Feasibility of Iterative Model Reconstruction for Unenhanced

Transcription

1 This copy is for personal use only. To order printed copies, contact Yuji Iyama, MD Takeshi Nakaura, MD Ayumi Iyama, MD Masafumi Kidoh, MD Kazuhiro Katahira, MD Seitaro Oda, MD Daisuke Utsunomiya, MD Yasuyuki Yamashita, PhD Feasibility of Iterative Model Reconstruction for Unenhanced Lumbar CT 1 Purpose: Materials and Methods: To evaluate the image quality and interobserver reproducibility of unenhanced lumbar spinal computed tomography (CT) images reconstructed with iterative model reconstruction (IMR). This prospective study was approved by the local ethics committee, and written informed consent was obtained from all patients. The study included 34 patients scanned with unenhanced CT and magnetic resonance (MR) imaging for lumbar canal spinal stenosis. The CT images were reconstructed with filtered back projection (FBP), hybrid iterative reconstruction (HIR), and IMR. Image noise and contrast-to-noise ratio (CNR) were compared among the three reconstruction techniques with the repeated oneway analysis of variance. The interobserver agreement of the dural sac on all CT image sets and T2-weighted images was also compared. Qualitative analysis of the three reconstruction techniques was performed by using Friedman test and the Wilcoxon signed-rank test with Holm correction. Original Research n Musculoskeletal Imaging Results: Conclusion: The image noise of IMR was significantly lower than that of FBP or HIR (P,.001 and P,.001). Pearson correlation analysis showed that the highest correlation coefficient with interobserver agreement was with IMR (r = 0.98) followed by MR imaging (r = 0.88), FBP (r = 0.41), and HIR (r = 0.33). It also showed that the narrowest Bland-Altman limit of agreement was achieved with IMR followed by MR imaging, FBP, and HIR. The qualitative image score using IMR was significantly higher than that using FBP or HIR (P,.001 and P,.001). IMR offers excellent noise reduction, higher interobserver reproducibility of canal stenosis, and improved image quality compared with FBP and HIR. 1 From the Department of Diagnostic Radiology, Kumamoto Chuo Hospital, Tainoshima 1-5-1, Kumamoto , Japan (Y.I., K.K.); Department of Diagnostic Radiology, Graduate School of Medical Sciences, Kumamoto University, Kumamoto, Japan (Y.I., T.N., M.K., S.O., D.U., Y.Y.); and Department of Diagnostic Radiology, National Hospital Organization Kumamoto Medical Center, Kumamoto, Japan (A.I.). Received August 19, 2016; revision requested November 1; final revision received November 29; accepted December 14; final version accepted December 20. Address correspondence to Y.I. ( iyamayuuji28@ gmail.com). q RSNA, 2017 Online supplemental material is available for this article. q RSNA, 2017 Radiology: Volume 284: Number 1 July 2017 n radiology.rsna.org 153

2 Lumbar spinal stenosis is a narrowing of the central spinal canal, lateral recess, or neural foramen. Medical imaging such as magnetic resonance (MR) imaging or computed tomography (CT) enhanced by means of myelography can help diagnose lumbar spinal stenosis (1 4). The symptoms of lumbar spinal stenosis include low back pain, buttock pain, and leg pain and numbness. MR images (especially T2 weighted) and CT myelography images can clearly show the range and degree of compression of the dural sac (5,6). Myelography, however, is associated with complications such as headache, allergy to iodinated contrast agents, nerve injury from the spinal needle with bleeding around the nerve roots, and infection (7). In recent years, MR imaging has become the standard of reference in the diagnosis of lumbar spinal stenosis because of its potential in visualizing soft tissues, which we cannot evaluate on radiographic images. MR imaging also has disadvantages, such as a long imaging time and motion artifacts, meaning that the image quality might be of nondiagnostic quality in some cases. In addition, MR imaging may be contraindicated in patients with pacemakers; moreover, Advances in Knowledge nn The iterative model reconstruction (IMR) technique can significantly improve image quality compared with the filtered back projection (FBP) and hybrid iterative reconstruction (HIR) techniques on lumbar spinal CT images (image noise: IMR, 10.4 HU 6 2.4; FBP, 55.8 HU ; HIR, 38.9 HU 6 8.9; contrast-tonoise ratio: IMR, ; FBP, ; HIR, ). nn The IMR technique provides high interobserver reproducibility of dural sac cross-sectional area, comparable to that of MR imaging; in contrast, FBP and HIR might not provide enough