Usefulness of model-based iterative reconstruction in low-dose lumbar spine CT after spine surgery with metal implant: a phantom study

Abstract

Background

The major problems of computed tomography (CT) imaging include radiation exposure and severe artifacts caused by operative implants.

Purpose

To evaluate the usefulness of the metal artifact reduction algorithm and model-based iterative reconstruction (MBIR) in postoperative low-dose (LD) spine CT.

Material and Methods

A CT torso phantom was scanned at standard-dose (SD) and LD settings. The CT images were reconstructed by three methods: hybrid iterative reconstruction (HIR); metal artifact reduction; and MBIR. The radiation dose of the phantom imaging was evaluated by volume CT dose index (mGy), dose length product (DLP, mGy × cm), and effective dose (mSv). The image quality of the six images was visually evaluated and analyzed using Scheffe’s paired comparison method. The average preference of each method was calculated based on the comparative scores. The task transfer function (TTF) and noise power spectrum for HIR and MBIR were also measured.

Results

The respective radiation-dose-related parameters of the SD and LD conditions were: volume CT dose index = 10.2 and 1.2 mGy; DLP = 277.9 and 33.9 mGy × cm; and effective dose = 4.2 and 0.5 mSv. The average preference for diagnostic acceptability of MBIR at LD was not significantly different from the other reconstructions of SD data. MBIR successfully reduced metal artifacts in the LD condition. The 10% TTF was higher for HIR at SD and higher for MBIR at LD.

Conclusion

MBIR is useful for LD spine CT after spine surgery with metal implant.

Keywords

X-ray computed tomography metal artifact image reconstruction spine lumbar vertebrae radiation dose

Introduction

Computed tomography (CT) images provide detailed morphology and size evaluation of bony and calcified structures and are widely used for spine surgery planning and postoperative follow-up. Follow-up CT images with metal implants are usually acquired after spine surgery to assess the pedicle screw (PS) placement and check for PS loosening, implant failure, and bony union. The major problems with CT imaging are radiation exposure and the severe artifacts caused by postoperative implants.

Certain techniques may reduce metal artifacts, such as increased tube voltage and tube current. However, these approaches result in increased radiation dose to the patient and only limited improvement of image quality. Hybrid iterative reconstruction (HIR), metal artifact reduction (MAR), and model-based iterative reconstruction (MBIR) are emerging CT technologies to obtain low-dose (LD) images with less artifacts.

HIR, represented by AIDR 3D (adaptive iterate dose reduction 3D; Canon Medical Systems Co., Tochigi, Japan), is a noise reduction technique applied during reconstruction of LD images. MAR can be added during the reconstruction process without changing the scan conditions, and several vendors have developed dedicated MAR algorithms. The effects of MAR on reducing artifacts have been reported using SEMAR (single-energy MAR; Canon Medical Systems Co., Tochigi, Japan) and other four vendors’ MAR algorithm (1,2). MBIR, represented by FIRST (Forward projected model-based Iterative Reconstruction SoluTion; Canon Medical Systems Co., Tochigi, Japan) is also used for MAR (3). MBIR can reduce streak artifacts, improve spatial resolution, and dramatically reduce noise using a statistical noise model, an anatomical model, a system model, and a cone beam model.

Although the usefulness of processing LD CT images of lumbar spine and cervical spine with HIR have been previously reported (4,5), there is a paucity of data about the impact of MAR and MBIR in LD spine CT. The aim of this phantom study was to evaluate the reduction of metal artifacts and image quality of MAR and MBIR in comparison with HIR in postoperative standard-dose (SD) and LD spine CT.

Material and Methods

Phantom

A custom-designed CT torso phantom (CTU-41, Kyoto Kagaku Co., Ltd., Kyoto, Japan) was scanned for the study (Fig. 1). The head-to-pelvis phantom is 48 cm wide, 44 cm long, and 95 cm high and made of polyurethane (soft tissue) and epoxy resin (synthetic bone). The phantom contains simulated pulmonary vessels, heart, liver, trachea, and chest wall. The titanium alloy PSs (diameter = 6.5 mm) were embedded into the vertebral bodies (L3 and L5), not the pedicle of vertebral arch, because of the lack of strength of the phantom’s pedicle.

