Iterative scatter correction for grid-less skeletal radiography allows improved image quality equal to an antiscatter grid in adjunct with dose reduction: a visual grading study of 20 body donors

Abstract

Background

Iterative scatter correction (ISC) is a new technique applicable to plain radiography; comparable to iterative reconstruction for computed tomography, it promises dose reduction and image quality improvement. ISC for bedside chest X-rays has been applied and evaluated for some time and has recently been commercially offered for plain skeletal radiography.

Purpose

To analyze the potential of ISC for plain skeletal radiography with regard to image quality improvement, dose reduction, and replacement for an antiscatter grid.

Material and Methods

A total of 385 radiographs with different imaging protocols of the pelvis and cervical spine were acquired from 20 body donors. Radiographs were rated by four radiologists. Ratings were analyzed with visual grading characteristics (VGC) analysis. The area under the VGC curve was used as a measure of difference in image quality.

Results

Without ISC, the grid-less images were rated significantly worse than their grid-based counterparts (0.389, P = 0.005); adding ISC made image quality equal (0.498; P = 0.963). In grid-less imaging, reduction of dose by 50% led to significant image quality impairment (0.415, P = 0.001); this was fully counterbalanced when ISC was added (0.512; P = 0.588).

Conclusion

ISC for plain skeletal radiography has the ability to replace the antiscatter grid without image quality impairment, to improve image quality in grid-less imaging, and to reduce patient radiation dose by 50% without substantial loss in image quality.

Keywords

Skeletal – appendicular skeletal – axial radiation safety digital radiography computer-assisted image processing technology assessments

Introduction

Plain radiography is an essential diagnostic tool for skeletal imaging. When X-rays penetrate an object, scattered radiation is generated which leads to deterioration of image quality with respect to contrast and signal-to-noise ratio (1). The scattered radiation recorded by the X-ray detector is the background signal accompanied by noise, superimposed on top of the primary image signal. The scatter signal causes the image to appear flat, with the risk of obscuring diagnostic information (2). This effect increases with greater object diameters. That is why plain radiography of proximal body parts in adults should be done with an antiscatter grid (3,4). However, grid-based images have several drawbacks, e.g. higher acquisition dose (factor 2 or 3) (1) when detector dose is kept equal. In addition, positioning of the grid is time-consuming in the case of free exposures, e.g. for critically ill patients in the intensive care unit (ICU). Finally, the enhancing effect turns into the opposite in cases of inaccurate grid positioning (5). This may lead to artifacts which interfere with diagnostic details. At worst, the radiograph has to be retaken, doubling patient dose (6). Because of these disadvantages, there are circumstances (i.e. ICU or emergency room) in which grid-less imaging is usually carried out. In case of bedside chest X-ray imaging recent studies showed that iterative scatter correction (ISC) software (SkyFlow®, Philips Healthcare, Hamburg/Germany) improves the image quality of grid-less images (2,7,8). So far there are no published results for this technique applied to skeletal plain radiography.

The purpose of this study was to evaluate the potential of software-based ISC for grid-less skeletal radiography with regard to image quality improvement, dose reduction, and supersedure of an antiscatter grid.

Material and Methods

Specimens

The database consisted of plain radiographs from 20 body donors fixed in paraformaldehyde (10 women, 10 men; 7 with obese build and 13 with normal build). Before death, all donors had given signed consent to dedicate their bodies to research. The local Ethics Committee gave approval for data analysis (No. 379/16, 8 December 2016).

Image acquisition and processing

Radiographs were acquired on a portable flat detector at a tube voltage of 70 kVp for spine imaging and 80 kVp for pelvis imaging (MobileDiagnost, Philips Healthcare, Hamburg, Germany). The employed stationary anti-scatter grid had a ratio of 8, strip frequency 40 l/cm (Philips Healthcare, Hamburg, Germany). For visual presentation all images were subject to identical standard post-processing (UNIQUE®, Philips Healthcare, Hamburg, Germany). As a measure for detector dose, the standardized exposure index (EI) (9,10) was recorded for every image and arithmetic averages were calculated. Under calibration conditions, the EI represents detector dose in μGy multiplied with 100.

