An Efficient Algorithm for Targeted Reconstruction of Tomographic Data

Applications such as cardiac imaging require high-resolution images of a targeted display field-of-view (DFOV) within the patient. While reconstructing only the targeted DFOV is straightforward for analytical reconstruction algorithms, conventional formulations of iterative algorithms require that the DFOV include the entire emitting object, since they work to match the measured data to an estimate formed by applying the system model to the current estimate of the image. If the estimated image does not include activity outside the DFOV, its forward projection cannot correctly match the measurement. A brute force solution is to reconstruct the full FOV at a pixel pitch equal to that desired in the targeted DFOV, then select the pixels corresponding to the targeted DFOV; this is a computationally expensive solution whose cost rises quadratically as the targeted DFOV decreases in size. We present a method to efficiently reconstruct only the DFOV after estimating the contribution of activity outside the DFOV from a first-pass, low-resolution reconstruction. We maintain high image quality and relatively consistent convergence characteristics in the targeted DFOV reconstruction by using the estimate of activity outside the DFOV as an additive correction term in the forward system model, rather than subtracting the estimate from the raw data. We demonstrate artifact-free, quantitatively accurate images with significant speed-up in reconstruction times over the brute-force method.