Fig. 1.

Custom-designed CT torso phantom CTU-41. (a) Photograph of the phantom. (b, c) Anteroposterior and lateral localizer images.

CT protocol and image reconstruction

All CT images were obtained using a 320-detector row CT scanner (Aquilion ONE ViSION Edition, Canon Medical Systems Co., Tochigi, Japan) in helical scan mode using the following parameters: detector configuration = 80 × 0.5 mm; tube voltage = 135 kVp; target noise level of tube current modulation with 5-mm slice using FC 13 (soft kernel), 14 HU, and 50 HU in the SD and LD CT conditions, respectively; gantry rotation time = 0.5 s; pitch factor = 0.813; and calibration field of view (FOV) = large.

A series of CT scans was reconstructed with slice thickness of 0.5 mm, increment of 0.4 mm, and FOV of 150 mm. SD and LD CT data were reconstructed with AIDR 3D (SD-HIR and LD-HIR), AIDR 3D with SEMAR (SD-SEMAR and LD-SEMAR) using a bone kernel (FC 30), and FIRST BODY SHARP (SD-MBIR and LD-MBIR). FIRST BODY SHARP is a high-resolution mode of BODY that targets the abdomen and pelvis. The AIDR 3D and FIRST methods were applied at the weak and standard noise reduction levels, respectively. Axial and sagittal multi-planar reconstructions with a thickness of 2 mm and increment of 2 mm were sent to the picture archiving and communication system (Figs. 2 –4).

Fig. 2.

Axial images at L2 level (without pedicle screws) in CT torso phantom reconstructed with (a) SD-HIR, (b) SD-SEMAR, (c) SD-MBIR, (d) LD-HIR, (e) LD-SEMAR, and (f) LD-MBIR. HIR, hybrid iterative reconstruction; LD, low-dose; MBIR, model-based iterative reconstruction; SD, standard-dose; SEMAR, single-energy metal artifact reduction.

Fig. 3.

Axial images at L3 level (with pedicle screws) in CT torso phantom reconstructed with (a) SD-HIR, (b) SD-SEMAR, (c) SD-MBIR, (d) LD-HIR, (e) LD-SEMAR, and (f) LD-MBIR. The angle of the axial image is adjusted to the L3 implant. Sharpness was evaluated at the spinous process (the arrow in (a)) and the ventral rim of the vertebral body (the arrowheads in (a)). There are black signals owing to metal artifacts around the screw, but in SD-MBIR (c) and LD-MBIR (f), the artifacts are less pronounced at both the spinous process and vertebral body. HIR, hybrid iterative reconstruction; LD, low-dose; MBIR, model-based iterative reconstruction; SD, standard-dose; SEMAR, single-energy metal artifact reduction.

Fig. 4.

Sagittal images of lumbar spine in CT torso phantom. reconstructed with (a) SD-HIR, (b) SD-SEMAR, (c) SD-MBIR, (d) LD-HIR, (e) LD-SEMAR, and (f) LD-MBIR. HIR, hybrid iterative reconstruction; LD, low-dose; MBIR, model-based iterative reconstruction; SD, standard-dose; SEMAR, single-energy metal artifact reduction.

Radiation dose

The radiation dose of the phantom imaging was evaluated by the volume CT dose index (mGy) and dose length product (DLP; mGy × cm). The effective dose (ED; mSv) was calculated by multiplying the DLP (mGy × cm) by a conversion factor (0.015 mSv × mGy⁻¹ × cm⁻¹).

Qualitative evaluation

From the six CT images, 15 pairs were assembled by Y.F., who was not involved in the image evaluation. Each pair of CT images was displayed side-by-side on a computer monitor in random order. All CT images were displayed with the window level and the width set to 500 and 2000, respectively. Three independent readers (Reader 1, M.H., a board-certified radiologist with 10 years of experience; Reader 2, S.F., a board-certified orthopedic surgeon with 27 years of experience; and Reader 3, K.M., a board-certified orthopedic surgeon with 11 years of experience) compared each pair of images, blinded in terms of dose and acquisition techniques. The qualitative evaluation points were based on those in a previously published report (6). The readers were asked to compare the image quality based on seven parameters: (i) subjective image noise; (ii) metal artifacts; (iii) metal contours; (iv) sharpness of spinous process in the slice with the PS; (v) sharpness of spinous process in the slice without the PS; (vi) sharpness of the ventral rim of the vertebral body (within a radius of 1 cm); and (vii) diagnostic acceptability, referring to the level of reader satisfaction with the diagnosis, which is derived from the overall image quality (Fig. 3). For each parameter, a comparative score was assigned to each image pair, as follows: 2 = the right image is definitely better; 1 = the right image is slightly better; 0 = the two images are equal; −1 = the left image is slightly better; and −2 = the left image is definitely better.

The scores assigned by the readers were analyzed by our original software using Scheffe’s paired comparison method with Nakaya’s variation, which is a modified form of Scheffe’s paired comparison method that excludes consideration of the ordering effect (7). The average preference of each method was calculated based on the scores, and the methods were compared with each other. We considered P values < 0.05 to be statistically significant.

Analysis of task-based transfer function and noise power spectrum

A TOS phantom with a diameter of 33 cm (Canon Medical Systems Co., Tochigi, Japan) was used for quantitative evaluation of spatial resolution and noise performance. As shown in Fig. 5a, this phantom was filled with water and contains five rods made of four different materials and one air cavity. Spatial resolution was evaluated in terms of the task-based transfer function (TTF) using the edges of the cylindrical contrast inserts in the phantom (8). The TTF provides a measure of how well the image transfers contrast across spatial frequencies, and frequencies associated with 50% and 10% of the TTF (50% TTF and 10% TTF, respectively) are recommended as indicators of spatial resolution. Noise performance was assessed in terms of the noise power spectrum (NPS) measured from the uniform section of the phantom. The NPS quantifies the noise signal’s frequency characteristics (9). Series of CT scans with two radiation dose settings (SD and LD: noise indexes of 14 HU, 220 mA and 50 HU, 30 mA, respectively) were reconstructed with a slice thickness of 0.5 mm and FOV of 150 mm. CT data acquired in the SD and LD conditions were reconstructed with AIDR 3D (SD-HIR and LD-HIR) using a bone kernel (FC 30) and with FIRST BODY SHARP (SD-MBIR and LD-MBIR).

Fig. 5.

Axial images of (a) TOS phantom, (b) insert with 350 HU in the TOS phantom and square ROI was used for task-based transfer function analysis, (c) water in the TOS phantom and square ROI was used for noise power spectrum analysis. ROI, region of interest.

We performed the NPS and TTF analyses with the “CT measure” image measurement software (10). To acquire the TTF, square regions of interest of 140 × 140 pixels were placed on inserts with averaged images totaling 350 HU across 10 CT images (Fig. 5b). The inserts were selected based on the CT values of lumbar vertebral bodies in the CT torso phantom. The TTF was calculated using the circular edge method (8). To acquire the NPS, as shown in Fig. 5c, a square region of interest of 256 × 256 pixels was placed on each reconstructed image. A total of 80 images was used to measure the NPS, which was calculated using a two-dimensional Fourier transform (9). An average NPS was calculated for each acquisition parameter. The peak frequency, f_P, and the average frequency, f_A, of noise texture were calculated as summary metrics.

Results

Radiation dose

The radiation-dose-related parameters of SD and LD were, respectively: volume CT dose index = 10.2 and 1.2 mGy; DLP = 277.9 and 33.9 mGy × cm; and ED = 4.2 and 0.5 mSv.

