Fast, automated optimization of virtual monoenergetic images with the dual-energy image synthesizer for cone-beam CT

Background

Dual-energy cone-beam CT (DE-CBCT) has become subject of recent interest due to the ability to produce virtual monoenergetic images (VMIs) with improved soft-tissue contrast and reduced nonuniformity artifacts. However, efficient production and optimization of VMIs remains an under-explored part of DE-CBCT's application.

Purpose

This work reports on the creation of DISC (dual-energy image synthesizer for CBCT), a newly developed, open-source user interface to efficiently produce and optimize VMIs with the eventual goal of clinical application.

Methods

Two sets of CBCT scans of a Catphan 604 phantom were acquired sequentially (80 and 140 kVp) using the on-board imager of a commercial linear accelerator. Material decomposition into aluminum (Al) and polymethyl-methylacrylate (PMMA) basis materials in the projection-domain and reconstruction with the Feldkamp–Davis–Kress (FDK) algorithm of basis material images were performed in the open-source Tomographic Iterative GPU-based REconstruction (TIGRE) Matlab toolkit. Using DISC, a series of VMIs were generated via linear combinations of the basis material images without reconstructing individual VMIs at different energies. Hounsfield units (HU) were computed using an energy-dependent fit over the range of 20–150 keV. VMI energies that maximized contrast-to-noise ratio (CNR) for various materials and minimized nonuniformity artifacts were determined with 1 keV precision.

Results

Optimal CNR values for all material inserts ranged from 55 to 62 keV, showing an average CNR enhancement of 25% over the polychromatic images. Optimal uniformity is observed at 65 keV. Computed HUs show good agreement with theoretical values, with root-mean-squared error of 16 HU across the range of energies and materials.

Conclusion

A spectrum of VMIs from DE-CBCT was efficiently produced with 1 keV precision using DISC. Optimal energies for both soft tissue contrast and nonuniformity reduction were quickly computed with high precision. Future work will expand DISC to generate other DE-derived image types and will explore the acquisition and optimization of DE patient images.