Use of a computer simulator to investigate optimized tube voltage for chest imaging of average patients with a digital radiography (DR) imaging system

Abstract

Objective:

The aim of this study was to investigate via computer simulation a proposed improvement to clinical practice by deriving an optimized tube voltage (kVp) range for digital radiography (DR) chest imaging.

Methods:

A digitally reconstructed radiograph algorithm was used which was capable of simulating DR chest radiographs containing clinically relevant anatomy. Five experienced image evaluators graded clinical image criteria, i.e. overall quality, rib, lung, hilar, spine, diaphragm and lung nodule in images of 20 patients at tube voltages across the diagnostic energy range. These criteria were scored against corresponding images of the same patient reconstructed at a specific reference kVp. Evaluators were blinded to kVp. Evaluator score for each criterion was modelled with a linear mixed effects algorithm and compared with the score for the reference image.

Results:

Score was dependent on tube voltage and image criteria in a statistically significant manner for both. Overall quality, hilar, diaphragm and spine criteria performed poorly at low and high tube voltages, peaking at 80–100 kVp. Lung and lung nodule demonstrated little variation. Rib demonstrated superiority at low kVp.

Conclusion:

A virtual clinical trial has been performed with simulated chest DR images. Results indicate mid-range tube voltages of 80–100 kVp are optimum for average adults.

Advances in knowledge:

There are currently no specific recommendations for optimized tube voltage parameters for DR chest imaging. This study, validated with images containing realistic anatomical noise, has investigated and recommended an optimal tube voltage range.

Introduction

Computer simulation of medical X-ray images are now used widely, largely for training and optimization purposes.^1–12 For the latter, it is essential that simulated images contain realistic projected anatomy (often described as anatomical noise) and system noise to ensure the optimization task is as realistic as possible.^13–22 A computer algorithm capable of simulating computed radiography (CR) chest, abdomen and pelvis images with digitally reconstructed radiographs (DRRs) created by ray tracing virtual x-rays through real patient CT data sets has been reported by our group²³ and has recently been improved and validated for simulation of digital radiography (DR) chest, abdomen and pelvis images.²⁴

Under the Ionising Radiation (Medical Exposure) Regulations 2017 (IR(ME)R17),²⁵ all medical radiation doses to patients must be kept as low as reasonably practicable (i.e. optimised), consistent with the intended purpose. The approach we have used with simulated DRR images for optimization has led to the derivation of optimized exposure parameters for chest CR imaging^26–28 and more recently abdomen, pelvis and spine CR imaging.²⁹ Changes to exposure technique (such as tube voltage) have since been applied in our radiology department as a direct result of this work. However, with the transition from CR to DR we deemed it necessary to derive optimized tube voltage (kVp) for chest imaging using simulated images of these DR flat panel systems.

It is widely acknowledged that chest radiography is one of the most frequently performed diagnostic X-ray examinations in the UK due to its excellent clinical value, such as the diagnosis of pulmonary problems and other diseases. The Health Protection Agency (now Public Health England, PHE) report HPA-CRCE-012³⁰ presented the frequency and collective dose for medical and dental examinations in the UK in 2008, and demonstrated that dental X-ray examinations was the most numerous at 26% of all X-ray examinations, but chest X-rays contributed the second largest at approximately 19.6%; optimization of DR chest imaging is therefore an essential requirement in modern medical imaging. Despite this, there are few studies that specifically investigate optimized tube voltage settings for adult chest radiography with digital imaging (DR) systems using clinically relevant physical or computerized phantoms. Doyle et al³¹ and Dobbins et al³² used physical phantoms to optimize a DR system and results were mixed; the former showed that lower tube voltages were superior for a matched effective dose in contrary to the latter. Low tube voltage was also shown to be optimum in a more recent study by Compagnone et al.³³ However, the phantoms used in these studies lacked anatomical detail and only measured physical metrics such as signal-to-noise ratio. Furthermore, there are few studies that use clinical observers to score and grade images containing the essential anatomical noise. Various Monte Carlo investigations have been carried out, largely for CR chest imaging, indicating low tube voltages are optimal.^34–37 However, the computerized "Zubal" voxel phantom used in most of the Monte Carlo studies are identified with only one of four tissue types, namely soft tissue, bone, bone marrow, lung tissue and air, limiting the contribution of anatomical noise.³⁸ The resolution of this voxel phantom is also very coarse (voxels are approx. 4 mm long x 3 mm wide x 3 mm thick) limiting the spatial resolution properties of the resulting images. Furthermore, there are currently no recommended optimum X-ray tube voltage parameters for chest imaging with DR systems; the Council of European Communities (CEC) Quality Criteria³⁹ recommend 125 kVp but this is for outdated film-screen technology.