image quality to evaluate the lumbar spinal canal. the presence of metal artifacts related to prior spinal instrumentation may decrease the accuracy of MR imaging in the diagnosis of postoperative spinal stenosis. In contrast, unenhanced CT has a short scan time and does not require a contrast agent. However, there have been few reports about the usefulness of unenhanced CT for diagnosis of lumbar spinal stenosis (8,9), possibly because the image noise associated with CT obscures any evaluation of the lumbar spinal canal. Recently introduced iterative reconstruction techniques for CT could reduce the quantum noise associated with filtered back projection (FBP) techniques (10 12). Previous reports suggested that a hybrid iterative reconstruction (HIR) technique could decrease image noise and provide better image quality on abdominal and thoracic scans compared with FBP techniques (13,14). In addition, iterative model reconstruction (IMR) (Philips Medical Systems, Cleveland, Ohio) was reported to be more useful than HIR for decreasing image noise and increasing the contrast-tonoise ratio (CNR) (15 17). To our knowledge, the usefulness of iterative reconstruction such as HIR or IMR for unenhanced lumbar spinal CT is not well established in the literature. We hypothesized that IMR could yield better image quality than FBP or HIR and that evaluation of the lumbar spinal canal via IMR technique has accuracy comparable to that of MR imaging. The purpose of our study was to evaluate the image quality and interobserver reproducibility of lumbar spinal canal stenosis on CT images reconstructed by using the IMR technique. Materials and Methods Our prospective study received institutional review board approval. Written informed consent to participate was obtained from all patients. Implication for Patient Care nn IMR of lumbar spinal CT images may be useful for accurately diagnosing lumbar canal stenosis. Patients Between March and July 2016, 34 patients who were suspected of having lumbar spinal canal stenosis or lumbar disc herniation, and who had undergone both preoperative MR imaging and unenhanced CT, were enrolled in our study. The patients were 16 men (mean age, 62.6 years; range, years) and 18 women (mean age, 63.9 years; range, years) aged years (mean age, 63.3 years) and weighing kg (mean, 62.2 kg). We compared the age between male and female patients. The maximum time interval between CT and MR imaging examinations was 30 days. MR imaging helped diagnose lumbar disc herniation in nine patients and lumbar spinal canal stenosis in 25 patients. The exclusion criteria were (a) presence of extensive vertebral fractures, which would result in difficult assessment of axial images, and (b) poor image quality because of postoperative metal artifacts or patient motion. No patients were excluded from our study on the basis of these exclusion criteria. CT Protocols All patients were scanned by using a 256-row multidetector CT scanner (Brilliance ict; Philips Healthcare, Content codes: Radiology 2017; 284: Abbreviations: CNR = contrast-to-noise ratio DCSA = dural sac cross-sectional area FBP = filtered back projection HIR = hybrid iterative reconstruction IMR = iterative model reconstruction Author contributions: Guarantor of integrity of entire study, T.N.; study concepts/ study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; agrees to ensure any questions related to the work are appropriately resolved, all authors; literature research, Y.I., T.N., M.K.; clinical studies, Y.I., A.I.; experimental studies, Y.I., A.I.; statistical analysis, Y.I., T.N., M.K.; and manuscript editing, Y.I., T.N., K.K., S.O., D.U., Y.Y. Conflicts of interest are listed at the end of this article. 154 radiology.rsna.org n Radiology: Volume 284: Number 1 July 2017

3 Table 1 Scan Parameters for the CT Protocol Parameter Value Tube voltage (kvp) 120 Field of view (cm) 30 Beam collimation (mm) Section thickness (mm) 0.9 Helical pitch CTDI vol (mgy) 15.6 Rotation time (sec) 0.75 Tube current time product Tube current (ma)* Scan time (sec) 10.8 DLP (mgy cm) Note. CTDI vol = volume CT dose index, DLP = doselength product. * Automatic exposure control on. Table 2 Scan Parameters for the Axial T2-weighted MR Imaging Protocol Parameter 3-T Imaging 1.5-T Imaging Sequence Axial T2 weighted Axial T2 weighted Repetition time (msec) Echo time (msec) Section thickness (mm) 4 4 Intersection gap (mm) 0 0 Flip angle (degrees) Matrix Field of view (mm) Section orientation Axial Axial Voxel size (mm 3 ) No. of acquisitions 2 2 Bandwidth (Hz) Mean acquisition