Synthesizing High-Resolution Dual-Energy Radiographs from Coronary Artery Calcium CT Images.

Abstract

Generating realistic radiographs from CT is mainly limited by the native spatial resolution of the latter. Here we present a general approach for synthesizing high-resolution digitally reconstructed radiographs (DRRs) from an arbitrary resolution CT volume. Our approach is based on an upsampling framework where tissues of interest are first segmented from the original CT volume and then upsampled individually to the desired voxelization (here ~1 mm → 0.2 mm). Next, we create high-resolution 2D tissue maps by cone-beam projection of individual tissues in the desired radiography direction. We demonstrate this approach on a coronary artery calcium (CAC) patient CT scan and show that our approach preserves individual tissue volumes, yet enhances the tissue interfaces, creating a sharper DRR without introducing artificial features. Lastly, we model a dual-layer detector to synthesize high-resolution dual-energy (DE) anteroposterior and lateral radiographs from the patient CT to visualize the CAC in 2D through material decomposition. On a general level, we envision that this approach is valuable for creating libraries of synthetic yet realistic radiographs from corresponding large CT datasets.