This paper describes the use of computer simulated chest images that contain clinically realistic projected anatomy (i.e. anatomical noise), and superior spatial resolution compared with the Zubal phantom, produced across a range of diagnostically relevant tube voltages to derive an optimized tube voltage range for caesium-iodide (CsI)-based DR chest imaging (without using an antiscatter grid, as per local exposure protocol). The evaluation of simulated images was carried out by experienced image evaluators (radiologists and reporting radiographers), and so this work presents the results of a virtual clinical trial. This study did not use images simulated with an anti scatter grid due to time constraints of the image evaluators; optimization of our current exposure technique was deemed an appropriate place to start the study. Furthermore, we did not use a specific physical phantom to carry out this study as it would give a sample size of only one, which is clearly not good enough, irrespective of how many image evaluators participate. It is acknowledged that clinical trials with real patients are likely to provide increased statistical significance than the virtual trial described here, but use of simulated images is much more cost effective, requires less time, rules out the need for research ethics committee approval and most importantly precludes the need for repeat patient exposures.

Methods and materials

Synopsis of the computer simulator

As described in the introduction, the X-radiograph simulator is based on a DRR algorithm, i.e. a computer simulation of a conventional two-dimensional X-ray image created from three-dimensional CT data. Although fully described and validated in previous studies,^23,24 a brief synopsis of the algorithm is outlined below (all software coding in this paper was carried out using Matlab v. 2012b, Mathworks, Natick, MA):

1. The computerized phantom is produced from real average (70 ± 10 kg—identified with input from the expert scanning radiographer) patient CT data sets. The voxel resolution of the phantom is 0.34 × 0.39 × 0.34 mm (width x height x depth). This is superior to the resolution of the Zubal phantom used in Monte Carlo studies.

2. Hounsfield units (HUs) of the phantom is converted into linear attenuation coefficient (LAC) using formulae derived from the Gammex RMI (Gammex-RMI, Broadway Business Centre, Nottingham, UK) tissue equivalent phantom (model no. 467). This is a solid water cylinder that contains 17 inserts, the attenuation properties of which mimic the range of attenuations of the various tissues found in vivo. This demonstrates an improvement over the computerized phantom used for Monte Carlo studies, which contains only four tissue types.

3. Virtual X-ray spectra are generated using the Spectrum Processor software⁴⁰ which is based on Report 78 published by the Institute of Physics and Engineering in Medicine.⁴¹

4. X-ray pencil beams are projected through the CT data set using a ray casting method of DRR production. X-rays are attenuated as they move through the CT data in an exponential manner. The intensity of photons in 0.5 keV energy bins emerging from the virtual phantom is calculated.

5. A DRR of pixel values in terms of absorbed energy (keV/mm²) in the DR system’s converter layer (CsI phosphor) is then created. This is the raw DRR.

