Image segmentation
View Patent ↗According to one embodiment there is provided a method of selecting a plurality of M atlases from among a larger group of N candidate atlases to form a multi-atlas data set to be used for computer automated segmentation of novel image data sets to mark objects of interest therein. A set of candidate atlases is used containing a reference image data set and segmentation data. Each of the candidate atlases is segmented against the others in a leave-one-out strategy, in which the candidate atlases are used as training data for each other. For each candidate atlas in turn, the following is carried out: registering; segmenting; computing an overlap; computing a value of the similarity measure for each of the registrations; and obtaining a set of regression parameters by performing a regression with the similarity measure being the independent variable and the overlap being the dependent variable. The M atlases are then selected from among all the N candidate atlases to form the multi-atlas data set, the M atlases being those atlases determined to collectively provide the highest aggregate overlap over all the training data image sets.
1. An image processing method for segmenting a novel image data set to mark an object of interest therein, the method comprising:
(a) providing a computer with a multi-atlas data set comprising a plurality of atlases, each atlas containing a reference image data set and segmentation data which marks an object of interest within the reference image data set;
(b) inputting into the computer a novel image data set for segmentation;
(c) registering in the computer the novel image data set with each of the atlases;
(d) selecting in the computer at least one atlas of the plurality of atlases to use for segmentation based on the registration performed in the registering step;
(e) segmenting in the computer the novel image data set based on the selected at least one atlas; and
(f) outputting in the computer the novel image data set with the associated segmentation data, so that the object of interest in the novel image data set is marked,
wherein each atlas of the plurality of atlases further comprises regression parameters representing results of a regression analysis between (i) real overlap between the atlas and a set of training image data sets, and (ii) a similarity measure between the atlas and the set of training image data sets; and
wherein the segmentation performed in the segmentation step is carried out by:
(a) computing a value of the similarity measure for each of the registrations, and (b) obtaining a predicted overlap for each of the registrations by applying the respective regression parameters to the respective values of the similarity measure; and
the selecting step includes selecting the at least one atlas of the plurality of atlases to use for the segmentation step based on the predicted overlaps.
2. The method of claim 1 , wherein the regression analysis is based on a linear regression function.
3. The method of claim 1 , wherein the similarity measure is one of: cross correlation, mutual information, and normalized mutual information.
4. The method of claim 1 , wherein the similarity measure is computed based on a region of interest within the image data set.
5. The method of claim 4 , wherein the region of interest corresponds to said object of interest or a further object of interest.
6. A non-transitory computer program product bearing machine readable instructions for carrying out the method of claim 1 .
7. A computer loaded with and operable to execute machine readable instructions for carrying out the method of claim 1 .