Dual-Source Single-Energy Multidetector CT Used to Obtain Multiple Radiation Exposure Levels within the Same Patient: Phantom Development and Clinical Validation

Data from our phantom experiment and preliminary clinical series of prospectively enrolled patients render a proof of concept and a foundation for the use of a dual-source single-energy CT technique to obtain multidetector CT data sets at various radiation exposure levels within the same patient.

Abstract

Purpose

To develop, in a phantom environment, a method to obtain multidetector computed tomographic (CT) data sets at multiple radiation exposure levels within the same patient and to validate its use for potential dose reduction by using different image reconstruction algorithms for the detection of liver metastases.

Materials and Methods

The American College of Radiology CT accreditation phantom was scanned by using a dual-source multidetector CT platform. By adjusting the radiation output of each tube, data sets at six radiation exposure levels (100%, 75%, 50%, 37.5%, 25%, and 12.5%) were reconstructed from two consecutive dual-source single-energy (DSSE) acquisitions, as well as a conventional single-source acquisition. A prospective, HIPAA-compliant, institutional review board–approved study was performed by using the same DSSE strategy in 19 patients who underwent multidetector CT of the liver for metastatic colorectal cancer. All images were reconstructed by using conventional weighted filtered back projection (FBP) and sinogram-affirmed iterative reconstruction with strength level of 3 (SAFIRE-3). Objective image quality metrics were compared in the phantom experiment by using multiple linear regression analysis. Generalized linear mixed-effects models were used to analyze image quality metrics and diagnostic performance for lesion detection by readers.

Results

The phantom experiment showed comparable image quality between DSSE and conventional single-source acquisition. In the patient study, the mean size-specific dose estimates for the six radiation exposure levels were 13.0, 9.8, 5.8, 4.4, 3.2, and 1.4 mGy. For each radiation exposure level, readers’ perception of image quality and lesion conspicuity was consistently ranked superior with SAFIRE-3 when compared with FBP (P ≤ .05 for all comparisons). Reduction of up to 62.5% in radiation exposure by using SAFIRE-3 yielded similar reader rankings of image quality and lesion conspicuity when compared with routine-dose FBP.

Conclusion

A method was developed and validated to synthesize multidetector CT data sets at multiple radiation exposure levels within the same patient. This technique may provide a foundation for future clinical trials aimed at estimating potential radiation dose reduction by using iterative reconstructions.

^© RSNA, 2016

Online supplemental material is available for this article.

Introduction

By decreasing statistical noise associated with low photon count, iterative reconstructions offer the potential for substantial radiation dose reduction while providing diagnostic-quality images. Cumulative evidence indicates that, for routine abdominal applications, radiation dose may be decreased up to 75% by using iterative reconstructions (1–4). Nevertheless, coherent integration of iterative reconstructions into routine clinical practice has been challenged by the substantial variability in radiation dose reduction potential among different studies. This phenomenon may be explained by the interplay of rapid advances in iterative reconstruction technology, diversity in radiation dose reference levels, and lack of data from prospective trials.

Newer-generation dual-source multidetector computed tomographic (CT) platforms enable the reconstruction of image data sets at different radiation dose levels from a single CT acquisition (5–8). Studies have shown that when both radiation tubes are operated at the same tube potential and tube current (ie, dual-source single-energy [DSSE] acquisition mode), half-dose and full-dose images can be reconstructed from the same CT acquisition by using the projection data of only one tube or both tubes, respectively (5–7). Of note, because each x-ray tube in a dual-source CT system has its own generator, it is technically possible to set up individual adjustments of the output (and thus radiation dose) of each radiation tube during a DSSE acquisition (9).

The separate reconstruction of the projection data of each tube independently or the combination of both tubes expands the number of different radiation exposure data sets that can be reconstructed from a single DSSE acquisition within the same patient, thus providing a powerful method for assessing the potential radiation dose reduction by using iterative reconstructions.

With this goal in mind, the purpose of this study was to develop, in a phantom environment, a method to obtain multidetector CT data sets at multiple radiation exposure levels within the same patient and to validate its use for potential dose reduction by using different image reconstruction algorithms for the detection of liver metastases.

Materials and Methods

One author of the study (J.C.R.G.) is an employee of Siemens Medical Solutions USA. All other authors are not employees of or consultants for industry and had control of inclusion of any data that could represent a conflict of interest.

