The fast development of hybrid imaging modalities in tomography, such as SPECT (single photon emission computed tomography)/CT (computed tomography), PET (positron emission tomography)/MRI (magnetic resonance imaging) and PET/CT, have increased an interest for reconstruction algorithms which are able to utilize a functional and anatomical information at the same time. In this paper a new method proposed for iterative reconstruction with anatomical prior in emission tomography (ET). The introduced regularization term is a modified anisotropic tensor diffusion filter which has shape-adapted smoothing properties. The filter accommodates available anatomical information which results in enhanced position and image dependent spatial resolution of emission images. Based on underlying orientations of normal and tangential vector fields for emission and anatomical images, the diffusion flux is rotated and scaled. Poisson likelihood fidelity and penalty terms are optimized separately by means of forward-backward splitting (FBS) technique. Presented approach is validated quantitatively using co-registered SPECT/MR synthetic data and compared with another anatomically penalized reconstruction as well as with iterative statistical reconstruction without regularization.