Image segmentation via multi-atlas fusion with context learning

A method for segmenting tissue within a computed tomography (CT) scan of a region of interest into one of a plurality of tissue classes, comprising:

• registering a plurality of atlases to the CT scan to produce a plurality of registered atlases, each registered atlas labelling at least a portion of a plurality of voxels in the CT scan as belonging to one of the plurality of tissue classes;
• selecting a proper subset of the plurality of registered atlases;
• and
• fusing the selected proper subset of registered atlases by minimizing a total expectation of labeling error, wherein a likelihood of any given set of atlases making an error is modeled as a joint probability of the set of atlases making a segmentation error at a given voxel, to produce a combined segmentation.

A system for segmenting tissue within a computed tomography (CT) scan of a region of interest into one of a plurality of tissue classes, comprising:

• a processor;
• a non-transitory computer readable medium operatively connected to the processor and storing machine executable instructions, the instructions comprising:
• a plurality of registered atlases representing the region of interest;
• a registration component configured to register each atlas to the CT scan to provide a registered atlas in which at least a portion of a plurality of voxels in the CT scan as belonging to one of the plurality of tissue classes;
• a context learning module configured to generate a context model, comprising respective likelihoods that each voxel of the CT scan is a member of each of the plurality of tissue classes, from at least one feature vector, comprising a plurality of image features, extracted from the CT scan;
• an iterative atlas selection configured to select a proper subset of the plurality of registered atlases according to the context model and the registered atlases;
• and
• a statistical fusion model configured to produce a combined segmentation as a probabilistic fusion from the selected proper subset of registered atlases.