Phantom Experiment

Data acquisition.—The American College of Radiology (ACR) CT accreditation phantom (model 464; Gammex-RMI, Middleton, Wis) was scanned by using a second-generation dual-source multidetector CT platform (Somatom Definition Flash; Siemens Healthineers, Forchheim, Germany), operating two x-ray tubes and two 64-row detectors with fully integrated electronics (Stellar Detector, Siemens Healthineers). The multidetector CT scanner had a research-only acquisition mode available, in which the system can be operated in DSSE mode and the effective tube current–time product can be adjusted by the user. From a single DSSE acquisition, three distinct data sets can be reconstructed: two corresponding to each x-ray tube (eg, tube “A” and tube “B”), as well as a combination of the data from both tubes (tubes A and B combined). This acquisition mode is not commercially available, and manufacturer assistance was required for installation. In all experiments, the DSSE acquisition was split, with 75% of the effective tube current–time product for tube A and 25% of the effective tube current–time product for tube B.

The phantom was scanned by using both single-source and DSSE acquisitions (Table 1). Six single-source CT scans were acquired by using 200, 150, 100, 75, 50, and 25 mAs (ie, reference tube current–time product) representing six radiation output levels at 100%, 75%, 50%, 37.5%, 25%, and 12.5%, respectively. With DSSE, only two scans were performed. In the first DSSE scan, tubes A and B were operated with 150 and 50 effective mAs, respectively, for a total combined output of 200 effective mAs. This approach represented equivalent radiation outputs of 75%, 25%, and 100%, respectively. For the second DSSE scan, tubes A and B were operated with 75 and 25 effective mAs, respectively, for a total combined output of 100 effective mAs, representing radiation output levels of 50%, 37.5%, and 12.5%. During all scans, the automatic exposure control (CAREDose4D; Siemens Healthineers) was turned off. Other scan parameters that affect total radiation exposure, including tube voltage, gantry revolution time, and pitch, were kept constant.

Table 1.

CT Acquisition and Reconstruction Parameters for Single-Source and DSSE Scans

Note.— FBP = filtered back projection, SAFIRE-3 = sinogram-affirmed iterative reconstruction with strength level of 3.

*The effective tube current–time product was automatically adjusted by the automatic exposure control system, depending on patient body size.

Image reconstruction.—Images for both single-source and DSSE data sets were reconstructed at 5.0- and 0.6-mm section thickness with conventional weighted FBP by using a medium-smooth B31 kernel and the SAFIRE-3 technique by using a medium-smooth I31 kernel. Images were reconstructed by using SAFIRE-3 on the basis of prior data that suggested optimal trade-off between noise reduction and minimal change in noise texture as compared with FBP (8).

Data analysis.—Data analysis was performed in the first, second, and third modules of the ACR phantom (Fig 1). Measurements of noise and image contrast were performed by using a dedicated secondary workstation unit (Core2 x6800; Intel, Santa Clara, Calif) equipped with previously validated custom Matlab-based software (Matlab, version 2009a; The MathWorks, Natick, Mass). To ensure consistency and reproducibility of the data, all measurements were repeated three times on three consecutive images along the z-axis of the phantom, and the 10% winsorized mean of the ROI measurements was used for statistical analyses.

Figure 1:

A, Photograph of the ACR accreditation phantom and, B–D, transverse CT images of the phantom. B, Module 1 has four rod inserts. Circular regions of interest (ROIs) enclosing the inserts were used to assess the contrast-dependent resolution properties as measured with the task-based modulation transfer function. C, Module 2 is the low-contrast section, which was used to assess contrast-to-noise ratio (CNR). The center circular ROI was the target, and the off-center ROI was the background. D, Module 3 is uniform. The four square ROIs were used to assess the noise properties in terms of noise power spectra and noise amplitude.

Quantitative image quality metrics were measured by using different components of the ACR phantom, including noise power spectrum, modulation transfer function, task-based modulation transfer function, and detectability index (Appendix E1 [online]) (9,10).

Clinical Study

This prospective, single-center, Health Insurance Portability and Accountability Act–compliant study was approved by the institutional review board of Duke University, and patients gave written informed consent before enrollment.

Sample size considerations.—A noninferiority statistical analysis was performed for comparing the reconstruction algorithm SAFIRE-3 at 50% and 37.5% radiation exposure levels against FBP at 100% by using the mean readers’ score of overall image quality as the primary outcome measure (see the Data Analysis section herein). The sample size and power analysis calculation was performed through a simulation study in R software (R Foundation, Vienna, Austria), with a code specifically designed for the analysis (Appendix E2 [online]). Our calculations indicate that, with a noninferiority margin of four units of image quality, four readers and 15 patients with only one lesion each would provide more than 80% power to test for noninferiority at a 5% significance level.