Qualitative evaluation

Fig. 6 summarizes the average preference results of the subjective image quality assessment. SD-SEMAR and SD-MBIR had higher overall average preference, whereas LD-SEMAR and LD-HIR had lower average preference. In particular, LD-SEMAR showed the lowest average preference for all parameters. SD-MBIR had the highest average preference for diagnostic acceptability. All average preference scores with statistical analysis are shown in the Supplemental material.

Fig. 6.

Average preferences for subjective image quality assessment. * indicates a significant difference (P value < 0.05) in comparison with low-dose CT data reconstructed with MBIR (LD-MBIR) using Scheffe’s paired comparison method.

We compared LD-MIBR with the other reconstruction methods and SD. LD-MBIR was not significantly different from LD-HIR, LD-SEMAR, or SD-HIR in terms of subjective image noise, and it was not significantly different from any of the reconstruction methods’ results using SD data in terms of metal artifacts or metal contours. Regarding the sharpness of the spinous process of the slices with and without the PS, LD-MBIR was inferior to any of the reconstructions using SD data but superior to LD-SEMAR and LD-HIR. LD-MBIR was also inferior to SD-MBIR and SD-SEMAR in terms of the sharpness of the ventral rim of the vertebral body, but it was comparable to SD-MBIR and superior to LD-SEMAR and LD-HIR. The diagnostic acceptability of LD-MBIR was significantly lower than that of SD-MBIR but not significantly different from that of other reconstructions using SD data.

Analysis of TTF and NPS

The TTF for each dose setting (SD or LD) and reconstruction method (HIR or MBIR) is shown in Fig. 7. The respective 50% and 10% TTF frequencies were: SD-HIR = 0.703 mm⁻¹ and 1.048 mm⁻¹; SD-MBIR =0.482 mm⁻¹ and 0.997 mm⁻¹; LD-HIR = 0.375 mm⁻¹ and 0.600 mm⁻¹; and LD-MBIR = 0.328 mm⁻¹ and 0.713 mm⁻¹. The 10% TTF value was higher for HIR at SD but higher for MBIR at LD. The 10% TTF value of HIR was 0.051 (5.1%) higher than that of MBIR at SD, but the corresponding MBIR value was 0.113 (18.8%) higher than that of HIR at LD.

Fig. 7.

Measured task-based transfer function for SD-HIR, SD-MBIR, LD-HIR, and LD-MBIR. HIR, hybrid iterative reconstruction; LD, low-dose; MBIR, model-based iterative reconstruction; SD, standard-dose.

The NPS for each dose setting (SD or LD) and reconstruction method (HIR or MBIR) is shown in Fig. 8. The NPS for MBIR was lower at medium–high spatial frequencies in both SD and LD conditions. In the LD condition, the peak of the noise distribution shifted in the low–frequency direction. The f_P values for SD-HIR, SD-MBIR, LD-HIR, and LD-MBIR were 0.48 mm⁻¹, 0.24 mm⁻¹, 0.24 mm⁻¹, and 0.16 mm⁻¹, respectively. The f_A values for SD-HIR, SD-MBIR, LD-HIR, and LD-MBIR were 0.54 mm⁻¹, 0.32 mm⁻¹, 0.27 mm⁻¹, and 0.21 mm⁻¹, respectively. Using MBIR greatly reduced the mid–high-frequency noise, and the peak of the NPS shifted in the low-frequency direction compared with HIR in both the SD and LD conditions.

Fig. 8.

Measured noise power spectrum for SD-HIR, SD-MBIR, LD-HIR, and LD-MBIR. HIR, hybrid iterative reconstruction; LD, low-dose; MBIR, model-based iterative reconstruction; SD, standard-dose.

Discussion

The results of the present study show that the use of MBIR in LD CT can reduce noise, reduce metal artifacts, and provide evaluable image quality. Masamoto et al. reported that thoracolumbar LD CT with MBIR can provide excellent images for measuring the pedicle diameter in preoperative CT (11). The current study demonstrates the possible utility of MBIR for LD postoperative lumbar spine CT with metal screws.

