Impact of iterative reconstruction on image quality of low-dose CT of the lumbar spine

Abstract

Background

Iterative reconstruction (IR) is a recent reconstruction algorithm for computed tomography (CT) that can be used instead of the standard algorithm, filtered back projection (FBP), to reduce radiation dose and/or improve image quality.

Purpose

To evaluate and compare the image quality of low-dose CT of the lumbar spine reconstructed with IR to conventional FBP, without further reduction of radiation dose.

Material and Methods

Low-dose CT on 55 patients was performed on a Siemens scanner using 120 kV tube voltage, 30 reference mAs, and automatic dose modulation. From raw CT data, lumbar spine CT images were reconstructed with a medium filter (B41f) using FBP and four levels of IR (levels 2–5). Five reviewers scored all images on seven image quality criteria according to the European guidelines on quality criteria for CT, using a five-grade scale. A side-by-side comparison was also performed.

Results

There was significant improvement in image quality for IR (levels 2–4) compared to FBP. According to visual grading regression, odds ratios of all criteria with 95% confidence intervals for IR2, IR3, IR4, and IR5 were: 1.59 (1.39–1.83), 1.74 (1.51–1.99), 1.68 (1.46–1.93), and 1.08 (0.94–1.23), respectively. In the side-by-side comparison of all reconstructions, images with IR (levels 2–4) received the highest scores. The mean overall CTDI_vol was 1.70 mGy (SD 0.46; range, 1.01–3.83 mGy). Image noise decreased in a linear fashion with increased strength of IR.

Conclusion

Iterative reconstruction at levels 2, 3, and 4 improves image quality of low-dose CT of the lumbar spine compared to FPB.

Keywords

X-ray computed tomography (CT)image manipulation/reconstruction spine lumbar vertebrae radiation dosage

Introduction

Iterative reconstruction (IR) is a computed tomography (CT) technique which can achieve radiation dose reduction and/or improve image quality (contrast, sharpness, and noise) (1) compared with traditional CT images reconstructed with filtered back projection (FBP).

Low-dose CT is a recent application of CT to examine the lumbar spine region instead of conventional radiography at a similar radiation level, i.e. around 1 mSv. This comes at the expense of increased image noise and reduced image quality compared with standard dose CT. According to a previous study, low-dose CT has superior image quality and gives a significant improvement in the anatomic and diagnostic information compared with radiography (2). In our institution, low-dose CT of the lumbar spine has now been implemented as routine instead of radiography, mainly for inpatients and patients referred from the emergency department. IR may improve the image quality of lumbar spine CT compared with FBP (3 –5). The aim of the current study was to evaluate the image quality of low-dose CT of the lumbar spine reconstructed with IR compared with FBP and to determine the optimal degree of IR, without further reduction of radiation dose.

Material and Methods

Patients

Patients who were referred for low-dose CT of the abdomen were invited to participate in the study. A power analysis indicated that if about 70% of cases had IR image quality better than the reference (FBP), 51 patients would be required to achieve 80% power. The estimation was that 55 patients needed to be included to compensate for dropouts. Exclusion criteria were: coma, dementia, age under 18 years, and inability to understand oral or written study information. The study was approved by the regional ethics board.

A convenience sample was collected during a period of 5 weeks. In total, 58 patients were asked to participate and three of them declined. Altogether 55 patients (22 men, 33 women) participated in the study. Mean age was 60 years (SD 21.2; age range, 20–94 years).

CT technique

The study was performed on a 2 × 128 channel Somatom Definition Flash CT scanner (Siemens, Erlangen, Germany). Based on settings used in a previous clinical study (2), the CT settings were: tube voltage, 120 kV; reference mAs, 30; collimation, 128 × 0.6 mm, and automatic dose modulation.

Image reconstruction

From raw CT data acquired for low-dose clinical abdominal CT, lumbar spine CT images were reconstructed for the current study. Therefore, patients were not subjected to any additional imaging or radiation. The reconstructions of lumbar spine images were performed with a medium filter (B41f) using FBP (6) and four different levels of IR, sinogram-affirmed iterative reconstruction (SAFIRE), the Siemens proprietary IR algorithm. Four levels of SAFIRE, IR2 (I41f 2), IR3 (I41f 3), IR4 (I41f 4), and IR5 (I41f 5), ranging from low to high noise reduction, were evaluated. The reconstructions were obtained in the axial and sagittal planes with 2 mm slice thickness and 2 mm increment. The lowest IR level (IR1) was excluded as it has a minimal effect on image quality compared with FBP. The settings were designed to deliver an effective dose of about 1 mSv as in previous phantom and clinical studies at our institution (2,6).

