Evaluation of three-dimensional iterative image reconstruction in virtual monochromatic imaging at 40 kilo-electron volts: phantom and clinical studies to assess the image noise and image quality in comparison with other reconstruction techniques

Abstract

Objective:

The purpose of this study was to evaluate the image quality in virtual monochromatic imaging (VMI) at 40 kilo-electron volts (keV) with three-dimensional iterative image reconstruction (3D-IIR).

Methods:

A phantom study and clinical study (31 patients) were performed with dual-energy CT (DECT). VMI at 40 keV was obtained and the images were reconstructed using filtered back projection (FBP), 50% adaptive statistical iterative reconstruction (ASiR), and 3D-IIR. We conducted subjective and objective evaluations of the image quality with each reconstruction technique.

Results:

The image contrast-to-noise ratio and image noise in both the clinical and phantom studies were significantly better with 3D-IIR than with 50% ASiR, and with 50% ASiR than with FBP (all, p < 0.05). The standard deviation and noise power spectra of the reconstructed images decreased in the order of 3D-IIR to 50% ASiR to FBP, while the modulation transfer function was maintained across the three reconstruction techniques. In most subjective evaluations in the clinical study, the image quality was significantly better with 3D-IIR than with 50% ASiR, and with 50% ASiR than with FBP (all, p < 0.001). Regarding the diagnostic acceptability, all images using 3D-IIR were evaluated as being fully or probably acceptable.

Conclusions:

The quality of VMI at 40 keV is improved by 3D-IIR, which allows the image noise to be reduced and structural details to be maintained.

Advances in knowledge:

The improvement of the image quality of VMI at 40 keV by 3D-IIR may increase the subjective acceptance in the clinical setting.

Introduction

Dual-energy CT (DECT), which allows virtual monochromatic imaging (VMI), has become a widely used imaging modality in clinical practice. ^1–3 Images in VMI appear as if the imaging is performed with a monochromatic beam, and can be reconstructed over a wide range of kilo-electron volt (keV) levels, from 40 to 140 keV. ^1–3 The energy levels for the imaging can be varied according to the clinical application. Lower energies are used to increase iodine contrast, whereas higher energies are used to reduce metal artifacts. ^1–4 A previous study reported that VMI at approximately 70 keV yielded a lower image noise and higher iodine contrast-to-noise ratio (CNR) than conventional CT at 120 kVp, and could replace 120 kVp CT as the standard CT imaging modality. ^2,3,5 In choosing the appropriate energy level, there needs to be a tradeoff between the iodine contrast and image noise. ^1–3,6 Higher iodine contrast can be obtained with VMI at lower keV. ^1–3,6 Recently, several reports have demonstrated the usefulness of VMI at 40 keV, especially for detecting low-contrast lesions. ^7–14 In addition, VMI at 40 keV has been utilized in routine clinical practice as a supplement to imaging at the standard energy level, to obtain additional useful information. ¹⁴ However, VMI at low keV values is associated with a marked increase in image noise, which leads to deterioration of the image quality. To decrease the effect of such increase in image noise, a higher radiation dose could be required. ^15,16 Therefore, it is desirable to reduce the image noise.

Several iterative reconstruction (IR) algorithms to provide better image quality with less image noise, as compared to filtered back projection (FBP), in CT images have been introduced. ^17,18 Three-dimensional iterative image reconstruction (3D-IIR) uses an algorithm to process DICOM image data and can be used with scanners from any vendor. ^19,20 We compared the images obtained using 3D-IIR with those obtained using adaptive statistical iterative reconstruction (ASiR; GE healthcare, Waukesha, WI) and FBP. To the best of our knowledge, this is the first study to evaluate the feasibility of VMI at 40 keV with image reconstruction using 3D-IIR to obtain images with reduced image noise and improved image quality, with diagnostic acceptability.

Methods

Phantom study to evaluate image noise and low-contrast resolution

We scanned a Catphan TM 500 phantom (The Phantom Laboratory, Greenwith, NY) to evaluate the CT image noise and CNR, and the noise power spectra (NPS) in the images, in the dual-energy scan mode with Discovery CT750 HD (GE Healthcare, Waukesha, WI). The CT imaging parameters were as follows: collimation, 64 × 0.625 mm; pitch, 0.984; gantry rotation, 1 s; scan field of view, 50 cm; tube voltage, fast kilovoltage switching of 80 and 140 kVp. The volume CT dose index was fixed at 35 mGy and the tube current was selected automatically according to the volume CT dose index. The images were reconstructed at 1.25 mm thickness and 1.25 mm intervals with a standard reconstruction kernel.

