In this presentation the authors described their method to segment a stack of similar images from different patients using a basic segmentation of one sample. They do this based on a weakly supervised classification algorithm to segment the complete set. Examples were shown to demonstrate the process where only a small amount of user interaction was required to achieve the segmentation.
They conclude that using over-segmentation through superpixels combined with local descriptors makes the labelling problem in multi-patient segmentations tractable.
One question after this presentation was about the clinical application of this kind of method. Most probably the highest value application will be in (retrospective) clinical research where often large sets of data from different patients have to be segmented to segment certain anatomical or pathological structures.
No comments:
Post a Comment