Label fusion is a core step of Multi-Atlas Segmentation (MAS), which has a decisive effect on segmentation results. Although existed strategies using image intensity or image shape to fuse labels have got acceptable results, there is still necessity for further performance improvement. Here, we propose a new label fusion strategy, which considers the joint information of intensity and registration quality.

The correlation between any two atlases is taken into account and the probability that two atlases both give wrong label is used to compute the fusion weights. The probability is jointly determined by the registration error and intensity similarity of the two corresponding atlas-target image pairs. The proposed label fusion algorithm is named Registration Error and Intensity Similarity based Label Fusion (REIS-LF).

Using 3D Magnetic Resonance (MR) images, the proposed REIS-LF algorithm is validated in brain structure segmentation including the hippocampus, the thalamus and the nuclei of the basal ganglia. The REIS-LF algorithm has higher segmentation accuracy and robustness than the baseline AQUIRC-W algorithm.

Taking the registration quality, the inter-atlas correlations and intensity differences into account in label fusion benefits to improve the object segmentation accuracy and robustness.