Template estimation plays a crucial role in computational anatomy since itprovides reference frames for performing statistical analysis of the underlyinganatomical population variability. While building models for templateestimation, variability in sites and image acquisition protocols need to beaccounted for. To account for such variability, we propose a generativetemplate estimation model that makes simultaneous inference of both bias fieldsin individual images, deformations for image registration, and variancehyperparameters. In contrast, existing maximum a posterori based methods needto rely on either bias-invariant similarity measures or robust imagenormalization. Results on synthetic and real brain MRI images demonstrate thecapability of the model to capture heterogeneity in intensities and provide areliable template estimation from registration.
展开▼
