A multiresolution approach to flow feature extraction from phase contrast magnetic resonance angiography

The authors considered a multi-resolution, model based segmentation method for magnetic resonance angiography (MRA). It is a spatial domain based technique and an extension to 3D of a 2D curve segmentation method reported elsewhere. The method has been demonstrated to produce a concise symbolic description of the MRA data (in the form of vessel centre lines) and is efficient in its computational complexity being equivalent in processing to filtering by a 3×3×3 kernel, and based on a generalised and flexible image model which has great potential as a basis for both qualitative and quantitative assessment of MRA data. The work and results presented thus far are preliminary and currently there are several areas where consolidation and enhancement is necessary. There is a need to assess the levels of noise in the data in situ, to better control the confidence levels used for the hypothesis testing. Curve tracing is currently done probabilistically based purely on the local curvature. By considering the physical measurements of the data being imaged, e.g. speed of blood and the vessel diameters, local connectivity could be established using a conservation of mass constraint. Also, there is need to explicitly define bifurcations as part of the signal model. With regards to visualisation, some experimentation has already been carried out to represent flow direction and using the multiresolution vectors for generating filtered maximum intensity projections and predicting probable flow. The segmentation is also being applied to the estimation of blood pressure gradients in vivo