6. Frequency-dependent noise, correct for detector incident dose and beam quality, is added to the raw DRR. To do this, we acquire a uniform noise image at approx. 5 µGy on a real DR system—this uniform noise image contains all of the relevant noise sources of a real DR system. The uniform noise image is then corrected to account for dose and beam quality using measured relationships between noise, beam quality and dose on a real DR system.²⁴

7. Scatter measured experimentally using an anthropomorphic phantom without an antiscatter grid on a real DR system are added to the raw DRR. We made measurements of scatter and scatter fractions (ration of scatter to total radiation) across the whole chest radiograph using an array of 224 lead beam stops, each of 6 mm in thickness and diameter, 25 mm apart from one another, suspended on a 1 mm thick poly-methyl methacrylate (PMMA) sheet. Measurements were made across the diagnostic tube voltage range using the chest portion of the Alderson RANDO phantom, with the lead stop array positioned in front of the phantom. This phantom consists of a natural human skeleton embedded in a synthetic isocyanate rubber with lung substitute and air cavities, simulating the average adult, approximately 70 kg. As primary X-radiation is absorbed by the lead stops, their shadows in a radiograph provide an estimate of scatter. An image of the beam stops with and without the phantom was acquired and the mean pixel value in the shadow of (scatter) and adjacent to (primary plus scatter) each lead stop was calculated. Scatter and SFs were subsequently measured at the position of each lead stop, and a two-dimensional interpolation program written in Matlab (bicubic interpolation) was used to calculate the values of scatter and SFs across the entire image. These measurements were used to add scatter over the raw DRRs.²³ Although only average-sized adults have been identified in this study (of which RANDO is modelled), anatomy differs slightly patient to patient and as such scatter and SF masks do not always fit accurately over raw patient DRRs. To overcome this, scatter and SF masks intended for patient DRRs were registered to them using the Matlab image registration function.

8. DRR pixel values are converted to DR system pixel values using the measured system transfer properties of the real imaging system.

9. The resulting DRR images were validated quantitatively with a chest, abdomen and spine phantom²⁴ and real patient²³ images. Signal-to-noise ratio values of the DRR images in the lung, spine and diaphragm regions agreed to within 10% across the diagnostic energy range when compared with the real images. Histograms were similar in shape and the dynamic range of the DRR images (minimum and maximum pixel values) were within two standard deviations of the mean of the corresponding values in the real images. Qualitative validation was carried out by expert image evaluators and they all agreed that the DRRs adequately simulated real images, and that they were acceptable to use for optimization studies. As well as normal chest anatomy, the model includes artificially added lung nodules. Lung nodules were chosen as they are indicative of common malignant disease such as cancer, and non-malignant diseases such as tuberculosis, pneumonia and sarcoidosis. Expert image evaluators were also asked to score out of 10 (1 = definitely not, 10 = definitely) if the position and appearance of the nodules were realistic. The mean (±1 standard deviation) score was 7.8 ± 1.2, thus validating the appearance and positioning of the nodules.

Simulation of a real DR imaging system

The DRR simulation algorithm is capable of simulating DR chest images of average adults at various tube voltages, detector air kerma values and with or without scatter rejection (using an antiscatter grid). For this study, the DRR algorithm was configured to simulate the Agfa DX-D400 (Agfa, Peissenberg, Germany) DR system with imaging panels 35 × 43 cm in size (pixel pitch 0.14 mm) utilizing a CsI converter layer. Furthermore, the associated ceiling suspended Toshiba X-ray tube with total filtration equivalent to 3.2 mm of aluminium and an anode target angle of 12° was incorporated into the algorithm.

Chest images produced by the DRR algorithm reconstructed with tube voltages 60 and 120 kVp (matched effective dose) are shown in Figure 1(a), (b) respectively. Figure 2(a) demonstrates a magnified low dose chest image as opposed to Figure 2(b) which is a "clinically correct" dose image. Both images were simulated with a tube voltage of 80 kVp.

Figure 1. .

(a) Simulated low tube voltage (60 kVp) chest image and (b) high tube voltage (120 kVp) chest (matched effective dose). Figure 1(a) shows increased contrast between bone and soft tissue background.

Figure 2. .

(a) Simulated magnified low dose chest image and (b) normal dose chest image (b). Figure 2(a) shows increased noise in the dense regions. Both images were simulated with a tube voltage of 80 kVp.

