Browse Prior Art Database

Using multi-atlas registration for a learning automatic segmentation system Disclosure Number: IPCOM000190369D
Publication Date: 2009-Nov-25
Document File: 3 page(s) / 424K

Publishing Venue

The Prior Art Database



This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 3



Using multi-atlas registration for a learning automatic segmentation system

    Delineation of certain objects in medical images is an important clinical task. It is needed, for instance, for tumor volumetry (therapy response assessment), to create new organ models for model-based segmentation, or for target definition in radiotherapy. However, without software support delineation of certain objects can be a time consuming and tedious task.

    Applications: Generally, the proposed method solves the problem of a coarse initial positioning of contours for segmentation. In the context of applications which require segmentation of organs (e.g. radiation therapy planning), the propagated contours serve as initial segmentations which are refined either automatically (e.g. by using Model Based Segmentation) or manually. Other applications might include the segmentation of certain tissue types for PET quantification (e.g. delineation of liver as reference tissue), or in combination with false-positive suppression in automatic hot spot detection.

    In a variation, the annotated images in the database are clustered based on their similarity. If the segmentations shall be propagated to a new image, the similarity computation is only performed between the new image and a representative of each cluster. The contours are then propagated from the most similar representative onto the new image. The annotations on the original images could be used to estimate an error of the registration (upon which the similarity computation is based) and hence to give an estimate of the expected error margin of the propagated contour; this information might be used in the further application of the contour (e.g. in order to identify regions which are with a high probability part of the organ of interest). In this scenario, addition of new datasets is difficult and could not be done by the user, such that the database would be fixed for a certain application.
- The key idea is to use a sufficiently large database of images with annotations. These annotations will typically consist of "approved" delineations of organs or anatomical structures.
- If the user likes to segment a certain organ in a new image the system automatically computes the most similar images from the databa...