Room: Track 1
Purpose: The finite size of x-ray focal spots is a fundamental limitation for spatial resolution in x-ray imaging. Image blur due to the focal spot is depth dependent. In this work, we demonstrate that the depth dependence of blur is minimal in common cone beam CT (CBCT) geometries and resolution can be recovered using the iterative Richardson-Lucy (RL) deconvolution algorithm.
Methods: A simulation study was conducted using a digital breast phantom with five 100 µm calcifications placed at different depths. A typical breast CT geometry (SID=923 cm, SDD=650 cm) was used. A pinhole image of a focal spot (0.3 mm) was used as a two-dimensional source model to generate projections with Poisson noise. Images were reconstructed with the FDK algorithm and ramp filter. Iterative RL deblurring was evaluated in both the projection and image domains, using a Gaussian-fitted focal spot and PSF at the phantom center as deconvolution kernels, respectively. The contrast to noise ratio (CNR) for five calcifications was compared for images generated using a point source, finite source, and images deblurred by projection and image-domain deconvolution. The method was also tested on real breast and knee CBCT acquisitions to see the effect of focal spot deconvolution.
Results: Calcification CNRs at different depths vary within 8% when using a finite source, similar to the results using a point source. Compared to the point source, the average calcification CNR decreased by 12%. At the same noise level, RL deblurring in the projection and image domain increased the calcification CNR by 8.51% and 70.2%, respectively. Qualitatively, the proposed method greatly improved the resolution of breast calcifications and trabecular bone structures (knee).
Conclusion: The work demonstrates that the depth dependence of focal spot blur is negligible for common CBCT geometries and system resolution can be substantially improved using iterative focal spot deconvolution.