Precision of Iodine Quantification in Hepatic CT: Effects of Iterative Reconstruction With Various Imaging Parameters

Transcription

1 Medical Physics and Informatics Original Research Iterative Reconstruction Algorithms in Hepatic CT Medical Physics and Informatics Original Research Baiyu Chen 1,2 Daniele Marin 3 Samuel Richard 2,3,4 Daniela Husarik 3 Rendon Nelson 3 Ehsan Samei 1,2,3, Chen B, Samei E, Marin D, Richard S, Husarik D, Nelson R Keywords: ASIR, dose reduction, hepatic CT, iodine quantification, iterative reconstruction, MBIR, precision DOI:.2214/AJR Received July 24, 12; accepted after revision October 26, 12. Supported in part by a grant from GE Healthcare, which also provided equipment and technical support for this study. R. Nelson and E. Samei are consultants for GE Healthcare. B. Chen received research funding from GE Healthcare. The remaining authors have no pertinent disclosures and maintained full control of the data and information submitted. 1 Medical Physics Graduate Program, Duke University, 2424 Erwin Rd, Ste 2, Durham, NC 277. Address correspondence to B. Chen (baiyu.chen@duke.edu). 2 Carl E. Ravin Advanced Imaging Laboratories, Duke University, Durham, NC. 3 Department of Radiology, Duke University, Durham, NC. 4 Present address: Carestream Health, Rochester, NY. Departments of Physics, Biomedical Engineering, and Electronic and Computer Engineering, Duke University, Durham, NC. WEB This is a Web exclusive article. AJR 13; :W47 W X/13/ W47 American Roentgen Ray Society Precision of Iodine Quantification in Hepatic CT: Effects of Iterative Reconstruction With Various Imaging Parameters OBJECTIVE. The objective of this study was to evaluate the feasibility of using iterative reconstructions in hepatic CT to improve the precision of Hounsfield unit quantification, which is the degree to which repeated measurements under unchanged conditions provide consistent results. MATERIALS AND METHODS. An anthropomorphic liver phantom with iodinated lesions designed to simulate the enhancement of hypervascular tumors during the late hepatic arterial phase was imaged, and images were reconstructed with both filtered back projection (FBP) and iterative reconstructions, such as adaptive statistical iterative reconstruction (ASIR) and model-based iterative reconstruction (MBIR). This protocol was further expanded into various dose levels, tube voltages, and slice thicknesses to investigate the effect of iterative reconstructions under all these conditions. The iodine concentrations of the lesions were quantified, with their precision calculated in terms of repeatability coefficient. RESULTS. ASIR reduced image noise by approximately 3%, and improved the quantitative precision by approximately %, compared with FBP. MBIR reduced noise by more than 6% and improved the precision by approximately % compared with the routine protocol. MBIR consistently showed better precision across a thinner slice thickness, lower tube voltage, and larger patient, achieving the target precision level at a dose lower ( 4%) than that of FBP. CONCLUSION. ASIR blended with % of FBP indicated a moderate gain in quantitative precision compared with FBP but could achieve more with a higher percentage. A higher gain was achieved by MBIR. These findings may be used to reduce the dose required for reliable quantification and may further serve as a basis for protocol optimization in terms of iodine quantification. I odinated contrast agents are commonly used with hepatic CT to enhance the tissue contrast between subtle lesions and liver parenchyma, with their concentrations being of quantitative interest in studies such as contrast bolus injection rate optimizations, arterial or portal phase differentiations, and noninvasive quantifications of perfusion [1 3]. Because a lesion s iodine concentration is proportional to the lesion s contrast, as measured by Hounsfield unit enhancement, it can be indirectly quantified from the Hounsfield unit count of the image. The reliability of this quantification, however, relies on the accuracy and precision of the quantification as a function of the imaging protocol. Accuracy is the degree to which measurements are close to the quantity s true value, which can be improved with calibrations. Precision is the degree to which repeated measurements under unchanged conditions show the same results. Precision is crucial to the reliability of an observed change in successive measurements, but it cannot be simply improved by calibrations. In addition, conventional CT reconstruction algorithms have been optimized for detection tasks, not quantification tasks. As a result, postprocessing algorithms are typically aimed at enhancing features of interest or providing artifact-free images, at the risk of sacrificing the quantitative precision [4]. A possible solution to better precision of iodine quantification is the deployment of new iterative reconstruction techniques, such as adaptive statistical iterative reconstruction (ASIR) and model-based iterative reconstruction (MBIR) provided by GE Healthcare. Iteratively reconstructed images have lower noise compared to images from conventional AJR:, May 13 W47