Patients.—Twenty-one consecutive patients who underwent clinically indicated contrast material–enhanced multidetector CT of the chest, abdomen, and pelvis for metastatic colorectal cancer were prospectively enrolled. Patients were eligible for enrollment if they were known to have or suspected of having liver metastases on the basis of the results of (a) previous multidetector CT or ultrasonographic examination findings (n = 15) or (b) increased carcinoembryonic antigen tumor marker levels (>5 ng/mL [>5μg/L]) (n = 6). Subjects were excluded if they had (a) a clinical history of anaphylactoid reaction to iodinated contrast media, (b) renal failure defined as serum creatinine level higher than 2.0 mg/mL (177 μmol/L), or (c) suboptimal image quality. Two subjects were excluded from the study owing to suboptimal image quality secondary to inadequate scanning time (n = 1) and contrast material extravasation (n = 1).

The subjects’ accrual flowchart (Fig 2) is based on the Standards for Reporting of Diagnostic Accuracy initiative (11) and proof of tumor burden. The final study cohort consisted of 19 patients (mean age ± standard deviation, 60 years ± 11; median age, 62 years; age range, 44–80 years), including nine men (mean age, 66 years ± 10; median age, 65 years; age range, 49–80 years) and 10 women (mean age, 54 years ± 9; median age, 53 years; age range, 44–69 years). The mean patient effective diameter was 27.6 cm ± 4.7 (median, 27.8 cm; interquartile range, 20.3–39.1 cm).

Figure 2:

Flowchart of study population enrollment. DECT = dual-energy CT, Mets = metastases.

Data acquisition and image reconstruction.—All scans were performed by using the same second-generation dual-source multidetector CT platform of the phantom experiment (Somatom Definition Flash; Siemens Healthineers). All patients were positioned supine, head first, on the scanning table and underwent scanning in the craniocaudal direction. After acquisition of anteroposterior and lateral digital localizer radiographs, a routine-dose (quality reference tube current–time product of 200 mAs) single–breath-hold acquisition of the chest, abdomen, and pelvis was performed during the hepatic parenchymal (portal venous dominant) phase (approximately 70 seconds after the start of contrast medium injection) by using a DSSE acquisition. All patients received 150 mL of an intravenous nonionic contrast medium with an iodine concentration of 300 mg of iodine per milliliter (iopamidol, Isovue 300; Bracco Diagnostics, Princeton, NJ). The bolus of contrast medium was injected through an 18–20-gauge angiocatheter inserted into an antecubital fossa vein by using a dual-chamber mechanical power injector (Empower; E-Z-Em, Lake Success, NY) at a flow rate of 3 mL/sec.

The initial diagnostic scan was followed by a second, half-dose DSSE scan (quality reference tube current–time product of 100 mAs) targeted to the liver (Table 1). The incremental radiation dose burden to the patient from this second, non–clinically indicated scan was estimated to be between 15% and 25% (50% reduction in volume CT dose index [CTDI_vol] and approximately 50%–65% reduction in the scanning range) relative to the initial diagnostic scan. To minimize the interscan delay, the second scan was performed in a caudocranial direction (the opposite direction of the initial diagnostic scan). This approach minimized the delay from repositioning of the CT table (typically between 3 and 6 seconds, depending on the scan length). Radiation exposure data were collected, including the CTDI_vol and dose-length product values provided by the scanner (Table 2). Size-specific dose estimate (SSDE in the following equation) values were calculated by using the effective diameter and CTDI_vol (12):

For comparative purposes, effective radiation dose values in millisieverts were estimated by multiplying the different scanner photon output levels (in terms of CTDI_vol) by using a fixed scan length of 26 cm and a conversion factor of 0.015 mSv/(mGy · cm) (13).

Table 2.

Radiation Dose for Each Dose Level Considered in This Study by Using a Dual-Source Scan

Note.—Data are means ± standard deviations, unless indicated otherwise.

*CTDI_vol values at these dose levels were inferred from the Digital Imaging and Communications in Medicine headers, as they are not routinely reported on the protocol page.

^†For comparative purposes, the estimations of effective dose were calculated by using the corresponding CTDI_vol at each radiation output level, a fixed scan length of 26 cm covering the abdominal region only, and a conversion factor of 0.015 mSv/(mGy · cm).

Similar to the phantom experiment, for both the conventional and half-dose acquisitions, the radiation dose was split between the two x-ray tubes by using a 75%–25% split. This approach resulted in a total of six different radiation dose levels within the same patient during two consecutive acquisitions, including 100%, 75%, and 25% radiation dose levels from the first acquisition and 50%, 37.5%, and 12.5% radiation dose levels from the second acquisition. For all clinical scans, the automatic exposure control (CAREDose4D; Siemens Healthineers) was turned on.

