Incorporation of cardiac deformation as a prior in the reconstruction of gated cardiac SPECT data

Objectives This work is aimed at improving precision in the reconstruction of 4D gated cardiac SPECT data using priors that model the deformation of the myocardium.

Methods A modified MAP reconstruction algorithm was derived with normal cardiac motion of the 4D XCAT phantom as a prior which models the mechanical deformation of the heart. The prior was designed so that the image intensity was constrained by a Gaussian distribution with mean equal to the expected reconstructed value and spread equal to various predetermined sigma values. The algorithm was verified by first simulating noisy SPECT projections of the first frame of 16 cardiac gates of the XCAT phantom. The data were reconstructed using the new modified MAP reconstruction method for various iterations, weights of the prior, and sigma values constraining the spread in the allowed reconstructed intensity values.

Results The weight of the prior and the sigma value have significant effect on the accuracy of reconstructed cardiac images. Sharper reconstructed edges are obtained by increasing the weight of the prior. A broader Gaussian spread requires a larger weighting of the prior to improve the sharpness of the edges. Simulations also demonstrated improved image quality with sharp edges at boundaries for all cardiac gates of the XCAT phantom when applying the XCAT model of cardiac motion as a prior in the 4D reconstruction of all cardiac gates.

Conclusions More accurate modeling of cardiac deformation can provide improved precision of the reconstructed gated cardiac images. Further work is needed to determine the sensitivity of the reconstructed accuracy to the accuracy of the cardiac deformation model. The challenge is combining these results into more accurate modeling of the physiology of the heart and imaging physics to improve bias and variance in the estimation of kinetic compartment model parameters from 5D dynamic gated cardiac SPECT data.

Research Support NIH R01EB000121, R01HL091036, and RO1HL068075