2 filtered back projection (FBP) reconstructions []. In particular, MBIR includes modeling of the image acquisition system, such as the limited size of the focal spot and the detector, which provides more accurate reconstruction, as well as improved resolution [6 9]. Iodine quantification in liver images reconstructed with iterative reconstruction might take advantage of these features and achieve the same precision as quantifications with FBP images at lower dose levels. Therefore, the purpose of this study was to investigate the impact of iterative reconstruction algorithms on the precision of iodine quantifications under various imaging conditions. Dose reduction potential of iterative reconstruction was also assessed in the context of precision of iodine quantification, as compared with FBP. Materials and Methods Liver Phantom With Iodinated Lesions An anthropomorphic liver phantom was designed at our institute and custom manufactured by CIRS to simulate hypervascular tumors imaged during the late hepatic arterial phase. As shown in Figure 1, the phantom is a slab with an elliptical cross-section (diameter, cm; thickness, 2. mm), composed of muscle background, a liver insert, 12 spherical lesions, and the spine. The lesions were of three diameters (1., 1, and. cm) and two contrast levels and were placed in concentric rings (diameter, 3. and 7. cm) around the isocenter to avoid uncertainties raised from locations. In addition, the phantom contained two 4-cm adipose rings that can be wrapped around the phantom, allowing the simulation of larger adults having either 4 or 8 cm of subcutaneous fat. To simulate the attenuation of hypervascular liver tumors and normal liver parenchyma during the late hepatic arterial phase after administration of iodinated contrast material, different concentrations of iodine were added to the liver parenchyma and lesions to provide the target Hounsfield unit counts listed in Table 1. The values in Table 1 were derived in three steps: first, the Hounsfield unit counts of the liver parenchyma at 8 and 14 kvp were experimentally measured from 6 consecutive patients who underwent clinically indicated dual-energy CT scans of the liver at our institution, as indicated in Table 1 [, ]. Second, the Hounsfield unit counts of the lesions with low and high contrast against the parenchyma background, as indicated in Table 1, were designed to correspond to.4 and.8 mg/ml additional iodine concentration. This allows the enhancement of the low-contrast lesion to be invisible (8 HU) at 14 kvp but visible (24 HU) at 8 kvp, because a previous study has shown that a minimum of HU is required to observe the lesion [11]. Finally, the rest of the data in Table 1 were mathematically interpolated from the previously calculated data according to tube voltage and iodine concentration. Scanning and Reconstruction Protocols The liver phantom without an additional adipose ring (representing a -kg patient), which is referred to as the small patient in the rest of this work, was scanned (Discovery CT7 HD, GE Healthcare). A routine abdominal protocol at our institution was used, with 4-mm beam collimation, 1 kvp, 1.37 pitch, tube current determined by automated tube current modulation (as determined by anteroposterior and lateral digital scout radiographs) with a 14.-HU noise index requirement (CT dose index [CTDI], 4. mgy), FBP reconstruction, and 2.-mm slice thickness. On the basis of the routine protocol, two additional reconstruction algorithms, ASIR and MBIR, were applied to the same acquisition dataset as that of FBP to assess their effect on quantification performance. Note that the ASIR used in this study was % blended with FBP to reduce the somewhat waxy appearance of iterative-reconstructed images, which is a typical clinical process to provide an im- Fig. 1 Anthropomorphic liver phantom used for liver image simulations. A and B, Diagram (A) and photograph (B) depict phantom, which was slab with liver insert and iodinated hepatic lesions of two concentrations. C, Two adipose rings added to periphery of phantom allowed flexibility of simulating larger patient sizes. W476 AJR:, May 13