VMI at 40 keV was obtained and the images were reconstructed using both FBP and 50% ASiR blended with 50% FBP (50% ASiR). Images reconstructed using FBP were transferred to a 3D-IIR server (SafeCT; Medic Vision, Tirat Carmel, Israel) for image-based iterative reconstruction.

To measure the image noise and calculate the CNR, regions of interest (ROIs), circular in shape and 10 mm in diameter were placed on a supra slice 15 mm diameter target of 1.0% contrast on 10 consecutive reconstructed images of the low-contrast module (CTP515) of the Catphan phantom (Figure 1a). Image noise was defined as the standard deviation (SD) of the CT value in the 10 mm circular ROI placed in the background.

Figure 1.

Figures demonstrating the phantom with example ROIs. The low-contrast module (CTP515) of the Catphan phantom (a) was used to measure the CT value and image noise, and calculate the CNR. ROIs 1 and 2, circular in shape and 10 mm each in diameter were placed on a 15 mm diameter supra slice target of 1.0% contrast and on a background, respectively. To calculate the NPS, the uniformity module (CTP486) of the Catphan phantom (b) was utilized. The acrylic phantom shaped like an elliptical cylinder (c) was scanned to calculate the MTFs using the circular edge method, targeting the hole containing the diluted contrast medium (5 mg/ml). CNR, contrast-to-noiseratio; MTF, modulation transfer functions; NPS, noise power spectra; ROI, region of interest.

CNR was calculated as CNR = (μ_A – μ_B) /ρ, where μ_A is the CT value of the supra slice 15 mm diameter target of 1.0% contrast, μ_B is that of the background, and ρis the image noise.

We calculated the NPS to evaluate the magnitude and spatial frequency characteristics of the image noise using the radial frequency method. We used 10 consecutive images of the uniformity module (CTP486) of the Catphan phantom (Figure 1b) for each reconstruction method and calculated the NPS for each image. The average value for 10 spectra was calculated to minimize the statistical error.

Phantom study to evaluate spatial resolution

To evaluate the spatial resolution, we calculated the modulation transfer functions (MTFs), targeting the hole containing diluted contrast medium (5 mg/ml) through the center of the elliptical cylinder acrylic phantom (Figure 1c). The scanning conditions were the same as those mentioned above. The images were reconstructed at 1.25 mm thickness and 1.25 mm intervals with a standard reconstruction kernel. For this evaluation, we used 25 consecutive axial images for each of the three reconstruction techniques. To decrease the effect of noise, we averaged every five consecutive axial images for each of the reconstruction techniques and calculated the MTFs on the averaged images using the circular edge method.

Clinical study

This retrospective single-center study was conducted with the approval of the Institutional Review Board of Tokyo women’s medical university, with a waiver obtained for obtaining informed consent. We conducted a retrospective search of our CT database for consecutive patients who underwent CT examination by the dual-energy scan mode for adrenal tumor, renal tumor or bladder tumor between October 21, 2011, and July 5, 2017. The study population consisted of 31 patients (mean age, 60.1), including 17 males and 14 females. The average body weight was 61.1 kg.

All patients were scanned in the Discovery CT750 HD system. Iodine contrast medium was injected at the dose of 600 mg/kg over a period of 30 s. The bolus tracking technique was used to determine the scan timing. The scanning conditions were the same as those mentioned above, except for the following parameters: pitch, 1.375; gantry rotation, 0.7 s. The tube current was automatically changed with the Auto Exposure Control with a 12 CT noise index. The median CTDIvol was 11.32 mGy (9.9–13.33 mGy). These images were reconstructed at 1.25 mm thickness and intervals with a standard reconstruction kernel. VMI at 40 keV was obtained with image reconstruction by FBP, 50% ASiR and 3D-IIR.