Free exposure images with antiscatter grid were taken of the pelvis (anteroposterior [AP] view) and cervical spine (2 planes); examples are shown in Fig. 1. Every radiograph was then repeated without antiscatter grid while other conditions remain constant. Afterwards the grid-less radiographs were twice again acquired, with reduced mAs for 37% and 50% dose reduction, respectively. This was done by deduction of two or three exposure points, respectively. On average, grid-based images were taken with 13.9 mAs (EI 224), grid-less standard dose images with 13.7 mAs (EI 609), grid-less 37% dose-reduced images with 8.6 mAs (EI 378), and grid-less 50% dose-reduced images with 6.9 mAs (EI 293). A scatter-corrected counterpart was calculated for every grid-less radiograph using ISC. Designations for the different imaging settings are shown in Table 1. The ISC software estimates the scatter signal using a physical scatter model based on pre-computed Monte-Carlo simulations (11) and partially subtracts it from the image. This simulates the contrast enhancement properties of a grid. Further technical details can be found in Mentrup et al. (8). Appropriate skeletal positioning is very difficult to achieve in cadaver specimens: five (1 × pelvis, 3 × spine AP, 1 × spine lat.) out of the 240 physically taken radiographs turned out to be inappropriate due to serious mistakes in body donor positioning. This lead to 35 excluded images; for statistical considerations, a complete set of seven images from each individual was necessary (A, B0, B0S, B2, B2S, B3, B3S). Eventually, 385 images were ready for further analysis.

Fig. 1.

Random image examples taken from the body donors. Most pelvis images had the artifact seen in (a) with a contrast step at the high of femoral neck, most likely as a result of body preservation; cervical spine images in AP view (b) sometimes revealed similar artifacts. (c) A limitation often occurring in lateral cervical spine images with shoulders overlaying lower cervical spine due to difficulties in body donor positioning.

Table 1.

Designations for the different imaging settings.

Designation	Description
A	Images with full dose and antiscatter grid
B0	Grid-less images with full dose
B0S	Grid-less images with full dose and ISC
B2	Grid-less images with 37% reduced dose
B2S	Grid-less images with 37% reduced dose and ISC
B3	Grid-less images with 50% reduced dose
B3S	Grid-less images with 50% reduced dose and ISC

Image quality evaluation

All images were evaluated in an absolute visual grading study by four observers with varying levels of radiological experience (two residents with five years and one resident with four years of experience as well as one final year student). The radiographs were presented and rated using the software ViewDEX 2.0 (12-14) on a calibrated medical-grade monitor (Coronis 3MP, Barco, Kortrijk, Belgium). This software was used in several recent studies (15–17). One single image was displayed at a time. The unlabeled images were presented in a unique random order for each reviewer. Zooming or adjusting the window level/width was not possible, as the setting should mimic a commonplace situation, e.g. a radiologist or surgeon quickly examines radiographs for fractures. Observers had to rate each image with the same quality criterion aligned to the European and German guidelines on quality criteria for diagnostic radiographic images (3,4): “Visually sharp reproduction of corticalis and spongiosa typical of the region as well as visually sharp reproduction of the bone contours near the joints.” A five-step rating scale was used based on Bath and Mansson (18), shown in Table 2.

Table 2.

Scale for rating the image quality criterion.

Rating	Description
1	Confident that the criterion is fulfilled
2	Somewhat confident that the criterion is fulfilled
3	Indecisive whether the criterion is fulfilled or not
4	Somewhat not confident that the criterion is fulfilled
5	Confident that the criterion is not fulfilled