time (sec) Cleveland, Ohio). All scans were obtained at 120 kvp, 0.75-second gantry rotation, and 127-mA tube current. The detailed scanning parameters are shown in Table 1. The patients were instructed to hold their breath, with deep inspiration during scanning. The total scan time was 10.8 seconds. CT Image Reconstruction Image reconstruction was performed in a cm display field of view. All images were reconstructed by using an FBP algorithm with a standard softtissue kernel (C), HIR (idose 4 ; Philips Medical Systems, Cleveland, Ohio) with a standard soft-tissue kernel (C), and IMR (soft-tissue level 1; Philips Medical Systems). We selected a hybrid iterative level 4 and an IMR level 1 as the recommended nominal settings by the vendor for soft-tissue imaging. We also used multiplanar reconstruction images of the lumbar spinal canal for quantitative image analysis using the picture archiving and communication system viewer (EV Client; PSP, Tokyo, Japan). MR Imaging All MR imaging examinations were performed by using a 3-T whole-body MR system (Ingenia; Philips Medical Systems, Best, the Netherlands) equipped with a dual-source, parallel, radiofrequency transmission system or 1.5-T whole-body MR system. Both MR imaging systems were used with a spinal coil. Patients were placed in the supine position with their arms raised. The following sequences were performed: sagittal T2-weighted and sagittal T1- weighted turbo spin-echo, axial T2- weighted, and coronal T2-weighted gradient-echo three-dimensional thinsection (1 mm) sequences. The section thickness of the axial T2-weighted images was 4 mm, with an intersection gap of 0.4 mm. Table 2 shows the parameters for the T2-weighted axial imaging. Quantitative Image Analysis A board-certified radiologist (T.N.) with expertise in orthopedic imaging and with 14 years of experience with lumbar spinal CT performed a quantitative image analysis on 0.9-mm-thick sagittal images with each reconstruction technique. The mean attenuation of the lumbar spinal canal was measured in a circular region of interest (ROI canal ) placed at the level of L4-5. This ROI canal was expected not to be so large that it included epidural fat or bone. If we could not measure the ROI canal because of severe canal stenosis, we selected another vertebral level (L1, L2, or L3). We also measured the attenuation of a lumbar disc in a circular region of interest (ROI disc ) placed at the levels of L3-4 and L4-5. This size of ROI disc was expected not to be so large that it included a vertebral body or the lumbar spinal canal. If we could not measure the ROI disc because of a severely thinned intervertebral disc, we selected another intervertebral level (L1-2, L2-3, or L3-4). The ROI canal and the ROI disc of 10 patients were measured at other intervertebral levels. The regions of interest of two patients were measured at the level of L1 2, those of another two patients were measured at the level of L2 3, and those of six patients were measured at the level of L3 4. To evaluate imaging noise, we measured the standard deviation of attenuation at the iliopsoas muscle. To minimize bias from single measurements, we measured imaging noise at three sequential sections and averaged the results. The CNR was calculated as follows: (ROI disc ROI canal )/ image noise. For all measurements, the size, shape, and position of the regions of interest were kept constant among the three reconstruction techniques by applying a copy-and-paste function at the workstation. Assessment of Interobserver Reproducibility for Evaluation of Dural Sac Cross-Sectional Area and Qualitative Image Analysis We performed the quantitative image analysis that included the MR images as the standard of reference. Radiology: Volume 284: Number 1 July 2017 n radiology.rsna.org 155

4 For assessment of interobserver reproducibility for evaluation of the dural sac cross-sectional area (DCSA) between MR images and 0.9-mm-thick axial CT images reconstructed with FBP, HIR, and IMR, two radiologists (Y.I. and A.I.) with expertise in orthopedic imaging and 7 years of experience with lumbar spinal CT and MR imaging performed the qualitative image analysis. Prior to the review, we prepared and stored 136 anonymized Digital Imaging and Communications in Medicine folders. Each contained axial CT images of one of our study subjects reconstructed with FBP, HIR, or IMR and axial T2-weighted MR images at the level of L4-5. The CT data sets were randomized. Qualitative image analysis and measurement of DCSA was performed with a soft-tissue window