Fast analytical approach of application specific dose efficient spectrum selection for diagnostic CT imaging and PET attenuation correction

Abstract

Computed tomography (CT) has been used for a variety of applications, two of which include diagnostic imaging and attenuation correction for PET or SPECT imaging. Ideally, the x-ray tube spectrum should be optimized for the specific application to minimize the patient radiation dose while still providing the necessary information. In this study, we proposed a projection-based analytic approach for the analysis of contrast, noise, and bias. Dose normalized contrast to noise ratio (CNRD), inverse noise normalized by dose (IND) and bias are used as evaluation metrics to determine the optimal x-ray spectrum. Our simulation investigated the dose efficiency of the x-ray spectrum ranging from 40 kVp to 200 kVp. Water cylinders with diameters of 15 cm, 24 cm, and 35 cm were used in the simulation to cover a variety of patient sizes. The effects of electronic noise and pre-patient copper filtration were also evaluated. A customized 24 cm CTDI-like phantom with 13 mm diameter inserts filled with iodine (10 mg ml^-1), tantalum (10 mg ml^-1), water, and PMMA was measured with both standard (1.5 mGy) and ultra-low (0.2 mGy) dose to verify the simulation results at tube voltages of 80, 100, 120, and 140 kVp. For contrast-enhanced diagnostic imaging, the simulation results indicated that for high dose without filtration, the optimal kVp for water contrast is approximately 100 kVp for a 15 cm water cylinder. However, the 60 kVp spectrum produces the highest CNRD for bone and iodine. The optimal kVp for tantalum has two selections: approximately 50 and 100 kVp. The kVp that maximizes CNRD increases when the object size increases. The trend in the CTDI phantom measurements agrees with the simulation results, which also agrees with previous studies. Copper filtration improved the dose efficiency for water and tantalum, but reduced the iodine and bone dose efficiency in a clinically-relevant range (70-140 kVp). Our study also shows that for CT-based attenuation correction applications for PET or SPECT, a higher-kVp spectrum with copper filtration is preferable. This method is developed based on filter back projection and does not require image reconstruction or Monte Carlo dose estimates; thus, it could potentially be used for patient-specific and task-based on-the-fly protocol optimization.

Figure 1

(A) In-house manufactured 24 cm CTDI-like phantom with contrast material insert. (B) Example cross-section image for a scan of the phantom with iodine insert in the center. Display window is 40 + [−400 400] HU. (C) ROI regions to be applied to images for measurement analysis, indicated by ROI1 (blue): 10-mm diameter circular ROI centered on the contrast sample vial; ROI2 (purple): inner background annulus ROI located closely surrounding the contrast sample with 20-mm inside diameter and 40-mm outside diameter; ROI3 (green): outer background ROI located in the uniform region of the phantom away from center with 130-mm inside diameter and 150-mm outside diameter.

Figure 2

Simulation results for the normalized CNRD curves with different object sizes. These data do not include electronic noise; therefore, the dose levels do not impact the simulation results. Water, bone, iodine and tantalum were evaluated. (A), (B), and (C) show the CNRD curves for 15-cm, 24-cm, and 35-cm water cylinders respectively. The CNRD curves for each contrast material are normalized to the maximum CNRD value without filtration for that contrast material.

Figure 3

Simulated CNRD curves at three different dose levels for (A) water, (B) bone, (C) iodine and (D) tantalum, using a 24-cm water cylinder with electronic noise. The CNRD curves for each contrast material are normalized to the maximum CNRD value without filtration at dose-100mA for that contrast material.

Figure 4

Simulated IND curves for three different dose levels with and without 1-mm Cu filter, for (A) 15-cm, (B) 24-cm, and (C) 35-cm water cylinders. Electronic noise is included in the simulation. The IND curves are normalized to the maximum IND value without filtration at dose100mA.

Figure 5

Bias (% error) in the 511 keV projection. (A) is for analysis with and without electronic noise. (B) is for analysis with and without 1mm Cu filter, including electronic noise. X-ray quantum noise is included in all simulations.

Figure 6

Images from scans of a 24-cm CTDI-like phantom using an iodine insert at two equal dose levels for different kVp settings. The top row shows the images from the ultra-low-dose scans. The bottom row shows the images from the standard-dose scans. From left to right are the images acquired using 80, 100, 120, 140 kVp. The display window is [−100 300] HU.

Figure 7

Images from scans of a 24-cm CTDI-like phantom using a tantalum insert at two equal dose levels for different kVp settings. The top row shows the images from the ultra-low-dose scans. The bottom row shows the images from the standard-dose scans. From left to right are the images acquired using 80, 100, 120, 140 kVp. The display window is [−100 300] HU.

Figure 8

Images from scans of a 24-cm CTDI-like phantom using a PMMA insert at two equal dose levels for different kVp settings. The top row shows the images from the ultra-low-dose scans. The bottom row shows the images from the standard-dose scans. From left to right are the images acquired using 80, 100, 120, 140 kVp. The display window is [−100 300] HU.

Figure 9

Images from scans of a 24-cm CTDI-like phantom using a water insert at two equal dose levels for different kVp settings. The top row shows the images from the ultra-low-dose scans. The bottom row shows the images from the standard-dose scans. From left to right are the images acquired using 80, 100, 120, 140 kVp. The display window is [−100 300] HU.

Figure 10

CNRD from scans using a 24-cm CTDI-like phantom with iodine, tantalum water and PMMA inserts, scanned at equal standard dose.

Figure 11

IND (A) and bias (B) analysis from scans of a 24-cm CTDI-like phantom with a different contrast material insert, at equal ultra-low dose.

Figure 12

Monte Carlo simulation results. Image (A) shows a schematic overview of the Monte-Carlo simulation using 120 incident X-ray photons. Image (B) shows the probability distribution of the total energy deposited to the detector. The probability distribution obtained using Monte-Carlo simulation with 1 million photons is compared to the numerical model used in our analytical method at two dose levels.