1. New Hybrid Monochromatic CT Images Achieve High Suppression of Metallic Artifacts While Preserving Soft Tissue and Iodine Contrast
Michal Kulon, Peter Komlosi
Department of Radiology and Medical Imaging
University of Virginia
Charlottesville, VA

2. No conflict of interest to disclose.
DISCLOSURES
No conflict of interest to disclose.

3. PURPOSE
To evaluate the effects of novel algorithm that uses adaptive, voxel-by- voxel blending of high and low-energy monochromatic CT images, with the aim of producing a single image set that:
Suppresses metallic streak artifacts, similar to high-energy images.
Retains good contrast between soft tissues and iodinated-contrast, similar to low-energy images.
Background:
Spinal metallic hardware reduces the quality of CT imaging because of beam hardening, photon starvation and scatter artifacts.
Dual energy CT has been previously proposed as a means to reduce metal artifacts via generating high-energy mono-energetic images.
However, high-energy images generally have relatively poor soft tissue and iodine contrast.

4. MATERIALS AND METHODS (1 of 2)
IRB and HIPAA compliant study with retrospective use of images.
Acquisition of cervical spinal CT myelographic data with a fast-kV switching dual-energy CT at 80 and 140 kVp from a single patient with anterior metallic fusion hardware, and reconstructed at 70-keV and 130-keV mono- energetic image datasets.
Custom software generated hybrid images (C++, DCMTK)
Each output voxel (V) is a weighted average of the 2 corresponding voxels from the 70-keV and 130-keV mono-energetic images
VOutput = w ∙ VLowEnergy + (1 - w) ∙ VHighEnergy
The weighting factor (w, range 0 to 1) is calculated individually for each output voxel, and favors the low-energy input voxel, unless the voxel is likely to be degraded by artifact, in which case the high-energy input voxel is favored. The likelihood and magnitude of artifact are considered to be increasing (leading to decrease in w) if:
The voxel is closer to other voxels containing metal (>1100 HU).
Attenuation value is significantly lower within the 70 keV voxel than within 130 keV voxel, implying an area of dark photon-starvation streak artifact.

5. MATERIALS AND METHODS Image Evaluation
The difference between the voxel intensities (Housfield Units) in contrast- filled thecal sac and muscle was used to measure iodine/soft-tissue contrast.
Blinded review and ranking of the images was performed by radiology trainees (n = 24) based on general appearance with attention to:
Conspicuity of intrathecal contrast material.
Severity of metallic streak artifacts
Contrast of soft tissues and bone.

6. Representative cervical myelographic CT images
Hybrid Algorithm
Low Energy (70 keV)
High Energy (130 keV)
Hybrid images aim to reduce metallic artifacts and retain contrast between soft tissues and iodine

7. Weighting Factor Calculation Ranging from 0 (black) to 1 (white)
Weighting Factors (W)
Ranging from 0 (black) to 1 (white)
Low Energy (70 keV)
Favored when W is near 1
High Energy (130 keV)
Favored when W is near 0

8. Reviewer Preferences Top Choice Among Reviewers (%) Iodine Conspicuity
Difference in Hounsfield Units between Thecal Sac and Muscle
Hybrid
62.5
472
70-keV
8.3
444
130-keV
29.2
148
CONCLUSION: Hybrid images were the most common top-choice among raters, followed by high-energy, followed by low-energy images.

9. Reviewer Preferences - Condorcet Method
Condorcet voting method uses ranking of choices.
Generally superior to pick-top-choice methods, especially in the presence of similar choices or more than 2 choices.
Comparison
(A vs. B)
Number of Raters Ranking A above B
(A : B)
Percentage of Raters
Favoring A over B
%, [Confidence-Interval]
Hybrid vs. 130-keV
17:7
70.8 [ ]
Hybrid vs. 70-keV
21:3
87.5 [ ]
130-keV vs. 70-keV
16:8
66.6 [ ]
Raters preferred the hybrid images to either low or high energy images.
Raters preferred the high-energy images to the low-energy images.

10. CONCLUSION
Most readers preferred the new hybrid algorithm over either a high or low energy reconstruction.
Metallic streak artifacts were most conspicuous on 70-keV images.
Visualization of iodinated contrast was worst on 130-keV images.