Dual energy CT: How well can pseudo‐monochromatic imaging reduce metal artifacts?

• Published in: Medical physics (Lancaster) Vol. 42; no. 2; pp. 1023 - 1036
• Main Authors: Kuchenbecker, Stefan, Faby, Sebastian, Sawall, Stefan, Lell, Michael, Kachelrieß, Marc
• Format: Journal Article
• Language: English
• Published: United States American Association of Physicists in Medicine 01-02-2015
• Subjects: 60 APPLIED LIFE SCIENCES; ALGORITHMS; Biological material, e.g. blood, urine; Haemocytometers; Biomedical imaging; Computed tomography; Computerised tomographs; computerised tomography; CONTRAST MEDIA; DECT; Digital computing or data processing equipment or methods, specially adapted for specific applications; Humans; Image data processing or generation, in general; IMAGE PROCESSING; Image reconstruction; Image restoration; Medical image artifacts; Medical image noise; medical image processing; Medical image reconstruction; Medical image segmentation; metal artifact reduction (MAR); METALS; monochromatic imaging; monoenergetic imaging; Nonlinear Dynamics; PATIENTS; PHANTOMS; Prostheses implantable into the body; prosthetics; RADIOLOGY AND NUCLEAR MEDICINE; REDUCTION; Signal-To-Noise Ratio; Tissues; Tomography, X-Ray Computed - methods
• ISSN: 0094-2405; 2473-4209
• DOI: 10.1118/1.4905106

Purpose:
Dual Energy CT (DECT) provides so‐called monoenergetic images based on a linear combination of the original polychromatic images. At certain patient‐specific energy levels, corresponding to certain patient‐ and slice‐dependent linear combination weights, e.g., E = 160 keV corresponds to α = 1.57, a significant reduction of metal artifacts may be observed. The authors aimed at analyzing the method for its artifact reduction capabilities to identify its limitations. The results are compared with raw data‐based processing.
Methods:
Clinical DECT uses a simplified version of monochromatic imaging by linearly combining the low and the high kV images and by assigning an energy to that linear combination. Those pseudo‐monochromatic images can be used by radiologists to obtain images with reduced metal artifacts. The authors analyzed the underlying physics and carried out a series expansion of the polychromatic attenuation equations. The resulting nonlinear terms are responsible for the artifacts, but they are not linearly related between the low and the high kV scan: A linear combination of both images cannot eliminate the nonlinearities, it can only reduce their impact. Scattered radiation yields additional noncanceling nonlinearities. This method is compared to raw data‐based artifact correction methods. To quantify the artifact reduction potential of pseudo‐monochromatic images, they simulated the FORBILD abdomen phantom with metal implants, and they assessed patient data sets of a clinical dual source CT system (100, 140 kV Sn) containing artifacts induced by a highly concentrated contrast agent bolus and by metal. In each case, they manually selected an optimal α and compared it to a raw data‐based material decomposition in case of simulation, to raw data‐based material decomposition of inconsistent rays in case of the patient data set containing contrast agent, and to the frequency split normalized metal artifact reduction in case of the metal implant. For each case, the contrast‐to‐noise ratio (CNR) was assessed.
Results:
In the simulation, the pseudo‐monochromatic images yielded acceptable artifact reduction results. However, the CNR in the artifact‐reduced images was more than 60% lower than in the original polychromatic images. In contrast, the raw data‐based material decomposition did not significantly reduce the CNR in the virtual monochromatic images. Regarding the patient data with beam hardening artifacts and with metal artifacts from small implants the pseudo‐monochromatic method was able to reduce the artifacts, again with the downside of a significant CNR reduction. More intense metal artifacts, e.g., as those caused by an artificial hip joint, could not be suppressed.
Conclusions:
Pseudo‐monochromatic imaging is able to reduce beam hardening, scatter, and metal artifacts in some cases but it cannot remove them. In all cases, the CNR is significantly reduced, thereby rendering the method questionable, unless special post processing algorithms are implemented to restore the high CNR from the original images (e.g., by using a frequency split technique). Raw data‐based dual energy decomposition methods should be preferred, in particular, because the CNR penalty is almost negligible.