Statistical analysis

An a priori power analysis was performed, using a two-tailed t-test for the difference between two dependent means. Assuming a type‐I error protection of 0.05 and an effect size of 0.4, a power of 0.8 produced a sample size of 52. Thus, each of the seven different groups (A, B0, B0S, B2, B2S, B3, B3S) should contain at least 52 pictures. We met this criterion, as every group had 55 pictures (7 × 55 = 385 overall images). The 1540 given ratings (four observers × 385 images) were on an ordinal scale; this indicated a non-parametric rank-invariant test. The visual grading characteristics (VGC) analysis fits this requirement (18–20) and was chosen as statistical method. In VGC analysis, the distributions of ratings for two conditions (a test condition on the y-axis and a reference condition on the x-axis) are used to calculate a VGC curve similar to a receiver operating characteristic (ROC) curve in ROC analysis. The area under the VGC curve (AUC_VGC) is used as a figure of merit in the range of 0.0–1.0. An AUC_VGC significantly > 0.5 indicates a higher image quality for the test condition. An AUC_VGC significantly < 0.5 indicates a lower image quality for the test condition. If AUC_VGC is not significantly different from 0.5, the two conditions cannot be separated statistically. The software used for calculation was VGC Analyzer (18,19,21,22). This is a statistical tool for fully crossed multiple-reader multiple-case VGC studies. It comes with a bootstrapping technique to determine the confidence interval (CI) of the AUC_VGC. In addition, it comes with a permutation resampling technique to determine the P value for testing the null hypothesis that the two compared systems are equal (H0: AUC_VGC = 0.5). A P value was considered significant at < 0.05. In this study, the AUC_VGC was based on a trapezoid VGC curve and random-reader model. Inter-rater reliability (IRR) was assessed using a two-way random, consistency, average-measures intraclass correlation coefficient (ICC) (23,24). This calculation was done with IBM SPSS Statistics for Windows, Version 20.0 (IBM Corp. Armonk, NY, USA).

Results

ICC was in the excellent range with 0.78 (25), indicating that observers had a high degree of agreement. The high ICC suggests that statistical power for subsequent analyses is not substantially reduced. Ratings were therefore deemed to be suitable for VGC analysis (26).

The results of the VGC analysis are presented in Fig. 2. Detailed numeric values are shown in Table 3. As an example, radiographs with different imaging settings of a body donor’s pelvis are shown in Fig. 3. Examples of plotted VGC curves are presented in Fig. 4.

Fig. 2.

AUC_VGC for each comparison. The dashed line represents equal image quality of the compared conditions. AUC > 0.5 favors the test condition, mentioned second in each pair comparison. Error bars represent the asymmetric 95% confidence interval. Dark bars stand for significant deviation from equal image quality between compared conditions (P < 0.05). Bright bars for comparisons without significant deviation from equal image quality. AUC_VGC, area under the visual grading characteristics curve; A, images with full dose and antiscatter grid; B0, grid-less images with full dose; B0S, grid-less images with full dose and ISC; B2, grid-less images with 37% reduced dose; B2S, grid-less images with 37% reduced dose and ISC; B3, grid-less images with 50% reduced dose; B3S, grid-less images with 50% reduced dose and ISC.

Table 3.

Values for AUC_VGC as a measure of difference in image quality between two compared imaging protocols, for all pair comparisons.

	AUC_VGC	SD	CI	P value
A–B0	0.39	0.04	0.31–0.47	0.01*
A–B0S	0.50	0.04	0.43–0.57	0.96
A–B2	0.34	0.05	0.25–0.43	<0.01*
A–B2S	0.44	0.04	0.36–0.51	0.08
A–B3	0.31	0.04	0.23–0.40	<0.01*
A–B3S	0.40	0.04	0.32–0.48	0.01*
B0–B0S	0.60	0.03	0.55–0.65	<0.01*
B0–B2	0.45	0.03	0.40–0.51	0.03*
B0–B2S	0.54	0.03	0.49–0.59	0.08
B0–B3	0.42	0.03	0.37–0.47	<0.01*
B0–B3S	0.51	0.03	0.46–0.57	0.59
B2–B2S	0.60	0.03	0.53–0.67	<0.01*
B3–B3S	0.60	0.03	0.54–0.66	<0.01*

*P value is significant (<0.05).

