Figure 2.

Schematic showing the use of supervoxels in preparing high precision phase map priors. The raw multiphase data (a) are initially fitted to the Fermi function with low precision priors (1) to yield a raw phase map (b). This raw phase map contains good spatial information but the phase values themselves are overestimated. The data are smoothed and clustered using supervoxels (2) to yield ROIs for each supervoxel phase cluster present (c). The raw multiphase data are then combined with the supervoxel ROIs (3) to yield high SNR supervoxel-ROI means of the multiphase data (d). This high SNR multiphase dataset is then fitted again (4) yielding a high precision phase map (e) which can be used to generate the final CBF maps (f), in combination with the raw data (5).