Structural SIMilarity and Spatial Frequency Motivated Atlas Database Reduction for Multi-atlas Segmentation
Multi-atlas segmentation has been proved to be an effective approach in biomedical images segmentation. In this approach, all atlases in atlas database participate in segmentation process. This results in the fact that the computational cost and the redundancy bias will increase correspondingly with the size of atlas database. To decrease the computational cost, many researches proposed atlas selection which selects limited atlases to register with the query image. This strategy effectively reduces the number of atlases in registration, but is ineffective in reducing redundancy bias. To address this problem, we propose a novel strategy that improving segmentation quality through analyse of atlas database. Our contributions are summarized as follow: (1) define optimal minimum reduced atlas database (MinRAD); (2) give an algorithm of constructing optimal MinRAD based structure similarity and spatial frequency; and (3) demonstrate validity of our strategy by experimental results.
KeywordsMulti-atlas segmentation Atlas database reduction Structural SIMilarity (SSIM) Spatial Frequency (SF)
- 1.Cuadra MB, Pollo C, Bardera A, Cuisenaire O, Thiran JP (2003) Atlas-based segmentation of pathological brain MR images. ICIP 573–576:2003Google Scholar
- 5.Akinyemi A, Plakas C, Piper J, Roberts C, Poole I (2012) Optimal atlas selection using image similarities in a trained regression model to predict performance. In: Biomedical Imaging (ISBI), 2012 9th IEEE International Symposium on IEEE, pp 1264–1267Google Scholar