AUC_VGC, area under the visual grading characteristics curve; SD, standard deviation; CI, confidence interval; A, images with full dose and antiscatter grid; B0, grid-less images with full dose; B0S, grid-less images with full dose and ISC; B2, grid-less images with 37% reduced dose; B2S, grid-less images with 37% reduced dose and ISC; B3, grid-less images with 50% reduced dose; B3S, grid-less images with 50% reduced dose and iterative scattered correction.

Fig. 3.

As an example, three radiographs of a body donor’s pelvis, all taken with 80 kVp and 20 mAs: (a) with grid (EI 282); (b) grid-less with otherwise unchanged settings (EI 992); (c) grid-less with ISC (EI 992). Contrast loss caused by omitting the antiscatter grid appears to be compensated when the antiscatter technique is added. kVP, peak kilovoltage; mAs, milliamperage-seconds; EI, exposure index.

Fig. 4.

Plotted VGC curves for three comparisons, based on the ratings of the four observers. Forgoing the grid led to quality loss (a). This was fully compensated by adding ISC (b). If grid-less images before and after ISC-treatment were compared, a gain in image quality became apparent with ISC (c). These observations proved statistically significant (also see Table 3). A, images with full dose and antiscatter grid; B0, grid-less images with full dose; B0S, grid-less images with full dose and ISC.

Without application of ISC, grid-less images were rated significantly worse than the grid-based counterparts (A–B0 with AUC_VGC = 0.39, P = 0.01). In addition, significant impairment in image quality was seen when grid-less images were compared to their 50% dose-reduced counterparts (B0–B3 with AUC_VGC = 0.42, P < 0.01). A statistically significant difference in favor of ISC-processed images was observed compared to the grid-less images without ISC (B0–B0S with AUC_VGC = 0.60, P < 0.01). This proportion was stable when antiscatter post-processing was applied to grid-less images acquired with reduced mAs for 37% and 50% patient dose reduction, respectively (B2–B2S with AUC_VGC = 0.60, P < 0.01; B3–B3S with AUC_VGC = 0.60, P < 0.01).

The pair comparison B0–B2S (AUC_VGC = 0.54, P = 0.08) as well as B0–B3S (AUC_VGC = 0.51, P = 0.59) both showed no significant deviation from 0.5. This means that the gain in image quality with ISC fully compensated the quality loss that comes with 50% dose reduction.

The AUC_VGC for A–B0S with 0.50 (P = 0.96) demonstrated no significant deviation from 0.5. This suggests that with the use of ISC it is possible to omit the grid without impairing image quality. Furthermore, no significant deviation from 0.5 was observed between A and B2S (AUC_VGC = 0.44, P = 0.08). This implies that ISC enables 37% dose reduction without substantial impairment in image quality compared to grid-based images. Further dose reduction led to statistically significant impairment in image quality (A–B3S with AUC_VGC = 0.40, P = 0.01).

Discussion

To our knowledge, software-based ISC has not yet been evaluated for skeletal radiography. There are comparable recent studies for bedside chest X-ray imaging that demonstrated an enhancing effect of software-based scatter correction (2,7,8). The results of the present study suggest that the application of ISC to skeletal radiography has a positive impact on image quality. Moreover, previous studies were made exclusively with chest phantoms (2,7,8). Hence our results may be considered as a valuable contribution to transfer the findings on real human bodies. Mentrup et al. (8) showed contrast improvement with ISC equal to an antiscatter grid. This was done by using a quantitative approach under experimental perfect conditions. Renger et al. (7) evaluated the applied ISC technique with a slightly more practical approach by assessing the visibility of different catheters often used in the ICU. In contrast to Mentrup et al. (8) they could not fully reach performance of an antiscatter grid. Both calibrated grid-based and grid-less images to equal EI. This allowed grid-based images to be about twice as high in mAs, leading to double patient dose. Considering these unequal conditions, it seems questionable claiming ISC cannot reach image quality of an antiscatter grid in visual grading.

