Image-Based Partial-Volume Correction in SPECT

Introduction and aim: In SPECT imaging, the finite spatial resolution of the imaging system leads to a blurring of the reconstructed image. Consequently, counts originating from each point in the source distribution will be spread out over a volume in the reconstructed image, leading to spill-over of counts between different structures and distortion of the apparent activity concentration. This process, which can seriously impair the accuracy in quantitative analysis, is termed the partial-volume effect. The magnitude of this effect depends on many factors but is generally determined by the size of the object of interest in relation to the spatial resolution of the system. At Skåne University Hospital, radionuclide therapy with 177 Lu-Tyr3-Octreotate is currently used for treatment of metastatic neuroendocrine tumors. Quantitative SPECT is utilized for treatment planning and dose assessment, focused on the kidneys which are the primary organs at risk. However, the activity concentration in the kidneys is believed to be underestimated because of the partial-volume effect. The aim of this work was to develop an accurate, robust and practical method to correct for the partial-volume effect in 177 Lu SPECT imaging. Material and Methods: An image-based method for partial-volume correction (PVC) was developed for the current application, based on the template projection – reconstruction method. In this method, a number of structures with expected homogenous activity uptake are outlined on an anatomical or functional tomographic image. All voxels within a particular structure is assigned a value of unity, to form a template. This template is then projected into a set of planar images with an analytical projector, at angles corresponding to those used in SPECT acquisition. The projector models the effects of attenuation and distance-dependent resolution, which are thus included in the projection images. After reconstruction of the projection images, voxel-specific correction factors are obtained and used in an iterative algorithm to perform PVC on the SPECT image. The templates were reconstructed using a perturbation based method. The developed PVC method was evaluated in a Monte Carlo simulation study with SIMIND, using the XCAT phantom to generate source distributions resembling a patient being treated with 177 Lu-Tyr3-Octreotate at four different times postinjection. Additional testing of the PVC algorithm and reconstruction methods for the templates was performed with a digital geometrical phantom. Finally, the PVC method was applied on four clinical images to assure its practical functionality and compare the results with the results from the phantom study. Results: In absence of noise and scattered photons, a nearly perfect correction was achieved for the geometrical phantom when the perturbation based reconstruction was used for the templates. In this case the source distribution and templates were perfectly spatially aligned and projected using the same projector. In the XCAT phantom study, the PVC reduced the error in kidney activity concentration from approximately -20 % to -1 % with templates acquired from the true organ configuration in the phantom. To study a more realistic scenario, the kidney VOI and template was also determined by automatic segmentation of the simulated SPECT image. In this case, the error was also considerably improved, from approximately -22 % to 2 % with PVC. For the clinical images, the count levels in the kidneys were in all four images approximately 5-10 % higher after PVC than the count levels obtained with the standard clinical method. The results were however sensitive to the outlining of the kidney VOI used to create the template. Similar results were obtained with both manual and automatic segmentation, as long as the shape and volume of the VOI corresponded to the anatomical shape of the kidney with reasonable accuracy. Conclusions: A robust and practical method to correct for the partial-volume effect in 177 Lu SPECT imaging has been developed and evaluated. The method requires outlining of a number of structures with expected homogenous activity uptake. It was shown to perform well in a simulation study, and considerably reduced the bias in estimated activity concentration. Further evaluation is needed to investigate the sensitivity of non-homogenous activity uptake within the defined structures and if the number of outlined structures can be reduced with preserved quantitative accuracy. Physical phantom measurements should also be performed to validate the method. The method can easily be extended and used for other applications of quantitative SPECT. Gustav Brolin MSc dissertation