Subjective evaluation

Two radiologists who were blinded to the image reconstruction methods evaluated the axial images of 31 patients reconstructed by the 3 methods. The images were displayed on the workstation (Advantage Workstation v. 4.6, GE Healthcare) with an appropriate window setting (window width, 800 HU; window level, 80 HU). The readers used 5-point scales to grade the subjective noise, artifacts, visualization of small or thin structures (common bile duct, adrenal glands and small vessels), visualization of the intrahepatic vasculature (portal vein and hepatic vein), and the diagnostic acceptability. For subjective noise and artifacts, the scoring was as follows: 5, minimal; 4, mild; 3, moderate; 2, marked but acceptable; 1, unacceptable. For visualization of small or thin structures, and visualization of the intrahepatic vasculature, the scoring was as follows: 5, completely defined; 4, almost defined; 3, slightly irregular; 2, quite irregular but acceptable; 1, unacceptable. For diagnostic acceptability, the scoring was as follows: 5, fully acceptable; 4, probably acceptable; 3, possibly acceptable; 2, only acceptable under limited conditions; 1, unacceptable.

Objective evaluation

A circular ROI measuring 10 mm in diameter was placed on the aorta (at the level of the orifice of the SMA) and the right psoas muscle to calculate the image noise and CNR.

Statistical analysis

We used analyses of variance to test the CT value, CNR and image noise measured on the Catphan phantom and clinical images. A p-value of less than 0.05 was considered as indicative of a statistically significant difference.Statistical analyses were performed using SPSS (SPSS statistics 25; IBM, Chicago, IL).

For subjective image analysis, we used Wilcoxon’s signed rank test corrected for multiple comparisons according to Bonferroni adjustment, to test the significance of differences in image rating. We considered a p-value of less than 0.0167 as being indicative of a statistically significant difference. Cohen weighted κ with linear weights was calculated to evaluate interobserver agreement for each subjective image rating. Values of 0–0.20 were considered to represent slight agreement, values of 0.21–0.40 as representing fair agreement, values of 0.41–0.60 as representing moderate agreement, values of 0.61–0.80 as representing good agreement; and values of 0.81–1.00 as representing almost perfect agreement.

Results

Table 1 summarizes the CNR and image noise in the phantom study. The CT values did not differ significantly among the three reconstruction methods. The CNR was significantly higher in the images reconstructed by 3D-IIR than in those reconstructed by 50% ASiR, and in images reconstructed by 50% ASiR than in those reconstructed by FBP (all, p < 0.001). The image noise was significantly smaller in images reconstructed by 3D-IIR than in those reconstructed by 50% ASiR (p < 0.026), and in images reconstructed by using 50% ASiR than in those reconstructed by FBP (p < 0.039).

Table 1.

CT value, contrast-to-noise ratio and image noise in the phantom study

	FBP	50% ASiR	3D-IIR
CT value (HU)	−35.7 ± 4.1	−36.4 ± 3.8	−34.9 ± 3.5
CNR	0.74 ± 0.21	1.23 ± 0.38	1.61 ± 0.91
Noise (HU)	15.7 ± 1.9	8.8 ± 0.9	6.0 ± 0.3

HU, Hounsfield unit; CNR, contrast-to-noise ratio; FBP, filtered back projection; ASiR, adaptive statistical iterative reconstruction; 3D-IIR, three-dimensional iterative image reconstruction

CT value, contrast-to-noise ratio and image noise are represented as the averages ± standard deviation.

According to the Figure 2, which shows the NPS curve, the image noise was low across the entire spatial frequency in images reconstructed by 3D-IIR than in those reconstructed by 50% ASiR, and in images reconstructed by 50% ASiR than in those reconstructed by FBP.

Figure 2.

Noise power spectra in the images obtained using the three reconstruction techniques. The magnitude of the image noise reduced uniformly across the spatial frequency spectrum in the order of 3D-IIR to 50% ASiR to FBP. 3D-IIR, three-dimensional iterative image reconstruction; ASiR, adaptive statistical iterative reconstruction; FBP, filtered back projection

Figure 3 illustrates the MTF for each reconstruction technique. The spatial resolution was almost the same for the images reconstructed using FBP, 50% ASiR and 3D-IIR.

Figure 3.

The MTF curves obtained using the three reconstruction techniques. The MTF curves were almost the same in the three reconstruction techniques. 3D-IIR, three-dimensional iterative image reconstruction; ASiR, adaptive statistical iterative reconstruction; FBP, filtered back projection; MTF, modulation transfer function.

The results of quantitative analysis in the clinical study are shown in Table 2. The CT values did not significantly differ among the three reconstruction methods. The CNR was significantly higher in the images reconstructed by 3D-IIR than in those reconstructed by 50% ASiR, and in the images reconstructed by 50% ASiR than in those reconstructed by FBP (all, p < 0.001). The image noise was significantly smaller in the images reconstructed by 3D-IIR than in those reconstructed by 50% ASiR, and in the images reconstructed by 50% ASiR than in those reconstructed by FBP (all, p < 0.001).