Image quality assessment

All images were stored and reviewed in a PACS. Five radiologists, with 9, 12, 13, 25, and 30 years of experience, reviewed the images, blinded to patient and image information.

In a first step, a randomized list of 275 CT reconstructions (axial and sagittal images), 55 patients by five reconstruction algorithms, was reviewed. All reviewers performed a visual grading for each reconstruction according to a modification of the European Guidelines on Quality Criteria for CT (EUR 16262) (7):

Criterion 1: sharp reproduction of cortical bone in the L3 vertebral body.

Criterion 2: sharp reproduction of trabecular bone in the L3 vertebral body.

Criterion 3: sharp reproduction of the intervertebral joints at the L3–L4 level on both sides.

Criterion 4: sharp reproduction of the intervertebral radicular canals at the L3–L4 level on both sides.

Criterion 5: sharp reproduction of the intervertebral disk profile at the L3–L4 level.

Criterion 6: reproduction of the paravertebral muscles (inter-muscular interface) at the L3–L4 level on both sides.

Criterion 7: acceptable noise level at L3–L4 level.

The reviewers scored each criterion as: 0, confident that the criterion is not fulfilled; 1, somewhat confident that the criterion is not fulfilled; 2, indecisive about whether the criterion is fulfilled or not; 3, somewhat confident that the criterion is fulfilled; or 4, confident that the criterion is fulfilled.

A second step was to perform a side-by-side visual grading of all five image reconstructions (FBP and IR2–IR5) for each patient in a list of 55 patients. Each reviewer compared all five reconstructions directly on two PACS monitors and ranked them from 1 to 5, 1 for worst perceived image quality and 5 for best image quality. The image reconstructions (FBP and IR2–IR5) were presented randomly for each patient.

Dose and noise estimation

The CTDI_vol for each patient was recorded. To estimate the noise, imaging of a Catphan phantom (Catphan® 504, The Phantom Laboratory, Salem, NY, USA) was performed with the same settings as for the clinical patients in the study. Images were reconstructed with FBP and IR2–IR5 and image noise was calculated as the average of ten manually selected regions of interest (ROI) in the uniformity module of the phantom.

Statistical analysis

Visual grading regression (VGR) (8), which is an ordinal logistic regression or proportional odds model, was used to compare the outcome image quality between different filter types. The outcome categories were combined as: 0, confident or somewhat confident that the criterion is not fulfilled; 1, indecisive; and 2, confident or somewhat confident that the criterion is fulfilled. Explanatory variables were filter type (using FBP filter as reference), reviewer, patient image, and image criteria. The regression was performed for all image criteria together, as well as for specific criteria, and also stratified by reviewer, the latter in order to examine the reviewer variability on image quality between different filter types. Ordinal regression was used to calculate odds ratios (ORs) with 95% confidence intervals (CIs) as the measure of association. An OR of 1 was interpreted as no difference of image quality between filter types; an OR > 1 was interpreted as meaning that the IR was rated better than FBP. If the 95% CI did not contain the value of 1.0, the association was statistically significant. All statistical analyses were performed using SPSS Statistics for Windows, version 22 (IBM Corp., Armonk, NY, USA).

Results

Image quality assessment

In the analysis of visual grading data collected from all reviewers of all criteria (C1–C7), there was a significant improvement of image quality with IR2, IR3, and IR4 compared with FBP, the reference. The same was found when the bone criteria (C1–C5) and the soft tissue criterion (C6) were analyzed separately. Overlapping confidence intervals indicated no significant differences of image quality between IR2, IR3, and IR4. Meanwhile, IR5 images were not significantly better than FBP images (Table 1). For the criteria cortical bone (C1) and trabecular bone (C2), the images with IR2 and IR3 had significantly better quality than FBP, while IR5 images were significantly worse than FBP (Table 1). An example of image quality of low-dose CT of the lumbar spine, reconstructed with FBP and IR2–IR5 filters, is shown in Fig. 1.

Table 1.

Scoring of image quality criteria for each reconstruction filter. The odds ratios (ORs) compare IR2–IR5 to filtered back projection (FBP) (the reference), according to ordinal logistic regression. An OR > 1 indicates that the test filter is better than FBP, while an OR < 1 indicates that the test is worse than FBP.