As shown in Fig. 3, both SD and LD CT images with MBIR showed less-pronounced metal artifacts at spinous processes and vertebral bodies. This is because MBIR reconstructs images by repeating the forward and back projections (12). During each iteration, the image is forward-projected into the sinogram and then compared with the measurement. An updated image is calculated by evaluating the mismatch between the forward projection and the original data.

LD-HIR and LD-SEMAR showed stronger metal artifacts and lower sharpness than the SD CT images on post-fixed spine phantom imaging. Generally, reducing dose results in noisy images with pronounced metal artifacts (13). When HIR or MBIR were used, the spatial resolution was reduced and the slice sensitivity profile was thickened dependent on both object contrast and image noise (8,14), resulting in blurred images and lower sharpness. The SEMAR algorithm automatically segments metal implants using the filtered back-projection original image for artifact reduction (15); however, its segmentation may be inaccurate when using LD scans, leading to low average preference.

The superiority of LD-MBIR to LD-HIR is explained by the TTF and NPS analyses. In the present study’s LD condition, the 10% TTF was slightly higher for MBIR than for HIR, although MBIR has similar noise dependence to HIR. The NPS was reduced over all frequency ranges > 0.1 mm⁻¹ when MBIR was used. These results are consistent with the previous paper that showed that MBIR can potentially make better use of projection data to reduce the CT radiation dose (16).

Our LD CT acquisition had a significantly reduced radiation dose compared with SD CTs (0.5 and 4.2 mSv for LD and SD, respectively). The effective dose of LD CT is less than one-third of that of a single X-ray, as the effective dose of a lumbar anteroposterior X-ray is 2.20 mSv, and that of a lumbar lateral X-ray is 1.5 mSv (17). Because multiple X-rays may be taken in a single examination (e.g. flexion–extension radiographs), a single CT examination may result in a smaller dose than an X-ray session. Although the image quality of LD spine CT is inferior to that of X-rays in terms of spatial resolution and metal artifacts, multi-planar reconstructions from LD CT provide 3D anatomical information that cannot be obtained from X-rays, such as the detailed positional relationships between bones and implants. Although the sharpness of the LD CT images is degraded compared with SD CT images, the LD spine CT images were acceptable for some purposes, such as postoperative assessments of implant position. In the future, it will be necessary to establish criteria to apply LD CT based on the scan region and examination purpose.

The present study has some limitations. First, we only assessed the image quality from a single CT manufacturer. Therefore, we have not evaluated the quality of CT images reconstructed with MBIR or MAR from other vendors. Second, this was a phantom study that did not include human subjects, so the possible effects of the geometry of metal implants and bone minerals inside the human body on our LD CT method are unknown. Third, the CT value of the insert used for TTF measurement was 350 HU, which is higher than the corresponding value of human lumbar spine, which is reported to be < 100 HU in patients aged > 70 years (18). We determined the value based on the CT value of lumbar spine in the current study’s CT torso phantom. Fourth, the MAR method using virtual monochromatic images generated from dual-energy CT was not compared with application of the same method to MBIR images using LD CT (6).

In conclusion, among the several ways to reduce metal artifacts, MBIR is unique in that it can effectively reduce metal artifacts in LD CT. Although the image quality is partially inferior to SD CT, the use of LD CT depending on the purpose of the CT examination could lead to a reduction in patient exposure.

Supplemental Material

sj-pdf-1-acr-10.1177_0284185120961424 - Supplemental material for Usefulness of model-based iterative reconstruction in low-dose lumber spine CT after spine surgery with metal implant: a phantom study

Supplemental material, sj-pdf-1-acr-10.1177_0284185120961424 for Usefulness of model-based iterative reconstruction in low-dose lumber spine CT after spine surgery with metal implant: a phantom study by Yasuhiro Fukushima, Akira Matsuda, Koji Koizumi, Maya Honda, Kazutaka Masamoto and Shunsuke Fujibayashi in Acta Radiologica

Footnotes

Acknowledgements

The authors thank Richard Lipkin, PhD, from Edanz Group () for editing a draft of this manuscript.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Yasuhiro Fukushima

Maya Honda

Supplemental material

Supplemental material for this article is available online.