Table 2.

CT value, contrast-to-noise ratio and image noise in the clinical study

	FBP	50% ASiR	3D-IIR
CT value (HU)	443.6 ± 57.9	437.2 ± 62.4	440.5 ± 61.6
CNR	6.55 ± 1.76	11.72 ± 2.89	18.43 ± 4.03
Noise (HU)	52.4 ± 11.5	29.3 ± 6.4	18.7 ± 4.1

HU, Hounsfield unit; CNR, contrast-to-noise ratio; FBP, filtered back projection; ASiR, adaptive statistical iterative reconstruction; 3D-IIR, three-dimensional iterative image reconstruction

CT value, contrast-to-noise ratio and image noise are represented as the averages ± standard deviation.

Table 3 shows the scores in the subjective evaluations. Interobserver agreement between the two radiologists were almost perfect for subjective noise (κ = 0.85), visualization of small or thin structures (κ = 0.86), visualization of the intrahepatic vasculature (κ = 0.89), and diagnostic acceptability (κ = 0.90), and good for artifacts (κ = 0.74).

Table 3.

Scores of subjective evaluations according to the reconstruction methods

	Reader 1a			Reader 2a			Pooled resultsb
	FBP	50% ASiR	3D-IIR	FBP	50% ASiR	3D-IIR	FBP	50% ASiR	3D-IIR
Subjective noise	0/30/1/0/0	0/0/25/6/0	0/0/0/0/31	0/1/29/1/0	0/0/1/30/0	0/0/0/0/31	2.5 ± 0.5	3.6 ± 0.5	5.0 ± 0.0
Artifacts	0/0/6/16/9	0/0/6/16/9	0/0/6/16/9	0/0/2/9/20	0/0/0/2/29	0/0/0/0/31	4.3 ± 0.7	4.5 ± 0.7	4.5 ± 0.7
Visualization of small or thin structures	0/0/14/17/0	0/0/0/10/21	0/0/0/6/25	0/0/10/20/1	0/0/3/28/0	0/0/0/4/27	2.2 ± 0.6	4.3 ± 0.6	4.8 ± 0.4
Visualization of the intrahepatic vasculature	0/1/14/16/0	0/1/8/20/2	0/0/2/10/19	0/0/8/23/0	0/0/4/27/0	0/0/0/5/26	2.2 ± 0.6	3.8 ± 0.5	4.7 ± 0.5
Diagnostic acceptability	0/3/24/4/0	0/1/8/22/0	0/0/0/7/24	0/0/14/17/0	0/0/2/29/0	0/0/0/2/29	1.9 ± 0.5	3.8 ± 0.4	4.9 ± 0.4

FBP, filtered back projection; ASiR, adaptive statistical iterative reconstruction; 3D-IIR, three-dimensional iterative image reconstruction

^aData show the frequencies of scores for each category (Grade 1/2/3/4/5).

^bData show the means ± standard deviation of the two readers’ scores for the 31 patients.

For all the subjective evaluation items other than artifacts, the scores were significantly higher for images reconstructed by 3D-IIR than for the images reconstructed by 50% ASiR (subjective noise, p < 0.001; visualization of small or thin structures, p < 0.001; visualization of the intrahepatic vasculature, p < 0.001; diagnostic acceptability, p < 0.001), 50% ASiR than FBP (subjective noise, p < 0.001; visualization of small or thin structures, p < 0.001; visualization of the intrahepatic vasculature, p < 0.001; diagnostic acceptability, p < 0.001). Figure 4 illustrates the subjective noise and the visualization of the intrahepatic vasculature. Figures 5 and 6 illustrates the visualization of small or thin structures. Figure 7 illustrates the artifacts.

Figure 4.

Abdominal CT images of a 62-year-old female. a, FBP; b, 50% ASiR; c, 3D-IIR

Image noise of the hepatic parenchyma is reduced in the order of c to b to a. In addition, the margins of the intrahepatic vasculature are defined more sharply in c than in b, and in b than. The window settings were as follows: width, 800 HU; level, 80 HU.

3D-IIR, three-dimensional iterative image reconstruction; ASiR, adaptive statistical iterative reconstruction; FBP, filtered back projection

Figure 5.