All images were reconstructed at 5.0-mm section thickness by using the same reconstruction methods (ie, FBP and SAFIRE-3) and kernel of the phantom experiment. A total of 12 data sets were generated for each patient (six radiation dose levels and two reconstruction algorithms per patient) for a total of 228 data sets for the entire study population.

Data analysis.—Subjective image quality was assessed in a blinded and randomized fashion by four independent radiologists (L.M.H., D.M., A.M., and A.B., with 9, 6, 5, and 3 years of experience in abdominal imaging, respectively). To minimize the effect of recall bias from the interpretation of multiple data sets within the same patient (14), data sets were presented to the readers during four evaluation sessions, separated by a 4-week interval. All patient identifiers (including name, age, sex, and medical record numbers) were removed from the images. Each reconstructed data set was initially presented by using a preset soft-tissue window (window width, 350 HU; window level, +40 HU). The radiologists had the ability to select or adjust clinical intensity windows for image viewing and interactively scroll through the images during the lesion detection task and subjective image quality evaluation.

Readers ranked image quality by using a slider bar and a 100-point scale according to the recommendations of the International Commission on Radiation Units and Measurements Image and the European Guidelines on Quality Criteria for Computed Tomography (15). The following attributes of image quality were evaluated: (a) sharpness (defined as the discreteness of the margins of liver and splanchnic vasculature); (b) noise (defined as the amount of graininess or mottle of the image); (c) artifacts (including beam hardening or streaks due to metal or high-density materials, such as residual barium within the colonic lumen); and (d) overall image quality (regarded as the reader’s desire to use a certain image appearance to confidently detect the presence or absence of liver lesions). Readers were instructed that an overall quality score of less than 30 was regarded as inadequate for diagnostic purposes. To maximize reproducibility of the reader’s interpretation of image quality, standardized criteria were presented to the readers in a training session held immediately prior to the first reading session.

After the image quality evaluation, readers were asked to comment on the presence or absence of liver lesions, as well as differentiate between benign and malignant liver lesions. Lesions were marked by circumscribing an ROI around the lesion. After identification of a lesion, the operator was required to answer questions regarding confidence in the lesion detection task (ie, lesion conspicuity) by using a 100-point scale, where a score of 0 indicated the absence of identifiable liver lesions and a score of 100 indicated the greatest reader’s confidence level for identifying a liver lesion.

Reference standard.—Proof of tumor burden was obtained by the principal investigator (D.B.), who was not involved in the imaging interpretation sessions. The presence of liver lesions was assessed by comparing routine-dose multidetector CT images with the results of previous and subsequent multidetector CT and/or magnetic resonance (MR) examinations, as well as the surgery and histopathology reports, if available (Fig 2). For malignant liver lesions, proof of malignancy was achieved with histopathologic findings or demonstration of lesion progression or response at serial MR or multidetector CT examinations. Progression was defined as an increase in the longest lesion diameter, while response was defined as a decrease of 5 mm or more in the longest lesion diameter at follow-up imaging or at the index examination in comparison with previous examinations. In patients with multiple liver lesions who underwent percutaneous biopsy, tissue samples were obtained from the largest lesion.

Sixty-three percent of enrolled patients (12 of 19) had a total of 35 liver lesions. The mean number of lesions per patient was 1.8 (range, 0–5), with a mean lesion size of 16 mm ± 8 (range, 6–36 mm). Thirty-four lesions in 12 patients were diagnosed as metastatic colorectal tumors on the basis of histopathologic analysis (n = 28) or imaging follow-up (n = 6) for a minimum of 13 months (mean follow-up time, 28.7 months; range, 13–47 months). One lesion in one patient with concurrent metastatic lesions was diagnosed as a benign hepatic cyst on the basis of typical MR imaging characteristics and stability at follow-up (24 months). Matching of reference and reader markings was performed by an independent radiologist (D.B.) who did not participate in the blinded interpretation.

Statistical analysis.—The phantoms were used to measure image contrast, noise, CNR, and detectability index as a function of dose level and source type. To determine whether the observations associated with each source type were significantly different from each other, a multiple linear regression model was fit for each response (or its logarithm) with dose level, source type, and the interaction between them as explanatory variables. The linear model was defined as , where y_ij(d) is the response variable of the jth observation from the ith source type (i = 1,2 and j = 1,2,3,4); is an intercept; s_ij is the source effect; d is the radiation dose, assumed to be continuous (with slope ); ds_ij is the dose-source interaction (with slope ); and E_ij is the measurement error, assumed to be normally distributed with zero-mean and constant variance, henceforth denoted as . Regressions were fit by using the least-squares method, and the difference between intercepts and slopes of the lines for each source type were assessed. In addition, to quantify the overall difference between source types regarding the response, the mean relative difference was calculated as

where y_s,i and y_d,i denote the ith measurement from a single and double source, respectively.