Criterion	Filter	Criterion not fulfilled (%)	Indecisive if fulfilled (%)	Criterion fulfilled (%)	Odds ratio (95% CI)	Interpretation
All criteria (C1–C7)	FBP	34	23	44	Reference
	IR2	26	25	50	1.59 (1.39–1.83)	S+*
	IR3	24	25	51	1.74 (1.51–1.99)	S+
	IR4	24	28	49	1.68 (1.46–1.93)	S+
	IR5	32	24	44	1.08 (0.94–1.23)	NS^†
Bone criteria (C1–C5)	FBP	27	23	50	Reference
	IR2	20	24	56	1.54 (1.30–1.82)	S+
	IR3	20	24	57	1.55 (1.31–1.83)	S+
	IR4	20	26	54	1.46 (1.24–1.73)	S+
	IR5	27	24	49	1.07 (0.80–1.11)	NS
Cortical bone (C1)	FBP	21	29	51	Reference
	IR2	17	23	61	1.83 (1.24–2.70)	S+
	IR3	17	24	60	1.62 (1.10–2.38)	S+
	IR4	19	29	52	1.20 (0.82–1.75)	NS
	IR5	33	22	45	0.55 (0.38–0.80)	S–*
Trabecular bone (C2)	FBP	45	22	34	Reference
	IR2	32	32	36	1.64 (1.14–2.34)	S+
	IR3	33	30	38	1.60 (1.12–2.29)	S+
	IR4	38	32	29	1.10 (0.77–1.58)	NS
	IR5	52	28	20	0.48 (0.33–0.69)	S–
Intervertebral joints (C3)	FBP	27	23	51	Reference
	IR2	19	22	59	2.47 (1.61–3.80)	S+
	IR3	18	23	59	2.52 (1.64–3.88)	S+
	IR4	18	25	58	2.30 (1.50–3.53)	S+
	IR5	22	27	52	1.48 (0.97–2.24)	NS
Intervertebral radicular canals (C4)	FBP	15	25	60	Reference
	IR2	13	25	63	1.33 (0.84–2.09)	NS
	IR3	12	23	66	1.72 (1.08–2.73)	S+
	IR4	11	23	66	1.97 (1.23–3.13)	S+
	IR5	14	25	61	1.09 (0.70–1.72)	NS
Disk profile (C5)	FBP	26	16	58	Reference
	IR2	20	19	60	1.77 (1.09–2.87)	S+
	IR3	19	20	62	2.11 (1.30–3.42)	S+
	IR4	14	20	66	3.98 (2.41–6.55)	S+
	IR5	16	16	68	4.22 (2.56–6.98)	S+
Soft tissue (C6)	FBP	62	24	14	Reference
	IR2	52	27	21	1.88 (1.31–2.69)	S+
	IR3	47	32	22	2.32 (1.62–3.32)	S+
	IR4	44	36	20	2.35 (1.64–3.36)	S+
	IR5	58	23	19	1.41 (0.98–2.03)	NS
Acceptable noise level (C7)	FBP	39	21	40	Reference
	IR2	29	24	47	1.99 (1.36–2.92)	S+
	IR3	25	22	53	3.03 (2.05–4.48)	S+
	IR4	22	28	50	2.92 (1.98–4.32)	S+
	IR5	28	27	44	1.81 (1.24–2.66)	S+

The test filter (IR) is significantly better/worse than the standard filter (FBP).

^†No statistical significance.

CI, confidence interval.

Fig. 1.

Axial reconstructions of low-dose computed tomography (CT) of the lumbar spine through the third lumbar vertebra with filtered back projection (FBP) and iterative reconstructions (IR) IR2–IR5 in a 28-year-old man.

Data for all criteria (C1–C7) and for bone criteria (C1–C5) were stratified according to reviewer. There was some variation among reviewers in the evaluation of image quality, mainly for IR3–IR5 but less for IR2 (Fig. 2). The same pattern of variability among reviewers was noted for bone criteria (C1–C5) after stratification of data according to reviewer (not shown).

Fig. 2.

Evaluation of all criteria (C1–C7) with iterative reconstructions IR2–IR5 for each reviewer (R1–R5). Filtered back projection (FBP) is used as reference. An odds ratio (OR) >1 indicates that the test filter is better than FBP, while an OR < 1 indicates that the test filter is worse than FBP.

In the side-by-side visual grading of all image reconstructions from worst to best, IR2, IR3, and IR4 images received the highest scores (Fig. 3).

Fig. 3.

Distribution of the scores of (best and second best) image quality according to five reviewers (R1–R5) in the side-by-side comparison of images reconstructed with all five filters (filtered back projection [FBP] and iterative reconstruction [IR] IR2–IR5). The total sum of the scores is 110.

Noise and dose estimation

The mean overall CTDI_vol was 1.70 mGy (SD 0.46; range, 1.01–3.83 mGy). The image noise, as measured in the Catphan phantom, was 19 HU in the FBP image and 15, 13, 11, and 9 HU for IR2, IR3, IR4, and IR5, respectively.