As illustrated in Figure 1(a), (b) contrast decreases as the tube voltage increases (with matched effecive dose), especially in the lung and ribs. This is expected due to the decrease in differences between the linear attenuation coefficients of different tissues with an increase in voltage. Figure 2(a), (b) show a higher level of noise in the heart and spine regions for the lower dose image Figure 2(a) compared to the higher dose image Figure 2(b).

Simulated image reconstruction and preparation

CT data sets from 20 average-sized patients (70 ± 10 kg—identified with input from the expert scanning radiographer) were used to create simulated DR images (note—permission from our local R&D committee was obtained prior to using retrospective CT data. Health Research Authority or ethics committee approval was not necessary). The images were simulated without any scatter rejection at seven different tube voltages –60 to 120 kVp in steps of 10 kVp. Each simulated image was generated with a matched effective dose of 0.01 (±1%) mSv by using an appropriate tube-current time product (mAs) required at each tube voltage to provide this effective dose. This value of effective dose was used as our institutuion’s local diagnostic reference level is 0.07 Gycm² for chest imaging. This converts to an effective dose of 0.01 mSv if one uses the conversion coefficient recommended in the Health Protection Agency’s (now Public Health England) report HPA-CRCE-028. The technique with no scatter rejection described here is consistent with the current chest imaging protocol used in our radiology department. Images of a given patient were transferred to a folder and given the name Patient_N, were N was the sequential patient number. For example, Patient 1 had seven images (images 60–120 kVp in steps of 10 kVp) transferred to its folder called Patient_1. This was repeated for all 20 patients. All patient folders were placed in a location accessible to image evaluators. It should be noted that we used a sample size of 20 patients as this was a sensible balance between the time available to the image evaluators and the required statistical accuracy.

Evaluation of clinical image quality using a blind observer study

Five experienced image evaluators (three Radiologists and two reporting Radiographers) graded all images on reporting Picture Archiving and Communications Systems (PACS) workstations with a dual monitor configuration (Barco Ltd, Brussels, Belgium). The monitors were calibrated to national standards⁴² and were kept in dedicated viewing rooms with lighting levels maintained at an acceptable level. On a third (general purpose) monitor, in-house software ("IQ Scoring") was used by the evaluators to open each patient image folder in sequential order. The software automatically opens three windows, one containing the reference image for grading, the second containing a test image and the third a scoring panel. The evaluators were instructed to place the reference window on the right hand PACS monitor, the test window on the left hand PACS monitor and the scoring panel on the general monitor.

The reference image was designated as that contained within the patient folder (see "Simulated image reconstruction and preparation") generated with a tube voltage of 90 kVp because this is in the centre of the tube voltage range; at the outset of this study our radiology department did not have any experience with DR imaging and therefore did not have a standard tube voltage for chest radiography, so a reference value based on local protocol could not be ascertained. None of the evaluators had any knowledge of what tube voltage the test or reference images represented. IQ scoring automatically displayed a random (in terms of tube voltage) test image in the test window and evaluators were asked to grade this image against the reference, using the grading panel. Evaluators were allowed to change the window and level settings of each image prior to grading to optimize the appearance of each, as per clinical practice. The grading panel was configured to allow each image evaluator to grade each test image on a continuous scale, along which any point may be selected by the evaluator, rather than using an ordinal scale. This approach has recently been recommended by Keeble et al⁴³ to circumvent some of the problems associated with the analysis of ordinal data for medical image optimization.

Slider bars were used on the grading panel by the evaluators for scoring each criteria. Placing the bar to the left of centre indicated inferior image quality of the test image with respect to the reference for the specific criteria being scored; the further bar was placed to the left denoted increasingly poorer quality than placed close to the centre. Conversely, placing the bar to the right of centre indicated superior image quality of the test image with respect to the reference. The exact positioning of each bar was dependent on each evaluator’s preference and therefore subjective. To account for this, we used a linear mixed algorithm to model the scores with evaluator as a random effect (see next section). The numerical value of the score was kept blind to each evaluator but the possible range spanned −3 to +3, e.g. if an evaluator placed the bar all the way to the left this would record a score of −3 for that criteria. When this process was finished for a specific test image, the next random test image was automatically displayed and the scoring process was repeated until all test images were graded for all 20 patients. Furthermore, each evaluator was forced to compare the reference image with itself to check for any observer bias and any potential non-uniformities between the two monitors.

