tubular Objects with Non-Circular Cross-Sections

Various anatomical objects are tubular in shape. These structures can be modeled by describing their curvilinear path and the cross-sectional shape along the path. However, most research on tubular object segmentation has focused on vascular systems, and often assumes a circular cross-section. These techniques are not readily applicable to anatomy such as the cochlea, which has a non-circular cross-sectional shape. We present the Principal Flow Filter, which calculates the local flow vector along a tubular object with a non-circular cross-section. It can be used to extract the centerline orientation and thus track along the tube, and as a basis for segmentation. We present results from generated data with a variety of cross-sectional shapes. The filter is shown to rapidly and robustly converge to the true orientation. We also analyse a CT scan of a human cochlea, with promising results.

• Type: Conference
• Year: 2004
• Where: ICPR
• Authors: Gavin Baker, Nick Barnes
• Keywords: Computer Vision, Cross-sectional Shape, ICPR 2004, Non-circular Cross-sectional Shape, Tubular Object Segmentation