Landmark Influence Visualization in Active Shape Models
K. Van Golen · Research Repository (Delft University of Technology) · 2014
Segmentation of shapes in Magnetic Resonance Image (MRI) data is a non-trivial problem. Active Shape Models (ASM) is an approach that attempts to fit a Statistical Shape Model to landmarks. One possible approach is to let a user fit the model onto the MRI data by placing landmarks for ASM. Usually, the user can only see the global fit of the Statistical Shape Model; adjustments to landmarks have a global effect. This hinders the user in time-efficiently constructing a set of landmarks that maximizes the accuracy of the segmented shape. A novel landmark influence visualization is proposed that improves the usability of Active Shape Models. Landmark influence is defined as the change the model fit will undergo if a landmark is removed. The landmark influence is visualized using disk glyphs around landmarks. A region-of-interest filter can be used to inspect influence on local regions. Distance maps and animations visualize how the shape will change. A user study was performed to evaluate the visualization. No correlation was found between segmentation accuracy and the visualization. Similarly, no correlation was found between the time required for segmentation and the visualization. Nevertheless, a questionnaire revealed that users found the disk glyphs and animations useful in their segmentation workflow.