Method for the fully automatic segmentation of an organ, in particular the renal parenchyma, from volume data sets in medical imaging

A method for segmenting an organ in volume data sets in medical imaging, wherein on the basis of segmentation results from training data sets, a region of interest is determined within a recorded volume data set which is most likely to contain the organ, and a likelihood data set is generated, wherein a likelihood of being associated with the organ is assigned to every voxel of the volume data set on the basis of its intensity values and the segmentation results from the training data sets, the method comprising:

• applying a threshold value method to the likelihood data set in the region of interest, by which an organ region is isolated from the background, in order to obtain a binary data set, in which voxels of the background and voxels outside the region of interest have a different value from voxels of the organ region,
• performing a step by step erosion of the organ region in the binary data set up to a predefinable number of steps, comprising
• removing, in every step starting from an outer limit of the organ region, a slice of the organ region,
• selecting, in the event of separation of the organ region into a number of subregions in one step, only one of the subregions which is most likely to contain the organ for the further erosion,
• wherein the subregion which is most likely to contain the organ is selected in each instance based on geometrical features in at least one slice of the binary data set, which are known from the training data sets,
• reducing, in the likelihood data set, the likelihood values of voxels of the subregions separated in the binary data set, so that a corrected likelihood data set results, and
• using, on the basis of the corrected likelihood data set, a segmentation technique to segment the organ from the likelihood data set and therefore also from the volume data set.