setting, window width of 300 HU, and window level of 30 HU by using a picture archiving and communication system viewer. Two readers measured the DCSA at the level of L4-5. All measurements were performed by manual selection of the evaluation area. To minimize the possibility of error, all measurements were performed three times, and the mean value was calculated. Adjustment of the window level and width was allowed during the qualitative assessment. The observers were blinded to the acquisition parameters; clinical symptoms; results of quantitative image analysis; and patient s age, sex, and follow-up clinical findings. The reading time was not limited. At the same time, two readers also evaluated the qualitative image analysis of the lumbar spinal canal using axial CT images reconstructed with FBP and HIR and MR images. They assessed the subjective image quality of the lumbar spinal canal stenosis using a four-point subjective scale as follows: score 4, excellent: no image noise with preserved spatial resolution, and the dural sac contour is smooth and clear, providing useful information for the lumbar canal stenosis; score 3, good: slight image noise with preserved spatial resolution, and the dural sac contour is clear, providing sufficient diagnostic information; Table 3 Quantitative Image Analysis Parameter FBP HIR IMR score 2, fair: image noise is present, and the dural contour is partially obscured due to image noise and impaired spatial resolution, providing acceptable diagnostic information; score 1, poor: image is very noisy or small with impaired spatial resolution, yielding insufficient diagnostic information. When the two observers disagreed, the qualitative image score was decided on consensus after discussion. Statistical Analysis Statistical evaluation was performed by using free statistical software (R statistical package, R Project for Statistical Computing; org/). All numerical values are reported as means 6 standard deviation. The Kolmogorov-Smirnov test was performed to test normal distribution. The Kolmogorov-Smirnov test indicates that the age, CNR, and image noise distribution is normal. We have compared the age between male and female patients using unpaired t test. We compared the CNR and image noise of each technique with the repeated one-way analysis of variance. If there was a statistically significant difference among the techniques, we performed pairwise comparisons using the paired t test with the Holm correction. For qualitative analysis, first the Friedman test was performed to test for significant differences in subjective image quality between the three reconstruction techniques. Differences with P,.05 were considered to indicate statistical significance. Second, the Wilcoxon signed-rank test with Holm correction was performed to compare the subjective score for lumbar canal stenosis between the three reconstruction techniques. The scale for the k coefficient was defined as follows: less than 0.40, poor; , intermediate to good; and , excellent agreement as previously suggested (18). The interobserver reliability of the measurement of DCSA on FBP, HIR, IMR, and T2- weighted MR images was analyzed by calculation of the Pearson correlation analysis. Agreement was considered poor if the Pearson coefficient (r) was less than 0.6, good if the coefficient was , and excellent if the coefficient was greater than 0.8. A Bland-Altman analysis was also performed on the FBP, HIR, IMR, and T2-weighted MR images. For the Bland-Altman plot, relative differences in measurements were calculated as the difference between the two measurements divided by the mean. Then, intermeasurement variability was defined as the 95% confidence interval of the relative differences (19). The 95% confidence interval of the relative difference was calculated as a mean relative difference In addition, we performed regression analysis to investigate the possible proportional bias involving the three reconstruction techniques and MR imaging. Results P Value (FBP vs HIR) P Value (FBP vs IMR) P Value (HIR vs IMR) Image noise (HU) ,.001*,.001*,.001* CNR *,.001*,.001* Note. Pairwise comparisons were performed by using the paired t test with Holm correction. Data are means 6 standard deviations. * Significant results. Patients There was no significant difference in age between men and women (P =.81). Quantitative Image Analysis The results of the quantitative image analysis are shown in Table 3. The image noise of IMR was significantly lower 156 radiology.rsna.org n Radiology: Volume 284: Number 1 July 2017