3 Iterative Reconstruction Algorithms in Hepatic CT age perception comparable to that of the FBP images to which most radiologists are accustomed. Also note that, because all three reconstruction algorithms were applied to the same acquisition dataset, they shared the same CTDI values. Other imaging and reconstruction parameters were also expanded into a full parameter space to assess the effect of iterative reconstruction under various circumstances, including four additional dose levels, corresponding to 7%, %, %, and % of the clinical dose level; two additional tube voltages, 8 and kvp; and one additional reconstruction slice thickness at.6 mm. For protocols of the same dose level but different tube voltages, the tube currents were adjusted accordingly to maintain the isodose condition. Each protocol was repeated three (for kvp) or ten (for 8 and 1 kvp) times for the assessment of precision. A phantom with two adipose rings (representing an 8-kg patient) was also investigated in our study. This phantom, referred to as the large patient, was scanned with a routine abdominal protocol corresponding to its size, with the range of parameters expanded similarly to those of the small patient. Table 2 illustrates the full parameter space explored for small and large patients in this study. Note that the clinical dose level for each phantom size corresponded to a predefined noise index, not a fixed dose. As a result, the % dose level for the large phantom (24. mgy) was much higher than the % dose level for the small patient (4. mgy). Data Analysis A customized code (MATLAB, Mathworks) was used for data analyses. To quantify the image noise of each dataset, spherical regions of interest (ROIs) were placed on the uniform muscle region, with the SD within the ROI recorded. To quantify the iodine concentration, the contrast enhancement of the lesions was measured from slices across the center of the lesions in each dataset (around 6 mm thick), with spherical ROIs placed on both high-contrast lesions and the nearby liver parenchyma to record the mean number of Hounsfield units within each ROI. Only the largest 1.-cm lesions were used, because those lesions provided the highest number of pixels TABLE 1: Target Attenuation Designed for Liver Parenchyma, High-Contrast Lesion, and Low-Contrast Lesion at Four Different Tube Voltages for best quantification. Spherical ROIs had a diameter of 1 cm to ensure sufficient confidence margins as well as sufficient number of pixels for statistical analyses (Fig. 2). The precision of contrast was further calculated in terms of repeatability coefficient. Repeatability coefficient represents the expected absolute difference between any two repeated quantifications of the same contrast, for 9% of cases [12]. Therefore, a lower repeatability coefficient (RC) indicates better precision and is calculated as follows: where Target Attenuation (HU) Tube Voltage (kvp) Liver Parenchyma Low-Contrast Lesion High-Contrast Lesion 14 6 a 73 b 79 b RC j = σWj 2 = 2.77σWj 2.77σˆ Wj (1), σˆ Wj a 111 b 1 b a Data are from 6 consecutive patients who underwent clinically indicated dual-energy CT of the liver at our institution. b Data are theoretically designed Hounsfield unit numbers of the low- and high- contrast lesions, respectively, corresponding to.4 and.8 mg/ml additional iodine concentration. n = WMS j = WMS ij / n = n K i = 1 k = 1 i = 1 (C ijk C ij ) 2 n(k 1) (2). WMS j is the estimate of the within-repeats variance (σ 2 Wj) for the contrast quantified from the jth protocol, K is the number of repeats for the jth protocol, C ijk is the contrast of the ith lesion measured from images acquired with jth protocol and kth repeat, C ij is the contrast of the ith lesion measured from images acquired with jth protocol, and averaged over all repeats, and n is the number of slices analyzed in each dataset. The two iterative reconstructions, ASIR and MBIR, were further assessed for their dose-reduction ability in terms of precision. This ability was defined as the ratio between the dose required by iterative reconstruction and the dose required by FBP to achieve the same threshold precision. Thus, given the same patient size, tube voltage, and slice thickness, if iterative reconstruction and FBP methods require A and B milligrays of dose, respectively, to achieve a threshold precision, the dose reduction ability of the iterative method was computed as (1 [A / B]) %. Note that this calculation relied on a proper choice of the threshold precision (i.e., repeatability coefficient) to represent the acceptable fluctuation in Hounsfield unit count, which was chosen to be HU in this study because it represents the minimum perceptible enhancement by human observers [11]. This calculation also required doses A and B to be properly interpolated from the predetermined dose levels described in Table 2. To do so, repeatability coefficient values of each reconstruction algorithm were fitted as a function of dose, with A and B interpolated accordingly from the fits. Results Phantom Images and Noises Images of the liver phantom under some typical imaging and reconstruction protocols are shown in Figure 3. Overall, thicker slice, higher dose, and iterative reconstruction show reduced noise. Compared with FBP, ASIR images slightly reduce the noise without obvious texture change. MBIR significantly reduces the image noise but also presents a waxier texture. In addition to the visual assessment, the noise was numerically assessed and listed TABLE 2: Scanning and Reconstruction Protocols Determined According to Routine Protocols at Our Institution for Specific Patient Sizes and Further Expanded to Investigate the Effect of Iterative Reconstruction Under Various Circumstances Phantom Size Dose Level Tube Voltage (kvp) Small % (noise index, 14.; CTDI, 4. mgy), 7%, %, %, and % Large % (noise index, 24.; CTDI, 24. mgy), 7%, %, %, and % Reconstruction Algorithm Slice Thickness (mm) No. of Repetitions 1,, and 8 FBP, ASIR, and MBIR.6 and 2. 3 ( kvp) and (8 and 1 kvp) 14, 1, and FBP, ASIR, and MBIR.6 3 Note CTDI = CT dose index, ASIR = adaptive statistical iterative reconstruction, FBP = filtered back projection, MBIR = model-based iterative reconstruction. AJR:, May 13 W477

4 in Table 3. A thicker slice reduces the noise by around 3%. ASIR reduces the noise by around %. MBIR reduces the noise by around 6% at clinical dose level and further reduces the noise by 8% at % of the clinical dose level. Different tube voltages show similar image noise in this study, because the tube current had been adjusted to maintain the same CTDI Fig. 2 Small patient phantom. A, Sample slice was acquired at 8 kvp,.6-mm slice thickness, and 4.-mGy CT dose index. B, To quantify iodine concentration (Hounsfield unit numbers) of lesions in panel A, spherical regions of interest (red circles) were placed on both 1.-cm high-contrast lesions and nearby liver parenchyma, with mean Hounsfield unit numbers recorded. for different tube voltages under the same patient size. Different patient sizes also show similar image noise, because the tube current has been adjusted to maintain similar noise indexes for the two patient sizes (small patient, 12. HU; large patient, 14. HU). Contrast The tissue contrast of the hyperenhancing lesions to the nearby liver parenchyma is shown in Figure 4. The contrast seems to be largely independent of the choice of reconstruction algorithm and dose level. Lower tube voltage yields higher contrast, as expected. Effects of Imaging and Reconstruction Parameters on Precision On the basis of the measured contrast, the precision of iodine concentration was further calculated. Overall, the repeatability coefficient decreased as dose increased, indicating a smaller variance (i.e., a better precision) Fig. 3 Sample images of liver phantom. Five columns represent four combinations of acquisition parameters and phantom size. Three rows represent three reconstruction algorithms applied to same acquisition dataset. Display window and level are 3 and 8. ASIR = adaptive statistical iterative reconstruction, CTDI = CT dose index, FBP = filtered back projection, MBIR = model-based iterative reconstruction. W478 AJR:, May 13

5 Iterative Reconstruction Algorithms in Hepatic CT TABLE 3: Noise of Images Acquired for Two Patient Sizes With Protocols of Various Slice Thicknesses, Reconstruction Algorithms, and Dose Levels Protocol, Reconstruction Algorithm Small patient phantom 2.-mm slice thickness 8 kvp Dose Levels % % FBP ASIR 4 MBIR 7 1 kvp FBP 9 18 ASIR MBIR mm slice thickness, 1 kvp FBP 98 ASIR MBIR Large patient phantom,.6-mm slice thickness, 1 kvp FBP ASIR MBIR Note In this article, precision is a second-order statistic that cannot be visually assessed from the image. Therefore, Figure 3 and Table 3 should not be interpreted as bases for judging the precision, but rather as illustrations of the visual attributes of the iterative reconstruction algorithms used in this study. ASIR = adaptive statistical iterative reconstruction, FBP = filtered back projection, MBIR = model-based iterative reconstruction. at higher dose. As mentioned in the Materials and Methods, the precision was empirically fitted to a function of dose. One example is illustrated in Figure A, where the repeatability coefficient values of a protocol with 1 kvp, 2.