Restoration of Thickness, Density, and Volume for Highly Blurred Thin Cortical Bones in Clinical CT Images

In clinical CT images containing thin osseous structures, accurate definition of the geometry and density is limited by the scanner's resolution and radiation dose. This study presents and validates a practical methodology for restoring information about thin bone structure by volumetric deblurring of images. The methodology involves 2 steps: a phantom-free, post-reconstruction estimation of the 3D point spread function (PSF) from CT data sets, followed by iterative deconvolution using the PSF estimate. Performance of 5 iterative deconvolution algorithms, blind, Richardson-Lucy (standard, plus Total Variation versions), modified residual norm steepest descent (MRNSD), and Conjugate Gradient Least-Squares were evaluated using CT scans of synthetic cortical bone phantoms. The MRNSD algorithm resulted in the highest relative deblurring performance as assessed by a cortical bone thickness error (0.18 mm) and intensity error (150 HU), and was subsequently applied on a CT image of a cadaveric skull. Performance was compared against micro-CT images of the excised thin cortical bone samples from the skull (average thickness 1.08 ± 0.77 mm). Error in quantitative measurements made from the deblurred images was reduced 82% (p < 0.01) for cortical thickness and 55% (p < 0.01) for bone mineral mass. These results demonstrate a significant restoration of geometrical and radiological density information derived for thin osseous features.