5 than that of FBP and HIR (P,.001 for both). The image noise of HIR was significantly lower than that of FBP (P,.001). The CNR of IMR was significantly higher than that of FBP and HIR (P,.001 for both). The CNR of HIR was significantly higher than that of FBP (P =.040). Assessments of Interobserver Reproducibility for Evaluation of DCSA and Qualitative Image Analysis At visual evaluation, the IMR score was significantly higher than the HIR and FBP score: median IMR, 4 (interquartile range, 3 4); median FBP, 1 (interquartile range, ); median HIR, 2 (interquartile range, ) (P,.001 and P,.001)]. The visual score of HIR was significantly higher than that of FBP (P,.001). There were substantial interobserver agreements regarding visual score (k = 0.68). The Pearson correlation analysis of interobserver agreement of DCSA between the two readers showed excellent correlation coefficient for IMR (r = 0.98) and MR imaging (r = 0.88). However, the correlation coefficients for FBP (r = 0.41) and HIR (r = 0.33) were poor. The Bland-Altman limits of agreement using IMR (mean difference 1.7 mm ; 95% limits of agreement, 13.7 to 17.2 mm 2 ) were narrower than those using MR imaging (11.3 mm ; 95% limits of agreement, 31.7 to 54.4 mm 2 ), which we regarded as the standard of reference. In addition, the Bland-Altman limits of agreement using IMR were narrower than those using FBP (13.8 mm ; 95% limits of agreement, 55.0 to 82.6 mm 2 ) and HIR (17.9 mm ; 95% limits of agreement, 63.8 to 99.6 mm 2 ). Regression analysis showed that there was no proportional bias in IMR (slope, ; P =.64) and MR imaging (slope, ; P =.57); on the other hand, there was a significant proportional bias in FBP (slope, ; P =.02) and HIR (slope, ; P =.03). Therefore, on FBP and HIR images, there was lower agreement between the two readers for larger mean DCSA values. The Bland-Altman comparisons of the two readers for measuring DCSA are shown in Figure 1. Representative cases are shown in Figure 2 and Figure E1 (online). Discussion Results of our study suggest that IMR could improve the image quality and reduce the image noise compared with the FBP and HIR techniques. In addition, the IMR technique yielded higher interobserver agreement than the FBP or HIR techniques. Moreover, interobserver agreement for the IMR technique is comparable with that for MR imaging. In the quantitative image analysis, the IMR technique improved image quality of unenhanced lumbar spinal CT better than the FBP and HIR techniques. The result is consistent with the results of previous reports in other regions (15 17). FBP and HIR techniques were based on several simplifying assumptions about the CT system: the point source focal spot, linear x- ray beam, and CT detector cells with uniform responses. In contrast, the IMR technique is based on accurate system models to produce imaging data that truly correspond to the measured projection data (20). These methods might dramatically reduce the image noise compared with the HIR technique and cause the results in our study. In the quantitative image analysis, the introduction of IMR in lumbar spinal CT was also very useful. Interobserver agreement in the evaluation of DCSA with IMR and MR imaging was higher than that with FBP or HIR. MR imaging is an excellent tool for evaluating soft tissues of ligaments, spinal cord, discs, and dura matter (4). Therefore, we can identify the dura matter and measure it accurately and achieve high interobserver agreement. IMR can decrease image noise and increase the CNR better than FBP and HIR according to the accurate system model. Therefore, we could identify the water density and margins of the dural sac accurately and achieve high interobserver agreement. In contrast, the image noise of HIR is 3.7 times higher than that of IMR (38.9 HU vs 10.4 HU 6 2.4), and the image noise of FBP is 5.4 times higher than that of IMR (55.8 HU vs 10.4 HU 6 2.4). We cannot measure the narrow area such as DCSA accurately on images reconstructed with FBP or HIR because of high image noise. For qualitative image analysis, the IMR technique had a higher qualitative image score than did the HIR and FBP techniques. The reason might be that readers can identify more accurately the soft tissue such as the disc and water density in the dural sac. A previous report suggested that CT could help evaluate not only bone spurs but also soft tissue (9). Our results suggest that the IMR technique might help with the identification of soft tissue at CT. To our knowledge, the usefulness of iterative reconstruction such as for unenhanced lumbar