-mm slice thickness, small patient, and all three reconstruction algorithms are plotted as a function of dose. Inspired by the decreasing trend, which is pronounced at an extremely low dose (< 1 mgy), still visible at low dose (1 2. mgy), but levels off at high dose levels (> 2. mgy), a function that closely fitted the data points was found to be RC = a [b / (Dose. + c)] (FBP, R 2 =.98; ASIR, R 2 =.9; MBIR, R 2 =.97). Other protocols not presented in Figure A also showed similar dose dependence, and the application of the function to all curves showed an average R 2 =.94 ±.8. Figure A also shows the effect of reconstruction algorithm. At the highest dose, which represents currently used clinical dose level, the precision of iterative reconstruction was not much different from that of FBP. As dose decreases, the two iterative reconstruction algorithms show distinct performances: ASIR remains similar to FBP, whereas MBIR shows a curve significantly lower and flatter than ASIR and FBP do (about % reduction in repeatability coefficient), indicating improved precision. Overall, the superior performance of MBIR confirms its dose reduction potential by providing the same quantification precision at a much lower dose, which was further quantified in the next section. The effect of reconstruction algorithm on quantification precision was further tested with thinner slices, where the quantum noise greatly increased. Figure B illustrates the precision of images with.6-mm slice thickness. Comparing Figure B to Figure A, which represents 2.-mm slice thickness, the precision for thinner slice thickness greatly deteriorates for all reconstruction algorithms, with an increase of % in repeatability coefficient. The advantage of MBIR, however, is more pronounced under such high image noise conditions. The effect of iterative reconstructions was also tested at 8 and kvp. Lower tube voltage is appreciated in hepatic imaging because it greatly enhances the lesion contrast; however, lower tube voltage also raises the baseline of image Hounsfield unit number, along with a possible byproduct of larger fluctuations in Hounsfield unit number. Figure 6 plots the precision for 8-, -, and 1-kVp protocols, with all other parameters fixed at.6-mm slice thickness and small patient. Results indicate that 8 kvp leads to higher repeatability coefficient (i.e., worse precision for FBP and ASIR). However, for MBIR, the precision is relatively stable across all tube voltages. Therefore, with MBIR, it is possible to get both enhanced image contrast and improved quantification precision at lower tube voltage. Finally, the comparisons between reconstruction algorithms were made at two distinct patient sizes, with other parameters fixed at 1 kvp and.6-mm slice thickness (Fig. 7). The repeatability coefficient values are slightly higher for the large patient but expected because the highest dose level involved in the large patient corresponded to a noise index of 24. HU, whereas the highest dose level in the small patient corresponded to a noise index of only 14. HU. In other words, the image noise in the large patient was slightly higher, which slightly increased the fluctuations in quantifications. In spite of the general higher repeatability coefficient, the advantage of MBIR is preserved with the large patient size. Dose Reduction Potential of Iterative Reconstructions ASIR showed a precision performance very similar to that of FBP, where the dose reduction potential was not statistically significant. Therefore, only MBIR was assessed for its dose-reduction potential. According to the fitted function of each repeatability coefficient dose relationship, the dose corresponding to HU repeatability coefficient was interpolated for each protocol, as listed in Table 4. The relative dose reduction potential of MBIR was further calculated and also listed in Table 4. A dose reduction level of at least 4% was achieved with MBIR across all protocols tested in this study. Discussion The precision of iodine quantification affects the usefulness of the quantification by affecting the reliability of a change observed in successive measurements. In this article, we focused AJR:, May 13 W479