Abdominal CT images of a 54-year-old male. a, FBP; b, 50% ASiR; c, 3D-IIR. The margin of the adrenal grand is clearer in c than in b, and in b than in a. The window settings were as follows: width, 800 HU; level, 80 HU. 3D-IIR, three-dimensional iterative image reconstruction; ASiR, adaptive statistical iterative reconstruction; FBP, filtered back projection

Figure 6.

Abdominal CT images of a 76-year-old female. a, FBP; b, 50% ASiR; c, 3D-IIR. Margin of the common bile duct (arrow) and the small vessel (arrowhead) are defined better in c than in b, and in b than in a. The window settings were as follows: width, 800 HU; level, 80 HU. 3D-IIR, three-dimensional iterative image reconstruction; ASiR, adaptive statistical iterative reconstruction; FBP, filtered back projection

Figure 7.

Abdominal CT images in a 55-year-old male. a, FBP; b, 50% ASiR; c, 3D-IIR. There are almost no differences in artifacts among the three reconstruction methods. The window settings were as follows: width, 800 HU; level, 80 HU. 3D-IIR, three-dimensional iterative image reconstruction; ASiR, adaptive statistical iterative reconstruction; FBP, filtered back projection

In regard to the diagnostic acceptability, all the images reconstructed by 3D-IIR were evaluated as fully or probably acceptable. On the other hand, approximately 35% of the images reconstructed by 50% ASiR were judged as being possibly acceptable or only acceptable under limited conditions, and approximately 66% the images reconstructed by FBP were evaluated as possibly acceptable or only acceptable under limited conditions.

Both readers assigned a score of more than three for each of the assessed items in the images reconstructed by 3D-IIR. The mean score for each category in the images reconstructed by 3D-IIR was more than 4.5, while that for the images reconstructed by 50% ASiR was less than 4.5, except for artifacts.

Discussion

VMI, which yields images as if they are obtained with a monoenergetic X-ray beam, can be performed with DECT. ^2,3 The energy levels for VMI range from 40 to 140 keV. ^1–3 According to previous studies, VMI at approximately 70 keV is the preferred energy level, yielding a higher CNR, and can replace conventional CT at 120 kVp. ^2,3,5 One of the features of VMI is to increase the iodine contrast at lower monoenergetic energy levels which yield improved detectability of low-contrast lesions. ^1,5,21 However, there is a tradeoff between iodine contrast and image noise, and image noise is also pronounced at lower keV levels. ^1–3,6

Various IR techniques have been developed for CT images. In general, IR offers better image quality with less noise than images obtained using FBP. ^17,18,22 ASiR is an image reconstruction technique which uses information obtained from FBP reconstruction as the initial building block. Post-processing algorithms iteratively reduce the noise in the resultant images and the process is repeated until specific criteria are met. ^22–25 ASiR is usually blended with the FBP image corresponding to an optimal percentage, and is therefore referred to as hybrid IR. ^22,23 Image noise is reduced with increasing the ASiR blending level for reconstruction. ²³ However, reconstruction with a high proportion of ASiR increases the rough texture of the image noise and decreases the spatial resolution. Therefore, 40–50% ASiR is preferable, as a routine. ²³ Whereas most IR techniques, including ASiR, are vendor-specific, 3D-IIR is vendor-neutral and can be used with scanners of any model from any vendor. ^19,20 It yields image data based on a reconstruction algorithm that works on the image space of reconstructed DICOM images. In volumetric iterative image reconstruction, an algorithm for post-scan processing of volumetric scans is employed. The algorithm deconstructs and then reconstructs the scanned volume in 3D space. The proprietary statistical model allows differentiation and separation of noise and signal, and images with less noise are reconstructed without loss of detail. ^19,20,26

SD is commonly used for evaluation of the image noise. ²⁷ In both the phantom and clinical studies, the SD was significantly lower in the images reconstructed by 3D-IIR than in those reconstructed by 50% ASiR, and in images reconstructed by 50% ASiR than in those reconstructed by FBP. However, SD has the limitation that it does not provide information about image noise appearance. In addition to SD, assessment of NPS is essential to analyze the noise texture. ²⁷ NPS at lower spatial frequencies corresponds to noise of rough texture, and that at higher spatial frequencies corresponds the noise of fine texture. ^27–30 In our study, the NPS of the phantom images showed a low image noise throughout the entire range of spatial frequency in the images reconstructed by 3D-IIR than in those reconstructed by 50% ASiR, and in the images reconstructed by 50% ASiR than in those reconstructed by FBP.