We followed another approach. Grid-based and grid-less images were not calibrated for equal EI but for equal patient dose. Thus, we compared antiscatter grid versus scatter correction software based on equal patient dose. Our approach coincides with a current study which indicates that the use of a grid improves diagnostic image quality without the need to increase tube mAs (27). If we would like to make the comparison analogous to the previous studies with equality in EI, then our pair comparison of grid-based images versus scatter-corrected 50% mAs-reduced grid-less images (A–B3S with EI225 and EI293, respectively) would approximately fit that approach. In this respect, our result is similar to Renger et al. (7): ISC images could not fully reach the image quality of grid-based images. However, we could reach grid-like image quality applying ISC to 37% mAs reduced grid-less images. Renger et al. (7) pointed out that their experiments were executed with nearly perfect tube-detector alignment and that grid image quality degradation due to misalignment, which often occurs in clinical practice, was not considered. In our study, tube-detector alignment with the body donors was inevitably sometimes difficult, close to circumstances in the ICU or trauma room. Investigating our grid-based images on a second look actually revealed a few grid-based images with typical artifacts caused by misalignment. Software-based ISC naturally does not have this shortcoming. This may be contributed to the performance of ISC in our study.

One limitation of our study is the sample size, which may be regarded small as the radiographs were obtained from only 20 body donors. However, it needs to be considered that body donors are not available in unlimited quantities. Only with body donors did we have the opportunity to repeatedly radiograph human skeletons for the purposes of study, without worrying about patient dose. Another limitation is the mentioned image artifacts due to human body preservation. Preparation is achieved by embalming with special chemicals (i.e. formaldehyde), resulting in changes in X-ray absorption. This effect can particularly be seen on most of our pelvis images that show a horizontal line at the high of the lower pelvis, causing a step in body density. This is probably due to imbalances in embalming chemicals according to body donor storage. Furthermore, interpretation of lateral view cervical spine images from body donors is restricted because the inevitably rigid joints lead to shoulders overlaying the lower cervical spine in lateral view.

It is most likely that the positive effect of software-based ISC is dependent on object size, analogous to an antiscatter grid, as the amount of scattered radiation increases with the diameter of the radiographed object. That is why our findings for pelvis and spine imaging are not transferable to radiographs of distal body parts, such as the wrist or ankle. But our results are probably transferable for other proximal body parts such as radiographs of the shoulder or femur. Further investigations should be done in this direction.

In conclusion, ISC for plain skeletal radiography has the ability to replace an antiscatter grid without image quality impairment, provided that patient radiation exposure is kept equal. A substantial radiation dose reduction in grid-less plain skeletal radiography is possible while preserving image quality. The ISC technique allows retrospective image quality improvement of given grid-less plain skeletal radiographs. It is assumed that the beneficial effects of ISC are especially present when radiographed body parts cause high amount of scatter radiation, such as the spine, pelvis, or proximal joints.

Footnotes

Acknowledgments

The authors thank Detlef Mentrup (Philips Healthcare, Hamburg, Germany) for providing the iterative reconstruction algorithm. They also thank Udo Schumacher (Department of Anatomy II, Experimental Morphology, University Hospital Hamburg-Eppendorf, Hamburg, Germany) for facilitating the image acquisition on body donors. The authors thank all the individuals who donated their bodies for the advancement of education and research.

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

Uffmann

Schaefer-Prokop

Digital radiography: the balance between image quality and required radiation dose.

Eur J Radiol 2009; 72:202–208.

Mentrup

Jockel

Neitzel

et al. Grid-like contrast restoration for non-grid chest radiographs by software-based scatter correction. Vienna: European Congress of Radiology (ECR), 2014.

European Commission. European guidelines on quality criteria for diagnostic radiographic images. 1st ed. Luxembourg: European Commission, 1996.

German Medical Association. Guidelines for quality assurance in X-ray diagnostics. 2007. Available at: http://www.bundesaerztekammer.de/fileadmin/user_upload/downloads/LeitRoentgen2008Korr2.pdf (last accessed 18 March 2018).

Wandtke

JC.

Bedside chest radiography.

Radiology 1994; 190:1–10.

Samei

Ranger

MacKenzie

et al . Effective DQE (eDQE) and speed of digital radiographic systems: an experimental methodology. Med Phys 2009; 36:3806–3817.