6 4 4 FBP, 8 kvp FBP, kvp FBP, 1 kvp 4 4 ASIR, 8 kvp ASIR, kvp ASIR, 1 kvp 4 4 MBIR, 8 kvp MBIR, kvp MBIR, 1 kvp Contrast (HU) A B C FBP, 2. mm ASIR, 2. mm MBIR, 2. mm 4 on achieving high quantitative precision with iterative reconstructions, which significantly reduced the noise and, hence, the noise-induced Hounsfield unit uncertainty. The overall results indicated that ASIR, one of the two iterative reconstruction algorithms tested in this study, had a limited gain in precision as compared with that of FBP. MBIR, the other iterative reconstruction algorithm, significantly improved precision and reduced the dose required for a reliable quantification. This information may serve as a guide when future protocol optimizations are tuned toward iodine quantification and patient dose reduction. Dose showed its strong effect on quantitative precision. The precision of quantification improved as dose increased but plateaued at high dose levels, highlighting the importance of sufficient but not excessive dose to diminish quantification uncertainty. Therefore, we not only compared the precision of iterative reconstruction and FBP at a Contrast (HU) FBP,.6 mm ASIR,.6 mm MBIR,.6 mm 4 given dose level but also compared the dose required by iterative reconstruction and FBP to achieve a given precision. With the routine protocol used clinically at our institute (Fig. A), ASIR showed moderate improvement in precision. MBIR significantly improved the precision of quantification and was able to achieve the same precision at a 2% lower dose as compared with FBP. This finding has important clinical implications in that, for most patients, MBIR may be able to maintain the quantification precision with a significant dose reduction. For example, when attempting to differentiate simple cysts, hemorrhagic cysts, papillary carcinomas, and clear cell carcinomas in the kidneys, the determination of contrast-enhanced enhancement is critical. This measurement is typically performed by comparing the attenuation of the mass in both unenhanced and contrast-enhanced datasets, the former of which is often obtained with a low radiation dose. Contrast (HU) Fig. 4 Tissue contrast of lesions with higher iodine concentration to nearby liver parenchyma. A C, Contrast was measured at images reconstructed from filtered back projection (FBP) (A), adaptive statistical iterative reconstruction (ASIR) (B), and model-based iterative reconstruction (MBIR) (C). Each subplot further contains measurements from three peak kilovoltages and five dose levels. CTDI vol = volume CT dose index. A B Fig. Precision (repeatability coefficient) of all three reconstruction algorithms. A, Graph shows repeatability coefficient fitted as function of dose, with other parameters fixed at small patient, 1 kvp, and 2.-mm slice thickness. Overall, higher dose led to better precision. ASIR = adaptive statistical iterative reconstruction, CTDI vol = volume CT dose index, FBP = filtered back projection, MBIR = model-based iterative reconstruction. B, Graph shows impact of reconstruction algorithms on precision with.6-mm slice thickness. The effect of iterative reconstruction was further investigated with thinner slice thickness, lower tube voltage, and larger patient sizes to ensure that the advantage of iterative reconstruction is generalizable. We chose this broad range of protocols for the following reasons: first, although thicker slices were typically used in the clinic to facilitate the workflow (i.e., fewer images to review) and reduce image noise, slices as thin as.6 mm are used routinely to reconstruct off-axis images, typically in the coronal plane, as well as CT angiographic images; second, lower tube voltage has recently been promoted for small patients to reduce both the iodine and radiation dose and to improve image quality [13 ]; and third, larger patient sizes represent a significant portion of the U.S. population and are raising concerns because the increased scattering in larger patients might deteriorate image quality and compromise diagnostic interpretation. With thinner slice thickness, the precision W48 AJR:, May 13