The response variables of sharpness, noise, artifacts, overall image quality, and conspicuity were analyzed independently by using linear mixed-effects models of the form

where Y_ijklm is the response of the mth observation from the lth patient, made by the kth reader by using the jth reconstruction algorithm and the ith radiation dose level. The model expresses the expected value of Y_ijklm as a function of a baseline mean () that represents the routine-dose (100%) FBP. The predictors are dose (d_i), reconstruction algorithm (a_j), and the interaction between them (da_ij). The reader and patient effects were considered random [, ] to account for the clustering nature of the observations; the last term [] is the random measurement error. The models were fit with the restricted maximum likelihood approach and were used to determine the appropriate dose reduction for SAFIRE-3 that provides a comparable overall image quality to routine-dose FBP. The Snedecor F test with Satterthwaite approximation was used to evaluate the image quality equality between FBP at 100% and SAFIRE-3 at every radiation level. The Mann-Whitney test was used to compare the responses between the reconstruction algorithms at every radiation dose level, and the P values were adjusted with Bonferroni correction (16,17). In all cases, the threshold for assessing statistical significance was set to an α level of .05. All statistical analyses were performed with SAS software (SAS Institute, Cary, NC) (18) and R statistical software (R Foundation) (19).

Results

Phantom Experiment

For each radiation dose level, there was no significant difference in measured CNR between the single-source and DSSE acquisitions (Fig 3). The comparison of intercepts and slopes showed a statistically significant difference in both noise and detectability index between the two acquisition methods. However, the relative percentage difference was negligible, measuring only 1.19% and 1.30% for noise and −0.28% and −0.26% for detectability index, respectively (Fig 3).

Figure 3a:

Plots show noise, CNR, and detectability index (for a 5-mm task) as a function of exposure level (logarithmic scale) according to source type for both reconstruction methods. The regression lines for noise and detectability index as a function of exposure level and reconstruction algorithm fit the data almost perfectly (R² ≈ 0.99) for both B31F and I31F phantoms. As a result, fit parameters are highly significant, and confidence intervals are narrow. (a, c) Differences in noise and detectability index were significant among methods based on intercept and slope comparisons. However, the lines are similar to each other, with small mean relative differences (on the order of 1% among methods), which should be considered negligible for clinical purposes. (b) For each radiation exposure level, there was no significant difference in measured CNR between single-source and DSSE acquisitions.

Figure 3b:

Plots show noise, CNR, and detectability index (for a 5-mm task) as a function of exposure level (logarithmic scale) according to source type for both reconstruction methods. The regression lines for noise and detectability index as a function of exposure level and reconstruction algorithm fit the data almost perfectly (R² ≈ 0.99) for both B31F and I31F phantoms. As a result, fit parameters are highly significant, and confidence intervals are narrow. (a, c) Differences in noise and detectability index were significant among methods based on intercept and slope comparisons. However, the lines are similar to each other, with small mean relative differences (on the order of 1% among methods), which should be considered negligible for clinical purposes. (b) For each radiation exposure level, there was no significant difference in measured CNR between single-source and DSSE acquisitions.

Figure 3c:

Plots show noise, CNR, and detectability index (for a 5-mm task) as a function of exposure level (logarithmic scale) according to source type for both reconstruction methods. The regression lines for noise and detectability index as a function of exposure level and reconstruction algorithm fit the data almost perfectly (R² ≈ 0.99) for both B31F and I31F phantoms. As a result, fit parameters are highly significant, and confidence intervals are narrow. (a, c) Differences in noise and detectability index were significant among methods based on intercept and slope comparisons. However, the lines are similar to each other, with small mean relative differences (on the order of 1% among methods), which should be considered negligible for clinical purposes. (b) For each radiation exposure level, there was no significant difference in measured CNR between single-source and DSSE acquisitions.

Clinical Study

The analysis of the subjective grading of image quality showed a gradual decrease in readers’ scores with the progressive reduction in radiation exposure for all image quality parameters and both reconstruction methods (Fig 4). For each radiation exposure level, all image quality parameters were ranked superior with SAFIRE-3 when compared with FBP (P ≤ .05 for all comparisons, Fig 4), with excellent agreement among the four readers (Pearson product-moment correlation, 0.77–1.00). Notably, the percentage of overall examinations deemed below a reasonable standard of quality was 9% (67 of 744 observations) for FBP but decreased to 5.8% (43 of 744 observations) for SAFIRE-3. The analysis of readers’ rankings of overall image quality showed that, for radiation exposure reductions of up to 62.5%, SAFIRE-3 was not inferior when compared with FBP routine exposure for a four-unit equivalence margin (P = .047) (Fig 5).

