| 摘要 |
A method is disclosed for segmenting an organ, more particularly the liver, in n-channel volume data records from magnetic resonance imaging. In at least one embodiment of the method, probability distributions are generated on the basis of segmentation results from n-channel training data records, which probability distributions specify probabilities of positions of voxels and intensity values, which have been reduced from n-dimensions to one dimension with the aid of discriminant reduction, of voxels in the training data records belonging to the organ. At least one 3D probability data record is then generated from the n-channel volume data records on the basis of the probability distributions, in which 3D probability data record each voxel is assigned a probability of belonging to the organ. A maximum value of the 3D probability data record is set as the start point for the segmentation, and the organ is subsequently segmented on the basis of the 3D probability data record using a region-growing technique. The method allows very reliable fully-automatic segmentation of the organ. |