References

Kidoh

Utsunomiya

Ikeda

, et al. Reduction of metallic coil artefacts in computed tomography body imaging: effects of a new single-energy metal artefact reduction algorithm. Eur Radiol 2016; 26: 1378–1386.

Bolstad

Flatabø

Aadnevik

, et al. Metal artifact reduction in CT, a phantom study: subjective and objective evaluation of four commercial metal artifact reduction algorithms when used on three different orthopedic metal implants. Acta Radiol 2018; 59:1110–1118.

Boudabbous

Arditi

Paulin

, et al. Model-based iterative reconstruction (MBIR) for the reduction of metal artifacts on CT. AJR Am J Roentgenol 2015; 205:380–385.

Becce

Ben Salah

Verdun

, et al. Computed tomography of the cervical spine: Comparison of image quality between a standard-dose and a low-dose protocol using filtered back-projection and iterative reconstruction. Skeletal Radiol 2013; 42:937–945.

Lee

Yun

Kim

, et al. Diagnostic usefulness of low-dose lumbar multidetector CT with iterative reconstruction in trauma patients: A comparison with standard-dose CT. Br J Radiol 2017; 90:20170181.

Jagoda

Schmitz

Wagenpfeil

, et al. Comparison of Metal Artifact Reduction in Dual-and Single-Source CT: A Vertebral Phantom Study. AJR Am J Roentgenol 2018; 211:1298–1305.

Nakaya

A modification of Scheffe’s method for paired comparisons. In: Proceedings of the 11th Meeting of Sensory Evaluation Convention. 1970:1–12.

Richard

Husarik

Yadava

, et al. Towards task‐based assessment of CT performance: system and object MTF across different reconstruction algorithms. Med Phys 2012; 39:4115–4122.

Boedeker

Cooper

McNitt-Gray

MF.

Application of the noise power spectrum in modern diagnostic MDCT: part I. Measurement of noise power spectra and noise equivalent quanta. Phys Med Biol 2007; 52:4027–4046.

10.

Ichikawa

CT measure. Tokyo: Japanese Society of CT Technology, 2012–2014. Available at: http://www.jsct-tech.org/.

11.

Masamoto K, Fujibayashi S, Otsuki B, et al. Utility of thoracolumbar low-dose CT with model-based iterative reconstruction for measuring pedicle diameter using a radiation dose less than a one-time lumbar X-ray. Spine 2020;45:38–47.

12.

Fujita

Higaki

Awaya

, et al. Lung cancer screening with ultra - low dose CT using full iterative reconstruction. Jpn J Radiol 2017; 35:179–189.

13.

Gjesteby

De Man

Jin

, et al. Metal artifact reduction in CT: where are we after four decades? IEEE Access 2016; 4:5826–5849.

14.

Goto

Tominaga

Taura

, et al. A method to measure slice sensitivity profiles of CT images under low-contrast and high-noise conditions. Phys Medica 2019; 60:100–110.

15.

Chang

Zamyatin

AA.

Metal artifact reduction algorithm for single energy and dual energy CT scans. IEEE Nucl Sci Symp Conf Rec 2012;3426–3429.

16.

Samei

Richard

Assessment of the dose reduction potential of a model-based iterative reconstruction algorithm using a task-based performance metrology. Med Phys 2015; 42:314–323.

17.

Simpson

Whang

Jonisch

, et al. The radiation exposure associated with cervical and lumbar spine radiographs. Clin Spine Surg 2008; 21:409–412.

18.

Schreiber

Anderson

Rosas

, et al. Hounsfield units for assessing bone mineral density and strength: A tool for osteoporosis management. J Bone Joint Surg Am 2011; 93:1057–1063.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.29 MB