Encrypted login | home

Program Information

Development and Evaluation of Iterative CT Image Reconstruction in Cylindrical Image Coordinate

W Rao

W Rao1,2*, R Pham1 , P Kandlakunta1 , M Anastasio1 , T Zhang1 , (1) Department of Radiation Oncology, Washington University, St. Louis, MO, 63110, USA; (2) State Key Laboratory of Laser Propulsion and Applications, Academy of Equipment, Beijing, 101416, China


SU-F-201-2 (Sunday, July 30, 2017) 2:05 PM - 3:00 PM Room: 201

Purpose: To develop and evaluate a fast iterative reconstruction (IR) that employs a precomputed system matrix in cylindrical image coordinate.

Methods: The system matrix of IR in cylindrical image coordinate is rotationally invariant. Thus its dimension is greatly smaller than that of conventional Cartesian image grid. We derived analytic formulae to calculate the system matrix exactly. Images are reconstructed with standard Simultaneous Algebraic Reconstruction Technique (SART) in cylindrical image coordinate, then interpolated to Cartesian grid for visualization and evaluation. For comparison, the same images were also reconstructed with conventional SART in Cartesian image grid and filtered back projection (FBP) algorithms. A simulated human body phantom (XCAT2) was used to generate projection data with Poisson noises of different levels. Three figures of merit (FOM) referring to the original image were employed to quantify the noises in reconstructed images. Another digital phantom with round patterns in different sizes, locations, and contrast was used to analyze the spatial resolution. The modulation transfer functions (MTF) were calculated from the fitting curves of the contrast-step edges. The area under MTF curve up to 0.5 mm-1 was calculated as a quantitative value of edge resolution.

Results: Images reconstructed by two SART methods show no significant difference in noise level; both were less noisy than the FBP method. The Cartesian method yields a homogeneous spatial resolution, whereas the spatial resolution by the cylindrical method is dependent on the locations and orientation.

Conclusion: IR can be greatly accelerated with the precomputed system matrix in cylindrical image grid. The noise level in reconstructed images is not affected by using different image matrices, but the spatial resolution for the cylindrical method is inhomogeneous and anisotropic. The grid resolution of cylindrical image grid should be carefully optimized to match the intrinsic limitation of the CT systems.

Contact Email: