PET Image Reconstruction Using Information Theoretic Anatomical Priors

Abstract

We describe a nonparametric framework for incorporating information from co-registered anatomical images into positron emission tomographic (PET) image reconstruction through priors based on information theoretic similarity measures. We compare and evaluate the use of mutual information (MI) and joint entropy (JE) between feature vectors extracted from the anatomical and PET images as priors in PET reconstruction. Scale-space theory provides a framework for the analysis of images at different levels of detail, and we use this approach to define feature vectors that emphasize prominent boundaries in the anatomical and functional images, and attach less importance to detail and noise that is less likely to be correlated in the two images. Through simulations that model the best case scenario of perfect agreement between the anatomical and functional images, and a more realistic situation with a real magnetic resonance image and a PET phantom that has partial volumes and a smooth variation of intensities, we evaluate the performance of MI and JE based priors in comparison to a Gaussian quadratic prior, which does not use any anatomical information. We also apply this method to clinical brain scan data using F18 Fallypride, a tracer that binds to dopamine receptors and therefore localizes mainly in the striatum. We present an efficient method of computing these priors and their derivatives based on fast Fourier transforms that reduce the complexity of their convolution-like expressions. Our results indicate that while sensitive to initialization and choice of hyperparameters, information theoretic priors can reconstruct images with higher contrast and superior quantitation than quadratic priors.

PET Image Reconstruction Using Kernel Method

“…A recent review on using anatomical prior information for PET image reconstruction can be found in [2]. Among different regularization methods [3]–[8], [10]–[13], the Bowsher method [9], which adaptively chooses neighboring pixels for each pixel in the image estimate using information from a prior image, was found to utilize anatomical prior information better than others in terms of performance and computational complexity [16]. The nonlocal regularization [17] can incorporate anatomical weights through a methodology similar to the neighborhood selection in the Bowsher method [14], [15] to improve emission reconstruction [18]–[20].…”

Resolution modeling in PET imaging: Theory, practice, benefits, and pitfalls

“…28 In fact, anatomy guided PET reconstruction allows concurrent improvements in both the effective spatial resolution and signal-to-noise ratios in the reconstructed images. Nonetheless, a number of simplifying assumptions are commonly made (e.g., uniformity in radiopharmaceutical uptake within anatomic labels): to this end, more sophisticated approaches [29][30][31][32][33][34] have been investigated, though they are often seen to introduce a number of additional parameters to be further fine-tuned for particular tasks of interest. This approach thus remains an open area of interest.…”

Fat-Constrained 18F-FDG PET Reconstruction in Hybrid PET/MR Imaging

“…Our approach of using the mDixon imaging technique with its inherent segmentation into watery and fatty compartments obviates additional image processing and may therefore be preferable. One standard approach to incorporate prior anatomic information into the reconstruction algorithm is technically motivated by the Bayesian framework, in which maximization of the logarithm of the posterior probability directly reveals the prior-here, the lower probability of tracer accumulation in fat tissue-as an additive penalty term (15,16,28). Typically, these schemes seek to establish contiguous regions of uniform uptake in the PET image that correspond to edge-separated regions in the MR or CT image, for instance, minimum cross-entropy reconstruction.…”

“…Augmentation of standard iterative reconstruction algorithms (11,12) with penalty and regularization terms to enforce desired image characteristics such as regional smoothness (13), or, equivalently, Bayesian motivated reconstruction techniques (14) using appropriate priors, has been proposed for this purpose. The insight that the spatial distribution of a PET tracer is constrained not only by the underlying physiology but also to some extent by anatomy has led to penalization approaches that use the anatomic information gained from CT or MR images, again with the rationale of achieving spatial correlation of continuities and discontinuities between metabolic and anatomic images (15)(16)(17). In this work, we incorporated into the PET reconstruction algorithm a soft constraint that down-weighs PET image values in regions of tissue types in which tracer uptake is not expected, in favor of surrounding regions.…”