PNNL

Abstract

Analysis of gene expression patterns in brain images ob- tained from high-throughput in situ hybridization to understand the function of the gene requires accurate and consistent annotations of anatomical regions/subregions. Such annotations are generally obtained by mapping an anatomical atlas onto the gene expression images through intensity- and/or landmark-based registration methods or deformable model-based segmentation methods. Owing to the complex appearance of the gene expression images, these approaches generally require a pre- processing step to determine landmark correspondences to incorporate landmark-based geometric constraints. In this paper we propose a novel method for landmark-constrained intensity-based registration without determining landmark correspondences a priori. The proposed method performs dense image registration and also identies the true landmark correspondences simultaneously by solving a single Markov Random Field model. Additionally, a machine learning approach is utilized to improve the discriminating properties of local descriptors for landmark matching by projecting them in an effcient Hamming space of lower dimension to overcome the uneven appearance of the gene expression images. We show that our method achieves desired results qualitatively and also compares well quantitatively with the expert's annotations.