A novel technique to incorporate structural prior information into multi-modal tomographic reconstruction

Kazantsev, Daniil and Ourselin, Sébastien and Hutton, Brian F and Dobson, Katherine J and Kaestner, Anders P and Lionheart, William R B and Withers, Philip J and Lee, Peter D and Arridge, Simon R (2014) A novel technique to incorporate structural prior information into multi-modal tomographic reconstruction. Inverse Problems, 30 (6). 065004. ISSN 0266-5611 (https://doi.org/10.1088/0266-5611/30/6/065004)

[thumbnail of Kazantsev-etal-IP-2014-A-novel-technique-to-incorporate-structural-prior-information]
Preview
Text. Filename: Kazantsev_etal_IP_2014_A_novel_technique_to_incorporate_structural_prior_information.pdf
Final Published Version
License: Creative Commons Attribution 3.0 logo

Download (2MB)| Preview

Abstract

There has been a rapid expansion of multi-modal imaging techniques in tomography. In biomedical imaging, patients are now regularly imaged using both single photon emission computed tomography (SPECT) and x-ray computed tomography (CT), or using both positron emission tomography and magnetic resonance imaging (MRI). In non-destructive testing of materials both neutron CT (NCT) and x-ray CT are widely applied to investigate the inner structure of material or track the dynamics of physical processes. The potential benefits from combining modalities has led to increased interest in iterative reconstruction algorithms that can utilize the data from more than one imaging mode simultaneously. We present a new regularization term in iterative reconstruction that enables information from one imaging modality to be used as a structural prior to improve resolution of the second modality. The regularization term is based on a modified anisotropic tensor diffusion filter, that has shape-adapted smoothing properties. By considering the underlying orientations of normal and tangential vector fields for two co-registered images, the diffusion flux is rotated and scaled adaptively to image features. The images can have different greyscale values and different spatial resolutions. The proposed approach is particularly good at isolating oriented features in images which are important for medical and materials science applications. By enhancing the edges it enables both easy identification and volume fraction measurements aiding segmentation algorithms used for quantification. The approach is tested on a standard denoising and deblurring image recovery problem, and then applied to 2D and 3D reconstruction problems; thereby highlighting the capabilities of the algorithm. Using synthetic data from SPECT co-registered with MRI, and real NCT data co-registered with x-ray CT, we show how the method can be used across a range of imaging modalities.