SU-E-I-96: Realistic Synthetic CT Imaging of Small Numerical Lesions with Simple CT Simulation Model Incorporating...

Quantitative analysis in CT imaging often requires image data set of lesions with ground truth, which is practically impossible in general. This study presents a simple CT simulation model which enables to create realistic synthetic CT images of numerical voxel phantoms for small lesionswith various shape and sizes.

Basic CT parameters were acquired from DICOM header of target CT data, and were used in the CT simulator. Frequency response of forward and backward projectors were determined by measuring NPS of reconstructed image from white Gaussian noise input data, which were then used to iteratively calibrate those projectors so as to produce white Gaussian NPS output. Then, the MTF of target equipment were measured and incorporated into the sinogram filter before back projection in order to reflect focal spot blurring and vendor specific recon kernel. For validation, physical and numerical phantoms with rod and tube shape of 2mm ∼ 10mm were created and imaged with a real CT (Sensation 16, Siemens) and the proposed simulated CT, respectively. Images were reconstructed with 8 different kernels (B10 ∼ B80) in both CT imaging technique, and their line profiles were compared.

Synthesized and physically scanned CT images of phantoms with different shape, sizes, and recon kernels agreed well in visual assessment, size measurement, and line profile comparison.

Proposed method could produce realistic synthetic CT images of numerical phantoms highly mimicking in visual appearance and quantitative comparison, which has a potential for use as ground truth data set in quantitative image analysis.