| Description |
: |
Magnetic Resonance Imaging, Computed Tomography, and other image modalities are
routinely used to visualize a particular structure in the patient's body. The classification
of the image region corresponding to this structure is called segmentation.
Magnetic Resonance Imaging, Computed Tomography, and other image modalities are routinely used to visualize a particular structure in the patient's body. The classification of the image region corresponding to this structure is called segmentation. For applications in Neuroscience, it is important for the segmentation of a brain scan to represent the boundary of the brain as a single folded sheet. However the segmentation of the brain generally exhibits many erroneous holes. Consequently we have developed an algorithm for automatically removing holes in segmented medical scans while preserving the accuracy of the image. Upon concepts of Discrete Topology, we correct the holes based on the smallest modification to the image. First we detect each hole with a front propagation and a Reeb graph. Then we search for a number of loops around the hole on the isosurface of the image. Finally we select the loop that minimizes the modification to the image and correct the hole in the image. At each step we limit the size of the data in memory. With these contributions our algorithm removes every hole in the image with high accuracy and low complexity even for images that do not fit into the main memory. To help doctors and scientists to obtain segmentations without holes, we make our software publicly available.
|