Renger

Brieskorn

Toth

et al . Evaluation of dose reduction potentials of a novel scatter correction software for bedside chest X-ray imaging. Radiat Prot Dosimetry 2016; 169:60–67.

Mentrup

Jockel

Menser

et al . Iterative scatter correction for grid-less bedside chest radiography: performance for a chest phantom. Radiat Prot Dosimetry 2016; 169:308–312.

Seibert

Morin

RL.

The standardized exposure index for digital radiography: an opportunity for optimization of radiation dose to the pediatric population.

Pediatr Radiol 2011; 41:573–581.

10.

International Electrotechnical Commission. Medical electrical equipment–Exposure index of digital X-ray imaging systems-Part 1: Definitions and requirements for general radiography. 1st ed. Geneva: IEC, 2008.

11.

Bertram

Hohmann

Wiegert

SU-FF-I-22: Scatter correction for flat detector cone-beam CT based on simulated sphere models.

Medical Physics 2007; 34:2342–2343.

12.

Borjesson

Hakansson

Bath

et al . A software tool for increased efficiency in observer performance studies in radiology. Radiat Prot Dosimetry 2005; 114:45–52.

13.

Hakansson

Svensson

Zachrisson

et al . VIEWDEX: an efficient and easy-to-use software for observer performance studies. Radiat Prot Dosimetry 2010; 139:42–51.

14.

Svalkvist

Svensson

Hakansson

et al . Viewdex: A status report. Radiat Prot Dosimetry 2016; 169:38–45.

15.

Soderman

Johnsson

Vikgren

et al . Effect of radiation dose level on accuracy and precision of manual size measurements in chest tomosynthesis evaluated using simulated pulmonary nodules. Radiat Prot Dosimetry 2016; 169:188–198.

16.

Farrugia Wismayer

Zarb

Radiography of the knee joint: A comparative study of the standing partial flexion PA projection and the standing fully extended AP projection using visual grading characteristics (VGC). Radiography 2016; 22:152–160.

17.

Ceder

Danielson

Kovac

et al . Thoracic spine imaging: a comparison between radiography and tomosynthesis using visual grading characteristics. Radiat Prot Dosimetry 2016; 169:204–210.

18.

Bath

Mansson

LG.

Visual grading characteristics (VGC) analysis: a non-parametric rank-invariant statistical method for image quality evaluation.

Br J Radiol 2007; 80:169–176.

19.

Bath

Hansson

Vgc Analyzer: a software for statistical analysis of fully crossed multiple-reader multiple-case visual grading characteristics studies.

Radiat Prot Dosimetry 2016; 169:46–53.

20.

Burmeister

Baltzer

Moslein

et al . Visual grading characteristics (VGC) analysis of diagnostic image quality for high resolution 3 Tesla MRI volumetry of the olfactory bulb. Acad Radiol 2011; 18:634–639.

21.

Bath

Evaluating imaging systems: practical applications.

Radiat Prot Dosimetry 2010; 139:26–36.

22.

Hansson

Mansson

Bath

The validity of using ROC software for analysing visual grading characteristics data: an investigation based on the novel software VGC analyzer.

Radiat Prot Dosimetry 2016; 169:54–59.

23.

McGraw

Wong

SP.

Forming inferences about some intraclass correlation coefficients.

Psychol Meth 1996; 1:30–46.

24.

Shrout

Fleiss

JL.

Intraclass correlations: uses in assessing rater reliability.

Psychol Bull 1979; 86:420–428.

25.

Cicchetti

DV.

Guidelines, criteria, and rules of thumb for evaluating normed and standardized assessment instruments in psychology.

Psychol Assess 1994; 6:284–290.

26.

Hallgren

KA.

Computing inter-rater reliability for observational data: an overview and tutorial.

Tutor Quant Methods Psychol 2012; 8:23–34.

27.

Moore

Wood

Avery

et al . Investigating the use of an antiscatter grid in chest radiography for average adults with a computed radiography imaging system. Br J Radiol 2015; 88:20140613.