Quantitative Assessment of Lesion Detection Accuracy, Resolution, and Reconstruction Algorithms in Neutron Stimulated Emission Computed Tomography

We present a quantitative analysis of the image quality obtained using filtered back-projection (FBP) with Ram-Lak filtering and maximum likelihood-expectation maximization (ML-EM)-with no postreconstruction filtering in either case-in neutron stimulated emission computed tomography (NSECT) imaging using Monte Carlo simulations in the context of clinically relevant models of liver iron overload. The ratios of pixel intensities for several regions of interest and lesion shape detection using an active-contours segmentation algorithm are assessed for accuracy across different scanning configurations and reconstruction algorithms. The modulation transfer functions (MTFs) are also computed for the cases under study and are applied to determine a minimum detectable lesion spacing as a form of sensitivity analysis. The accuracy of NSECT imaging in measuring relative tissue concentration is presented for simulated clinical liver cases. When using the 15th iteration, ML-EM provides at least 25% better resolution than FBP and proves to be highly robust under low-signal high-noise conditions prevalent in NSECT. However, FBP gives more accurate lesion pixel intensity ratios and size estimates in some cases; due to advantages provided by both reconstruction algorithms, it is worth exploring the development of an algorithm that is a hybrid of the two. We also show that NSECT imaging can be used to accurately detect 3-cm lesions in backgrounds that are a significant fraction (one-quarter) of the concentration of the lesion, down to a 4-cm spacing between lesions.