Discussion

In the present study, the effect of IR was tested to investigate whether IR could improve the image quality of an already established low-dose protocol for lumbar spine CT. The results showed that IR2, IR3, and even IR4 can improve the image quality significantly compared with standard FBP images. In the side-by-side visual grading there were more individual variations among reviewers in the grading of image quality. Most of the reviewers preferred IR2–IR4 to FBP, but reviewers varied regarding which IR level was judged the best (Fig. 3). This may be the result of side-by-side visual grading having a higher degree of subjective assessment since there was no systematic comparison of specific anatomical structures. Furthermore, reviewers could identify some differences in image quality and noise level when they compared images reconstructed by all five filters, side-by-side, on the same screen. However, both times, the assessments of image quality showed higher scores for IR compared to FBP images.

In CT of the cervical spine, Omoumi et al. found that the optimal strength level of IRs (SAFIRE, Siemens, Forchheim, Germany) for soft tissues and trabecular bone is lower while a higher level is better for the evaluation of the intervertebral discs and the neural foramina, i.e., the choice of IR level depends on the anatomical structure (9). They reported no improvement for cortical bone using IR compared with FBP. Our results for low-dose CT of the lumbar spine show a significantly improved image quality with IR (mainly IR2 and IR3, but also IR4) compared with FBP, with significantly higher scores for sharp reproduction of all selected anatomical structures, i.e. cortical bone, trabecular bone, intervertebral joints, intervertebral radicular canals, disk profile, and paravertebral muscles. Iterative reconstruction images also have a more acceptable overall noise level. Therefore, we can conclude that the optimal strength of IRs in low-dose CT of the lumbar spine seems to be independent of anatomical structures of the lumbar spine.

The CTDI_vol in the current study, 1.70 mGy (SD 0.46; range, 1.01–3.83 mGy), has an equivalent level compared with CTDI_vol of our previous study which was 2.03 mGy (SD 0.56; range, 1.22–3.85 mGy) (2). To our knowledge, few studies have been performed previously to evaluate the effect of IR on CT of the spine and there is no previous study that evaluates the use of IR in low-dose CT of the lumbar spine, at an effective dose of about 1 mSv.

Several studies have shown the benefits of IR in CT images. Most were designed to minimize the noise level and improve image quality compared to standard FBP. In a previous phantom study, IR (ADIR, Toshiba Medical Systems, Toshiba Corp., Tokyo, Japan) improved image quality significantly and was reported as having the potential to decrease the radiation dose compared with FBP (3). In another clinical study, the radiation level for full dose CT of the lumbar spine decreased after the use of IR (adaptive statistical iterative reconstruction; ASIR; GE Healthcare, Milwaukee, WI, USA). Dose levels measured as CTDI_vol (mGy) were 22.6 ± 8.2 (8.2–47.4) for images with reconstructed with FBP and 18.0 ± 6.3 (10.0–40.9) for images reconstructed with IR (4). A study by Yang et al. compared image quality and radiation level in lumbar spine CT in a control group of patients with full dose lumbar spine CT with FBP (tube voltage 120 and mAs 300) with two groups of low-dose CT examinations, at about 50% lower dose than the full dose CT using two CT settings (tube voltage 120 kV and 100 kV and tube loading 150 mAs and 230 mAs, respectively) which were reconstructed with IR (iDose4, Philips Healthcare, Cleveland, OH, USA). The result showed that the low-dose CT reconstructions with IR provided equivalent image quality and diagnostic reliability compared with full dose CT protocols with standard FBP. The CTDIvol (mGy) was 18.43 ± 5.17 for full dose group with FBP and 9.97 ± 3.67 and 7.33 ± 2.25, respectively, for the low-dose CT groups with IR (5).

In the literature, there are many other clinical studies about the impact of IR on CT image of other body parts than lumbar spine. A study showed that the use of IR (ASiR, GE Healthcare, Milwaukee, WI, USA) minimized the noise level and improved the diagnostic value in abdominal CT. Iterative reconstruction using IRIS (Siemens, Erlangen, Germany) improved image quality of low-dose chest CT in comparison with standard-dose CT with FBP, whereas yet another study showed that, in low-dose CT of the lungs, IR (SAFIRE, Siemens, Erlangen, Germany) allowed a dose reduction down to a submillisievert level, but still gave diagnostic image quality. Iterative reconstruction reduces not only the image noise but also the artifacts from calcifications, which improves diagnostic accuracy of coronary CT angiography in patients with heavily calcified coronary arteries. Routine use of IR for neuroradiology CT examinations (head, cervical, and intracranial angiography and lumbar spine) is recommended because of a significant dose reduction while preserving image quality.