7 Iterative Reconstruction Algorithms in Hepatic CT A A FBP, 1 kvp ASIR, 1 kvp MBIR, 1 kvp 4 FBP, small patient ASIR, small patient MBIR, small patient 4 B B FBP, kvp ASIR, kvp MBIR, kvp 4 FBP, large patient ASIR, large patient MBIR, large patient C FBP, 8 kvp ASIR, 8 kvp MBIR, 8 kvp Fig. 6 Impact of iterative reconstructions on precision at different peak kilovoltages. A C, Graphs show precision investigated at 1 kvp (A), kvp (B), and 8 kvp (C). Other parameters were fixed at.6-mm slice thickness and small patient size. ASIR = adaptive statistical iterative reconstruction, CTDI vol = volume CT dose index, FBP = filtered back projection, MBIR = model-based iterative reconstruction. 4 Fig. 7 Impact of iterative reconstructions on precision by patient size. A and B, Graphs show precision at small (A) and large (B) patient size. Other parameters were fixed at 1 kvp and.6-mm slice thickness. Note that highest dose levels of two patient sizes correspond to different noise index, with 14. HU for small patient and 24. HU for large patient. ASIR = adaptive statistical iterative reconstruction, CTDI vol = volume CT dose index, FBP = filtered back projection, MBIR = model-based iterative reconstruction. TABLE 4: Threshold Dose for All Protocols and Patient Sizes and Relative Dose Reduction Ability of Model-Based Iterative Reconstruction (MBIR) Derived According to Threshold Dose Dose, Protocol Threshold dose (mgy) a Small Patient Large Patient 8 kvp kvp 1 kvp kvp 1 kvp 14 kvp FBP.6 mm b 26. b 3.7 b 2. mm MBIR.6 mm mm.7. <. c Relative dose reduction (%), MBIR vs FBP.6 mm mm > 2 d Note Dashes indicate no data available. FBP = filtered back projection. a The threshold dose is the CT dose index corresponding to a repeatability coefficient of HU. b Even the maximum dose tested in this study did not fulfill the precision requirement, so the threshold dose was extrapolated rather than interpolated from the function. c Even the minimum dose tested in this study fulfilled the precision requirement, so the threshold dose was expressed as the minimum dose tested, conservatively. d To be conservative, the dose reduction ability was calculated on the basis of the minimum dose tested in this study, which already fulfilled the precision requirement. AJR:, May 13 W481

8 is usually deteriorated by the increased image noise. However, the gain in precision with ASIR became more noticeable. An even more pronounced improvement was observed with MBIR. The precision of MBIR with.6 mm was not only superior to that of FBP with.6 mm but even comparable to the precision of FBP with 2.-mm slices. With lower tube voltage, the precision of both FBP and ASIR deteriorated significantly because higher Hounsfield unit values led to higher fluctuations in quantifications, but the precision of MBIR was stable, allowing the combination of better image contrast with better quantification. With a large patient, MBIR was significantly better than FBP and ASIR and was comparable to its precision with a small patient, given that enough dose was delivered to ensure similar noise level. Across all circumstances tested in this study, ASIR reduced noise by around 3% but did not improve quantitative precision by the same order (around % reduction in repeatability coefficient). Therefore, although ASIR was shown by previous studies to be very promising in detection tasks [], it did not bring great improvement to the quantification task used in this study. MBIR, however, reduced the noise by more than 6% and consistently improved precision (around % reduction in repeatability coefficient) with a dose reduction potential of 4% or more. This would lower the risk of radiation-induced cancer and, thus, would enable more frequent follow-up CT examinations and more pediatric examinations. This distinct behavior between ASIR and MBIR mentioned already may be partially attributed to the modeling process included in the MBIR reconstruction and partially to the fact that this ASIR reconstruction was % blended with FBP. The % blending ratio was chosen because it was commonly used at our institution to mitigate the waxy look of iterative reconstructed images. However, to our knowledge, this ratio is not standardized and can range from % to % in clinical practice. The precision of ASIR with blending ratio higher than % can be linearly approximated from the precision of FBP and % ASIR and is expected to provide better performance Although this study focused on the quantification of iodine concentration, the precision of iodine quantification is essentially the precision of Hounsfield unit quantification, as explained in the introduction to this article. Therefore, the dependency of precision on imaging and reconstruction parameters shown in this study can be directly applied to all Hounsfield unit quantification tasks, such as the detection of subtle hepatocellular carcinoma in the setting of cirrhosis, the assessment of diffuse or focal deposition of fat or iron in the liver, or the characterization of cystic renal lesions. Furthermore, the dependency of precision indicated in this study might also