Figure 4a:

Box plots of (a) sharpness, (b) noise, (c) artifacts, and (d) overall image quality as a function of exposure level and reconstruction algorithm. The Mann-Whitney test was used to compare image quality metrics between reconstruction algorithms at every exposure level, and the P values (at the bottom of the plots) were adjusted with Bonferroni correction for multiple comparisons. Note that SAFIRE-3 and FBP are significantly different from each other at every exposure level, with SAFIRE-3 always higher than FBP. Also note that higher values on the y-axis correspond to lower levels of perceived noise (a). * = extreme outliers (ie, values with more than three times the height of the boxes).

Figure 5:

Regression curve shows the smoothed conditional mean of overall image quality, along with 95% confidence interval, as a function of exposure level for the SAFIRE-3 reconstruction algorithm. The curve was obtained by using local polynomial regression fitting. The horizontal line is the mean of the FBP algorithm at the 100% exposure level. The curves intersect at around the 50% exposure level. The mean of FBP at the 100% exposure level and the mean of SAFIRE-3 at the 50% exposure level are not significantly different (F test, P = .897). Similarly, the mean of FBP at the 100% exposure level and the mean of SAFIRE-3 at the 37.5% exposure level are not significantly different, indicating the potential for up to 62.5% radiation exposure reduction (F test, P = .100).

Figure 4b:

Box plots of (a) sharpness, (b) noise, (c) artifacts, and (d) overall image quality as a function of exposure level and reconstruction algorithm. The Mann-Whitney test was used to compare image quality metrics between reconstruction algorithms at every exposure level, and the P values (at the bottom of the plots) were adjusted with Bonferroni correction for multiple comparisons. Note that SAFIRE-3 and FBP are significantly different from each other at every exposure level, with SAFIRE-3 always higher than FBP. Also note that higher values on the y-axis correspond to lower levels of perceived noise (a). * = extreme outliers (ie, values with more than three times the height of the boxes).

Figure 4c:

Box plots of (a) sharpness, (b) noise, (c) artifacts, and (d) overall image quality as a function of exposure level and reconstruction algorithm. The Mann-Whitney test was used to compare image quality metrics between reconstruction algorithms at every exposure level, and the P values (at the bottom of the plots) were adjusted with Bonferroni correction for multiple comparisons. Note that SAFIRE-3 and FBP are significantly different from each other at every exposure level, with SAFIRE-3 always higher than FBP. Also note that higher values on the y-axis correspond to lower levels of perceived noise (a). * = extreme outliers (ie, values with more than three times the height of the boxes).

Figure 4d:

Box plots of (a) sharpness, (b) noise, (c) artifacts, and (d) overall image quality as a function of exposure level and reconstruction algorithm. The Mann-Whitney test was used to compare image quality metrics between reconstruction algorithms at every exposure level, and the P values (at the bottom of the plots) were adjusted with Bonferroni correction for multiple comparisons. Note that SAFIRE-3 and FBP are significantly different from each other at every exposure level, with SAFIRE-3 always higher than FBP. Also note that higher values on the y-axis correspond to lower levels of perceived noise (a). * = extreme outliers (ie, values with more than three times the height of the boxes).

The analysis of the lesion detection task showed, for both reconstruction methods, a decrease in readers’ sensitivity, with progressive reduction in radiation exposure (Table 3). Although we found no statistically significant difference in sensitivity and specificity between SAFIRE-3 and FBP for each radiation exposure level, the results of the fit logistic mixed-effects model showed a significantly higher sensitivity with SAFIRE-3 for the combined data at different radiation exposures (P = .037; odds ratio = 1.43; 95% confidence interval: 1.02, 2.01). Furthermore, readers’ perception of lesion conspicuity improved significantly with SAFIRE-3 at radiation exposure levels of 75% (P < .001), 50% (P = .009), 37.5% (P < .001), and 12.5% (P < .001) (Fig 6). Our data also suggest that, for radiation dose reductions of up to 50%, there was no significant difference in readers’ perception of lesion conspicuity with SAFIRE-3 when compared with routine exposure with FBP (Figs 6, 7).

Table 3.