CNR measurement is used to assess low-contrast resolution. ^18,31 The SD decreased in the order of 3D-IIR to 50% ASiR to FBP, while preserving the CT attenuation in both the phantom and clinical studies, resulting in a higher CNR in the images reconstructed by 3D-IIR than in those than in those reconstructed by 50% ASiR, and in the images reconstructed by 50% ASiR than in those reconstructed by FBP. The improved low-contrast detectability in the images reconstructed by 3D-IIR probably contributed to the better delineation of the intrahepatic vasculature in the 3D-IIR images as compared to that in the images reconstructed by 50% ASiR or FBP in this clinical study.

MTF is one of the most common used metrics to evaluate the spatial resolution. ^18,27 In this phantom study, use of 3D-IIR had no obvious effect on the MTF measurement, which indicates that the image noise reduction by 3D-IIR is achieved while the spatial resolution is preserved. The results are consistent with Padole’s report, which showed the capability of 3D-IIR for reducing image noise while allowing the structural details to be preserved. ^20,32 The better visualization of small or thin structures and diagnostic acceptability in clinical images obtained by 3D-IIR are also considered to reflect the noise reduction achieved by 3D-IIR while allowing the structural details to be maintained, as shown above.

In regard to the diagnostic acceptability, all images obtained by 3D-IIR were evaluated as being fully or probably acceptable. On the other hand, approximately 35% of the images reconstructed by 50% ASiR were judged as being possibly acceptable or only acceptable under limited conditions, and approximately 66% the images reconstructed by FBP were evaluated as possibly acceptable or only acceptable under limited conditions. These results indicate that the images obtained by 3D-IIR have the potential to be utilized in routine clinical practice, while the images reconstructed by other techniques present with some diagnostic difficulties.

Thus, the results of this study show that 3D-IIR for image reconstruction in VMI at 40 keV improves the image quality by reducing the image noise, and yields an image quality that is sufficiently acceptable for diagnosis. Previous studies have reported the effectiveness of VMI at 40 keV for detecting low-contrast lesions, and of using it as a supplement to VMI at 65 keV, as the standard energy level, for obtaining additional useful information in the evaluation of neck and head pathologies. ^7–14 The detectability of the lesions can be improved by enhancing the iodine contrast, however, in some cases, the image noise in VMI at 40 keV was still rated, in subjective evaluation, as being substantial, resulting in poor and non-diagnostic image quality. In this context, in our study, the image noise in all the clinical images reconstructed by 3D-IIR was subjectively rated as being minimal and the diagnostic acceptability was rated as being fully or probably acceptable. 3D-IIR may contribute to further reduction of the image noise in VMI at 40 keV, allowing better image quality to be obtained. Image noise reduction also contributes to radiation dose reduction. ^15,16

An added advantage of 3D-IIR is that it can be used with scanners of any model from any vendor, ^19,20 including for images reconstructed using FBP in the past, if DICOM data are available. As shown in this study, the image quality of VMI at 40 keV was improved, as compared to that obtained with FBP, by the use of 50% ASiR, which is a vendor-specific technique, although there were some diagnostic difficulties. On the other hand, 3D-IIR, which uses a vendor/model-neutral algorithm, significantly improved the image quality as compared to that obtained with either FBP or 50% ASiR, with significantly improved diagnostic acceptance. Needless to say, reference to past CT images is essential for follow-up studies and for evaluation of the response to treatment. To reduce the image noise in images obtained by VMI at 40 keV in the past with reconstruction using FBP may improve the clinical diagnostic acceptance. In this regard, 3D-IIR has the potential to further expand the clinical applicability of VMI at 40 keV.

This study had several limitations. One of the limitations was that we used the DECT from a single vendor. Differences in the principle of image acquisition of the DECT can affect the results. Therefore, evaluation of DECT from other vendors is desirable. Several IR techniques, besides ASiR, have recently been introduced commercially, and these image quality obtained using the other IR algorithms, including model-based iterative reconstruction, need to be discussed. Furthermore, our study was retrospective and population of clinical study was relatively small. Prospective study with more patients may be preferred to confirm our results.

Conclusion

3D-IIR improves the image quality in images obtained by VMI at 40 keV, with reduced image noise and maintained structural details. The imaging is feasible for routine clinical practice and can contribute to expansion of the clinical application of DECT.