Consistent with previous studies our measurements show that an increase in IR level gave a reduction of noise. Others have also shown that there is a shift in the noise spectrum towards lower frequencies (9 –11). This implies a shift towards an “over-smooth” appearance of the images at high IR levels.

The data used in the image quality evaluation are ordinal data types. Each image is graded based on one or more criteria by a number of observers who select a score reflecting image quality based on a specific criterion such as the visibility of a certain anatomical structure as the EU criteria (7). Visual grading regression is an ordinal logistics regression model that has recently increased in popularity as a useful statistical method for analyzing image quality and can be considered as an appropriate statistical method in image quality assessment studies (8,12).

There was obvious inter-observer variation in scoring which can be explained by the small differences in image quality between the various IR levels. Furthermore, visual grading analysis is based on a subjective assessment of image quality which could be affected by other factors such as rater experience and personal preference. For subjective evaluations, the fact that different observers have different opinions regarding image quality cannot be ignored (13). Furthermore, images reconstructed with strong levels of IR (SAFIRE) have an “over-smooth” appearance which may affect the image quality assessment depending on how familiar the reviewers are with that appearance.

A limitation of the current study is that the result is mainly based on a subjective image quality assessment. Another limitation is that pathological findings and diagnostic performance were not assessed. Third, we used IR from Siemens, while the impact of IR and its optimal level for low-dose CT of the lumbar spine may differ depending on the IR algorithm and software developed by different manufacturers.

In conclusion, IR 2, 3, or 4 (i.e. I41f2, 3, or 4) improved the image quality of low-dose CT of the lumbar spine, and should be included as routine.

Footnotes

Acknowledgments

We wish to thank the staff of the radiology department at Örebro University Hospital, with special thanks to M. Gustavsson and E. Holmvall.

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.

References

Grant K, Raupach R. SAFIRE: sinogram affirmed iterative reconstruction. Available at: http://usa.healthcare.siemens.com/siemens_hwem-hwem_ssxa_websites-context-root/wcm/idc/groups/public/@us/@imaging/@ct/documents/download/mdaw/ndq2/∼edisp/safire-00308312.pdf. (accessed 29 January 2016).

Alshamari

Geijer

Norrman

. Low dose CT of the lumbar spine compared with radiography: a study on image quality with implications for clinical practice. Acta Radiol 2016; 57: 602–611.

Gervaise

Osemont

Lecocq

. CT image quality improvement using Adaptive Iterative Dose Reduction with wide-volume acquisition on 320-detector CT. Eur Radiol 2012; 22: 295–301.

Komlosi

Zhang

Leiva-Salinas

. Adaptive statistical iterative reconstruction reduces patient radiation dose in neuroradiology CT studies. Neuroradiology 2014; 56: 187–193.

Yang

Chiou

. Imaging quality and diagnostic reliability of low-dose computed tomography lumbar spine for evaluating patients with spinal disorders. Spine J 2014; 14: 2682–2690.

Alshamari

Geijer

Norrman

. Low-dose computed tomography of the lumbar spine: a phantom study on imaging parameters and image quality. Acta Radiol 2014; 55: 824–832.

Bongartz G, Golding SJ, Jurik AG, et al. European Guidelines for Multislice Computed Tomography, European Commission. Available at: http://www.msct.eu/CT_Quality_Criteria.htm. (accessed December 2014).

Smedby

Fredrikson

. Visual grading regression: analysing data from visual grading experiments with regression models. Br J Radiol 2010; 83: 767–775.

Omoumi

Verdun

Salah

. Low-dose multidetector computed tomography of the cervical spine: optimization of iterative reconstruction strength levels. Acta Radiol 2014; 55: 335–344.

10.

Ghetti

Palleri

Serreli

. Physical characterization of a new CT iterative reconstruction method operating in sinogram space. J Appl Clin Med Phys 2013; 14: 4347–4347.

11.

Holmquist

Nyman

Siemund

. Impact of iterative reconstructions on image noise and low-contrast object detection in low kVp simulated abdominal CT: a phantom study. Acta Radiol 2016; 57: 1079–1088.

12.

Smedby

Fredrikson

De Geer

. Quantifying the potential for dose reduction with visual grading regression. Br J Radiol 2013; 86: 20110784–20110784.

13.

Obuchowski

. How many observers are needed in clinical studies of medical imaging? Am J Roentgenol 2004; 182: 867–869.