spinal CT is not well established in the literature. The most important finding in our study was that both the qualitative and quantitative studies showed that the IMR technique could increase image quality better than FBP or HIR. IMR also provided high interobserver reproducibility of DCSA, which was comparable to that of MR imaging. In contrast, FBP and HIR might not produce good enough image quality to evaluate the lumbar spinal canal because of the high image noise. Therefore, IMR is useful for evaluating the lumbar spinal canal by using unenhanced lumbar CT. Our study findings suggest the feasibility of IMR CT for high interobserver reproducibility of canal stenosis comparable to MR imaging. However, IMR CT might not replace MR imaging in the assessment of spinal stenosis and radiculopathy for most patients, because unenhanced CT cannot be used to evaluate the lumbar spine and spinal root. However, IMR CT might be useful for patients with a contraindication to MR imaging. In such a situation, IMR CT may be a valid alternative to CT myelography. In addition, IMR CT may be useful for patients with metallic implants. Our study had limitations. First, the small size of the patient sample could Radiology: Volume 284: Number 1 July 2017 n radiology.rsna.org 157

6 Figure 1 Figure 1: Graphs show interobserver agreement in the DCSA using the Bland-Altman technique. (a) CT images reconstructed with FBP. (b) CT images reconstructed with HIR. (c) CT images reconstructed with IMR. (d) T2-weighted MR images. It also shows the narrowest Bland- Altman limits of agreement with IMR (mean difference, 1.7 mm ; 95% limits of agreement, 13.7 to 17.2 mm 2 ) followed by MR imaging (11.3 mm ; 95% limits of agreement, 31.7 to 54.4 mm 2 ), FBP (13.8 mm ; 95% limits of agreement, 55.0 to 82.6 mm 2 ), and HIR (17.9 mm ; 95% limits of agreement, 63.8 to 99.6 mm 2 ). The mean difference line is solid and 95% limits of agreement lines are dashed. limit the generalization of our results. Second, we used only one parameter for setting the middle sharpness and middle noise reduction during reconstruction using IMR, although the IMR technique offers nine settings for reconstruction parameters. However, a comparison of all parameters with other reconstruction techniques would be extremely difficult because it is statistically complicated. Third, it is possible that, because of the unique appearance of the images, the readers were not completely blinded when they were presented with the IMR images. Fourth, we did not evaluate the diagnostic performance of lumbar spinal canal stenosis among the three reconstruction techniques. Fifth, in our study, the radiation dose might affect the image quality and interobserver 158 radiology.rsna.org n Radiology: Volume 284: Number 1 July 2017

7 Figure 2 Figure 2: CT and T2-weighted MR images in a 77-year-old woman (body weight, 42 kg) with severe lumbar canal stenosis at the level of L4-5. (a) Axial CT image reconstructed with FBP. (b) Axial CT image reconstructed with HIR. (c) Axial CT image reconstructed with IMR. (d) Axial T2-weighted MR image. FBP and HIR images cannot depict the lumbar dural sac clearly because of high image noise. MR and IMR images clearly depict the lumbar dural sac. reproducibility of three reconstruction techniques. At this radiation dose (dose-length product = mgy cm), the IMR technique can reduce image noise while yielding higher interobserver reproducibility of canal stenosis compared with FBP and HIR. However, a previous report suggested that the merits of iterative reconstruction, especially for low-contrast objects, can be lost at extremely low radiation doses (21). Therefore, our study cannot provide proof that IMR can improve the image quality of the spinal canal compared with FBP and HIR at any scan setting. Sixth, we performed conventional quantitative analysis such as for image noise in our study. Many reports have evaluated image noise by using the standard deviation of CT attenuation (22,23). However, it is based on the assumption of a linear reconstruction algorithm. Therefore, to regard image noise as a standard deviation may not be inadequate for nonlinear reconstruction algorithms such as IMR. Last, unenhanced CT images cannot evaluate the lumbar spine or spinal root. In conclusion, IMR offers excellent noise reduction, high interobserver reproducibility of canal stenosis, and improved image quality compared with that produced with FBP or HIR. Disclosures of Conflicts of Interest: Y.I. disclosed no relevant relationships. T.N. disclosed no relevant relationships. A.I. disclosed no relevant relationships. M.K. disclosed no relevant relationships. K.K. disclosed no relevant relationships. S.O. disclosed no relevant relationships. D.U. disclosed no relevant relationships. Y.Y. disclosed no relevant relationships. Radiology: Volume 284: Number 1 July 2017 n radiology.rsna.org 159

