Figure 1. Interactive refinement segmentation of a vertebral body contour in 2D from a magnetic resonance imaging (MRI) acquisition.

The leftmost image presents the native scan and the second image from the left shows the initial user-defined seed point (white) that has been placed inside the vertebral body for the interactive segmentation. The third image from the left presents the segmentation outcome for the current position of the user-defined seed. However, due to the bright region inside the vertebral body, the average gray value – which is automatically calculated from the region around the user-defined seed – is not calculated “correctly”, and thus the resulting contour (red) leaks in the upper area and misses an edge in the lower left (note: for the interactive segmentation of the vertebral body, a rectangle was used as template to construct the graph. Thereby, the center of the rectangle is the user-defined seed point and the yellow crosses in the two rightmost images display the four corners of the rectangle). Finally, the rightmost image presents the result of the refined segmentation. Therefore, the user simply placed three additional seeds (white dots on the contour of the vertebral body), and thus forced the algorithm to perform the min-cut at these positions – which also influences the cuts along the neighboring rays. Furthermore, additional gray value information can be extracted around these extra seeds that the user placed on the contour of the vertebral body.