Sensitivity and Specificity for Lesion Detection as a Function of Dose and Reconstruction Algorithm with Clopper-Pearson 95% Confidence Intervals

Note.—Numbers in parentheses are the data used to calculate percentages. Numbers in brackets are 95% confidence intervals. The comparison of sensitivity and specificity showed no statistically significant difference between SAFIRE-3 and FBP at each radiation dose level. Clopper-Pearson 95% confidence intervals were calculated according to reference 17.

Figure 6:

Box plot of conspicuity scores as a function of exposure level and reconstruction algorithm. The P values at the bottom correspond to the Mann-Whitney tests for comparing conspicuity metrics between reconstruction algorithms at every exposure level, adjusted with Bonferroni correction. * = extreme outliers (ie, values with more than three times the height of the boxes).

Figure 7a:

Transverse contrast-enhanced CT sections in a 55-year-old man with liver metastases from colon cancer. Liver images were obtained during the portal phase at (a) 100%, (b) 75%, (c) 50%, (d) 37.5%, (e) 25%, and (f) 12.5% radiation exposure levels. Note the substantial decrease in noise with SAFIRE-3 compared with FBP for each exposure level. This effect resulted in substantially improved visualization of a (1.3-cm) hypoattenuating metastatic lesion in the posterior right hepatic lobe (arrow).

Figure 7b:

Transverse contrast-enhanced CT sections in a 55-year-old man with liver metastases from colon cancer. Liver images were obtained during the portal phase at (a) 100%, (b) 75%, (c) 50%, (d) 37.5%, (e) 25%, and (f) 12.5% radiation exposure levels. Note the substantial decrease in noise with SAFIRE-3 compared with FBP for each exposure level. This effect resulted in substantially improved visualization of a (1.3-cm) hypoattenuating metastatic lesion in the posterior right hepatic lobe (arrow).

Figure 7c:

Transverse contrast-enhanced CT sections in a 55-year-old man with liver metastases from colon cancer. Liver images were obtained during the portal phase at (a) 100%, (b) 75%, (c) 50%, (d) 37.5%, (e) 25%, and (f) 12.5% radiation exposure levels. Note the substantial decrease in noise with SAFIRE-3 compared with FBP for each exposure level. This effect resulted in substantially improved visualization of a (1.3-cm) hypoattenuating metastatic lesion in the posterior right hepatic lobe (arrow).

Figure 7d:

Transverse contrast-enhanced CT sections in a 55-year-old man with liver metastases from colon cancer. Liver images were obtained during the portal phase at (a) 100%, (b) 75%, (c) 50%, (d) 37.5%, (e) 25%, and (f) 12.5% radiation exposure levels. Note the substantial decrease in noise with SAFIRE-3 compared with FBP for each exposure level. This effect resulted in substantially improved visualization of a (1.3-cm) hypoattenuating metastatic lesion in the posterior right hepatic lobe (arrow).

Figure 7e:

Transverse contrast-enhanced CT sections in a 55-year-old man with liver metastases from colon cancer. Liver images were obtained during the portal phase at (a) 100%, (b) 75%, (c) 50%, (d) 37.5%, (e) 25%, and (f) 12.5% radiation exposure levels. Note the substantial decrease in noise with SAFIRE-3 compared with FBP for each exposure level. This effect resulted in substantially improved visualization of a (1.3-cm) hypoattenuating metastatic lesion in the posterior right hepatic lobe (arrow).

Figure 7f:

Transverse contrast-enhanced CT sections in a 55-year-old man with liver metastases from colon cancer. Liver images were obtained during the portal phase at (a) 100%, (b) 75%, (c) 50%, (d) 37.5%, (e) 25%, and (f) 12.5% radiation exposure levels. Note the substantial decrease in noise with SAFIRE-3 compared with FBP for each exposure level. This effect resulted in substantially improved visualization of a (1.3-cm) hypoattenuating metastatic lesion in the posterior right hepatic lobe (arrow).

Discussion

Data from our phantom experiment and preliminary clinical series of prospectively enrolled patients render a proof of concept and a foundation for the use of a DSSE technique to obtain multidetector CT data sets at various radiation exposure levels within the same patient. This technique may provide a powerful tool for research and optimization of clinical protocols aimed toward the reduction of radiation dose. Although in prior studies, investigators have assessed the utility of a DSSE technique in reconstructing full-dose (two tubes) and half-dose (one tube) radiation levels (5–8), investigators in only one preliminary phantom study with first-generation dual-source CT determined the feasibility of obtaining different radiation exposure levels between the two tubes during a DSSE acquisition (9). Our results corroborate this preliminary experimental data by Werncke and colleagues (9) and demonstrate comparable image quality (ie, less than 5% difference for both first- and second-order image quality metrics) between conventional single-source and DSSE acquisitions. These results were confirmed at different dose-matched radiation exposure levels, ranging from routine-dose abdominal CT (effective dose of 4 mSv) to submillisievert CT. This information is crucial for the use of a DSSE technique in the assessment of technologies and strategies that may enable routine CT to be performed below an effective dose of 1 mSv, a major public heath priority in the United States (20).