8 References 1. Epstein NE, Epstein JA, Carras R, Hyman RA. Far lateral lumbar disc herniations and associated structural abnormalities: an evaluation in 60 patients of the comparative value of CT, MRI, and myelo-ct in diagnosis and management. Spine 1990;15(6): Schnebel B, Kingston S, Watkins R, Dillin W. Comparison of MRI to contrast CT in the diagnosis of spinal stenosis. Spine 1989;14(3): Bolender NF, Schönström NS, Spengler DM. Role of computed tomography and myelography in the diagnosis of central spinal stenosis. J Bone Joint Surg Am 1985;67(2): Wittenberg RH, Lütke A, Longwitz D, et al. The correlation between magnetic resonance imaging and the operative and clinical findings after lumbar microdiscectomy. Int Orthop 1998;22(4): Modic MT, Masaryk T, Boumphrey F, Goormastic M, Bell G. Lumbar herniated disk disease and canal stenosis: prospective evaluation by surface coil MR, CT, and myelography. AJR Am J Roentgenol 1986;147(4): Tsuchiya K, Katase S, Aoki C, Hachiya J. Application of multi-detector row helical scanning to postmyelographic CT. Eur Radiol 2003;13(6): Sandow BA, Donnal JF. Myelography complications and current practice patterns. AJR Am J Roentgenol 2005;185(3): Tallroth K. Plain CT of the degenerative lumbar spine. Eur J Radiol 1998;27(3): Eun SS, Lee HY, Lee SH, Kim KH, Liu WC. MRI versus CT for the diagnosis of lumbar spinal stenosis. J Neuroradiol 2012;39(2): Vorona GA, Ceschin RC, Clayton BL, Sutcavage T, Tadros SS, Panigrahy A. Reducing abdominal CT radiation dose with the adaptive statistical iterative reconstruction technique in children: a feasibility study. Pediatr Radiol 2011;41(9): Silva AC, Lawder HJ, Hara A, Kujak J, Pavlicek W. Innovations in CT dose reduction strategy: application of the adaptive statistical iterative reconstruction algorithm. AJR Am J Roentgenol 2010;194(1): Singh S, Kalra MK, Hsieh J, et al. Abdominal CT: comparison of adaptive statistical iterative and filtered back projection reconstruction techniques. Radiology 2010;257 (2): Song JS, Lee JM, Sohn JY, Yoon JH, Han JK, Choi BI. Hybrid iterative reconstruction technique for liver CT scans for image noise reduction and image quality improvement: evaluation of the optimal iterative reconstruction strengths. Radiol Med (Torino) 2015;120(3): Kligerman S, Mehta D, Farnadesh M, Jeudy J, Olsen K, White C. Use of a hybrid iterative reconstruction technique to reduce image noise and improve image quality in obese patients undergoing computed tomographic pulmonary angiography. J Thorac Imaging 2013;28(1): Iyama Y, Nakaura T, Yokoyama K, et al. Impact of knowledge-based iterative model reconstruction in abdominal dynamic CT with low tube voltage and low contrast dose. AJR Am J Roentgenol 2016;206(4): Iyama Y, Nakaura T, Yokoyama K, et al. Lowcontrast and low-radiation dose protocol in cardiac computed tomography: usefulness of low tube voltage and knowledge-based iterative model reconstruction algorithm. J Comput Assist Tomogr 2016;40(6): Yuki H, Oda S, Utsunomiya D, et al. Clinical impact of model-based type iterative reconstruction with fast reconstruction time on image quality of low-dose screening chest CT. Acta Radiol 2016;57(3): Fleiss JL. Statistical methods for rates and proportions. 2nd ed. New York, NY: Wiley, 1981; Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res 1999;8(2): Brown KM, Zabic S, Koehler T. Acceleration of ML iterative algorithms for CT by the use of fast start images. In: Pelc NJ, Nishikawa RM, Whiting BR, eds. Proceedings of SPIE: medical imaging 2012 physics of medical imaging. Vol Bellingham, Wash: International Society for Optics and Photonics, 2012; Solomon J, Mileto A, Ramirez-Giraldo JC, Samei E. Diagnostic performance of an advanced modeled iterative reconstruction algorithm for low-contrast detectability with a third-generation dual-source multidetector CT scanner: potential for radiation dose reduction in a multireader study. Radiology 2015;275(3): Leschka S, Stolzmann P, Schmid FT, et al. Low kilovoltage cardiac dual-source CT: attenuation, noise, and radiation dose. Eur Radiol 2008;18(9): Goldman LW. Principles of CT: radiation dose and image quality. J Nucl Med Technol 2007;35(4): ; quiz radiology.rsna.org n Radiology: Volume 284: Number 1 July 2017