To browse our segmented slices, click here. To watch an animation of our segmentation, click here We examined our algorithm performance at different thresholds. Each slice was examined visually to characterize the quality of its segmentation. The number of "good frames" were tallied, where a good frame was define as one which had not included an adjacent organ while having solid (>80%) coverage of the kidney.  We discovered that there were optimal thresholds, and were able to conclude that the number of good frames was primarily a function of the threshold and the median value of the voxel block surrounding the seed point.  Examples of a bad and good segmentation: On the left we see a slice from an overgrown region. On the right we see a well-segmented slice. The overgrown slice has mistakenly included an adjacent organ.  By examining an unsegmented abdominal slice closely we can see that the edge between adjacent organs is much weaker than the noise (if not non-existent). One weakness of 3D region growing is that once incorrect voxels are included, they continue to grow and can negatively affect other slices. |
