Fig. 2.

Semi-automatic image processing is used to create a mask. The (fat+water) (a) and fat-only images are transformed into estimates of fat-only and water-only signals as per Eq. 3 and 4 (b, c, resp.). The water-only signal is thresholded to remove air (d). Ghosting artifacts are substantially reduced following morphological opening and closing (e). Disconnected ghosts of high signal are removed by a region growing seeded from the center pixel of the image volume (f). A human observer edits the image map to fix any remaining problems (g). Finally, the mask is applied to the water only signal to show how artifacts have been removed (h).