provide insights on other types of Hounsfield unit based quantifications, such as the volume quantification of lung nodules based on intensity thresholding and the density quantification of lung airway that also heavily relies on Hounsfield units. The advantage of MBIR could be extended to this broader spectrum of quantifications. Nonetheless, this study has several limitations. First, only precision was considered as the criterion of quantitative process, whereas in clinical practices, many factors may affect the quantitative performance, such as the texture and resolution of the image. Therefore, the dose reduction percentage of MBIR deducted in this study might be overly optimistic, and studies from other perspectives are required to further evaluate it. Second, some protocols only have a limited number of repetitions (three times). Although this was somehow compensated by the curve fitting process, the statistical power of repeatability coefficient was diminished. Finally, the two adipose rings were uniformly added to the peripheral of the phantom to simulate larger patients, whereas in reality, adipose tissue can accumulate predominantly within the abdomen, particularly in men. To what extent this adipose distribution can affect the quantification remains the effort of future studies. In conclusion, this study provided a framework for the evaluations of iterative reconstructions in quantitative hepatic imaging. ASIR did not show much gain in quantitative precision when it was % blended with FBP, but it could achieve more with a higher percentage. MBIR, a new reconstruction algorithm, showed a strong improvement in the precision of iodine quantifications across various doses, slice thicknesses, tube voltages, and patient sizes, and achieved the same precision at a dose 4% less than that of FBP. References 1. Blomley MJ, Coulden R, Dawson P, et al. Liver perfusion studied with ultrafast CT. J Comput Assist Tomogr 199; 19: Miles KA, Hayball MP, Dixon AK. Functional images of hepatic perfusion obtained with dynamic CT. Radiology 1993; 188: Schramm P, Huang Y, Erb G, Klotz E, Heiland S. How does the injection protocol influence the attenuation-time curve in CT perfusion measurements: comparison of measured and simulated data. Med Phys 9; 36: Colsher JG, Jiang H, Thibault J-B, et al. Ultra low dose CT for attenuation correction in PET/CT. In: Sellin P, ed. Nuclear Science Symposium conference record, 8. Dresden, Germany: IEEE, 8:6 11. Marin D, Nelson RC, Schindera ST, et al. Lowtube-voltage, high-tube-current multidetector abdominal CT: improved image quality and decreased radiation dose with adaptive statistical iterative reconstruction algorithm initial clinical experience. Radiology ; 4: Thibault J-B, Sauer KD, Bouman CA, Hsieh J. A three-dimensional statistical approach to improved image quality for multislice helical CT. Med Phys 7; 34: Nelson RC, Feuerlein S, Boll DT. New iterative reconstruction techniques for cardiovascular computed tomography: how do they work, and what are the advantages and disadvantages? J Cardiovasc Comput Tomogr 11; : Katsura M, Matsuda I, Akahane M, et al. Modelbased iterative reconstruction technique for radiation dose reduction in chest CT: comparison with the adaptive statistical iterative reconstruction technique. Eur Radiol 12; 22: Husarik DB, Marin D, Samei E, et al. Radiation dose reduction in abdominal computed tomography during the late hepatic arterial phase using a modelbased iterative reconstruction algorithm: how low can we go? Invest Radiol 12; 47: Marin D, Nelson RC, Samei E, et al. Hypervascular liver tumors: low tube voltage, high tube current multidetector CT during late hepatic arterial phase for detection initial clinical experience. Radiology 9; 1: Maki DD, Birnbaum BA, Chakraborty DP, Jacobs JE, Carvalho BM, Herman GT. Renal cyst pseudoenhancement: beam-hardening effects on CT numbers. Radiology 1999; 213: Barnhart HX, Barboriak DP. Applications of the repeatability of quantitative imaging biomarkers: a review of statistical analysis of repeat data sets. Transl Oncol 9; 2: Nakayama Y, Awai K, Funama Y, et al. Lower tube voltage reduces contrast material and radiation doses on 16-MDCT aortography. AJR 6; 187:W49 W Yu L, Li H, Fletcher JG, McCollough CH. Automatic selection of tube potential for radiation dose reduction in CT: a general strategy. Med Phys ; 37: Schindera ST, Nelson RC, Mukundan S Jr, et al. Hypervascular liver tumors: low tube voltage, high tube current multi-detector row CT for enhanced detection phantom study. Radiology 8; 246:1 132 W482 AJR:, May 13