In our experimental and preliminary clinical studies, we investigated the feasibility of using a variable-dose split DSSE protocol to generate, within the same patient, multidetector CT data sets of the liver at six different radiation exposure levels, with only an 18% increase in the dose-length product (ie, 123 mGy · cm ± 51 for the second, limited-coverage DSSE acquisition, compared with 672 mGy · cm ± 191 for the first, diagnostic DSSE acquisition). This strategy overcomes some of the major limitations of previous studies in which different radiation exposure levels were compared from separate study cohorts (an approach that is confounded by potentially large interpatient variability) or the use of artificial noise insertion tools to simulate reduced-dose CT data sets (21,22). Furthermore, when compared with multiple single-energy CT acquisitions at different radiation dose levels, a variable dose-split DSSE protocol eliminates the need for multiple intravenous contrast material injections and avoids potential variability due to misregistration among different data sets secondary to patient motion or differences in contrast material timing.

Our study may have important clinical implications. By generating data sets at different radiation dose levels (including submillisievert levels) from a single DSSE acquisition within the same patient, the dose-split DSSE technique minimizes the effect of important confounding variables for the assessment of dose reduction strategies, including interpatient or interscan variability. This will enable more accurate and precise optimization of CT radiation dose according to patient size, diagnostic task, and reconstruction algorithm. It will also accelerate the development and validation of novel reconstruction or noise reduction algorithms, facilitating determination of minimally acceptable dose levels for a wide range of diagnostic tasks (20).

Our preliminary clinical data consistently demonstrated superior reader perception of image quality with SAFIRE-3 when compared with FBP across a broad spectrum of different radiation exposure levels. This improvement was accompanied by an increased overall sensitivity for the detection of colorectal liver metastases, as well as a significantly higher confidence level for successfully detected lesions at each radiation exposure level. Perhaps even more important, our results for reader perception of image quality suggested the possibility of up to 62.5% radiation dose reduction with SAFIRE-3 when compared with a routine-dose FBP acquisition. Our results compare favorably with those in prior investigations (23–25) and were confirmed among four readers whose clinical experience ranged from 3 to 9 years of experience. Future larger clinical studies, in which readers’ diagnostic accuracy for liver lesion detection is used as a metric of image quality, are warranted to validate our preliminary clinical results.

Some potential limitations of our study merit consideration. First, our clinical findings reflect a preliminary experience from a small patient cohort. Further validation of our data in a larger and more diverse patient population is needed to generalize our results. Second, we did not investigate the feasibility of using a variable dose-split strategy at low tube voltages (lower than routine 120 kV) or in obese patients. According to a previous study, this technique may not be feasible below 100 kV or in larger patients (5). Finally, although hepatic parenchymal enhancement shows a uniform steady-state plateau (a flat broad peak) during the hepatic parenchymal phase (60–80 seconds after the start of contrast material injection), we cannot rule out the possibility that differences in enhancement between the first and second DSSE acquisitions might have affected our clinical observations.

In summary, the results of our study indicate that multidetector CT acquisition strategies in which a DSSE technique is used can reliably generate data sets at multiple radiation doses within the same patient. This approach may provide a foundation for future clinical trials aimed toward estimation of potential radiation dose reduction by using iterative reconstruction methods.

Advances in Knowledge

• ■ We developed and validated a method to synthesize multidetector CT data sets at six different radiation exposure levels within the same patient.

• ■ Our preliminary clinical study demonstrated consistently superior reader perceptions of image quality, conspicuity of metastatic liver lesions, and overall sensitivity for lesion detection by using sinogram-affirmed iterative reconstruction with strength level of 3 (SAFIRE-3) compared with filtered back projection (FBP) across a broad range of radiation exposure levels (P ≤ .05 for all comparisons).

• ■ Our clinical data also suggest the possibility of up to 62.5% reduction in radiation exposure by using SAFIRE-3 compared with routine-dose FBP for liver imaging.

Implication for Patient Care

• ■ A dual-source single-energy CT technique may provide a powerful tool for accurate and precise assessment of potential radiation dose reduction by using iterative reconstruction methods, allowing improved optimization of multidetector CT protocols.