Image grading criteria were based on the Council of European Communities Quality Criteria,³⁹ revised to reflect modern diagnostic requirements and experiences of our group (Table 1).

Table 1. .

The image criteria used for grading simulated images

Image Criteria
Overall quality of image compared to reference
Quality of lung region
Quality of hilar/mediastinum
Quality of spine
Quality of ribs
Quality of diaphragm/retrodiaphragmatic
Quality of lung nodule

Analysis of evaluator scores using a linear mixed effects algorithm

As recommended by Keeble et al⁴³ we modelled the relationship between average evaluator score for each criteria and tube voltage with a linear mixed effects (LMEs) algorithm using the R statistical software package⁴⁴ (R Foundation for Statistical Computing, Vienna, Austria) and lme4 (Bates et al⁴⁵). The linear mixed effects algorithm adjusts the average evaluator score for each criteria by correcting for random effects, such as within-subject and between-subject variability. It also allows the use of all data available and account for the correlations between data coming from the evaluators and patients, even when there is a relatively low sample size for this structured data (i.e. there are numerous covariates to be fitted). The following full model was formulated in R:

$${\displaystyle \mathrm{S}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}\sim \mathrm{T}\mathrm{u}\mathrm{b}\mathrm{e}\mathrm{V}\mathrm{o}\mathrm{l}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e}\ast \mathrm{I}\mathrm{m}\mathrm{a}\mathrm{g}\mathrm{e}\mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{a}+(1|\mathrm{P}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{e}\mathrm{n}\mathrm{t})+(1|\mathrm{E}\mathrm{v}\mathrm{a}\mathrm{l}\mathrm{u}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r})}$$ (1)

where "Score" is the image quality score given by the observer (in the range −3 to +3), "Tube Voltage" is the tube voltage being evaluated, "Image Criteria" is the structure that the score is being given for from Table 1 e.g. lung, "Evaluator" is the person scoring the image and "Patient" is the specific patient being evaluated. In this model, "Tube Voltage" and "Image Criteria" are the fixed effects, as the hypothesis is that the tube voltage settings will determine the score given by the observer but will depend on the region of the image being evaluated. In this model, it is expected that there will be an interaction between these two fixed effects (as denoted by the *), as the evaluator may give different scores for a given tube voltage setting depending on which region of the image they are looking at, e.g. 120 kVp may be good for spine detail, but poor in the lungs. The random effects are the "Patient" and "Evaluator" as they may influence the score (and hence the model), but it is not possible to fully control for this as the study only samples 20 patients and 5 evaluators out of a much larger population; these are random effects as some patients may give particularly good scores compared with others, whilst the sample of evaluators may give scores on a slightly different scale to each other (some may score "slightly better" as scores of around 0.5 on the sliding scales, whilst others may consider this to be closer to a score of 1.0, and thus have different "baselines"). The aim of the study is to find the best option overall for all image evaluators. It should be noted that these underlying statistical models are complex but in its simplest sense we have used a random-intercepts linear mixed model which accounts for "baseline" differences between evaluator, but assumes that whatever the effect of tube voltage is, it will be the same for all evaluators (i.e. the linear model assumes inter evaluator regression lines are parallel). A random-intercepts model is where the outcome variable Y (score) is a function of predictor X (tube voltage), with a random intercept (C) for evaluator, using the underlying statistical model Yi = BXi+ Cj. It would have been preferential to have used a random-slope model where evaluators are not only allowed to have differing intercepts, but where they are allowed to have differing slopes (i.e. whatever the effect of tube voltage is, it will not be perceived the same for all evaluators), but there were not enough patient data sets to do this.

To determine if there is a statistically significant relationship between score and both tube voltage and image criteria, the following null models were created:

$${\displaystyle \mathrm{S}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}\sim \mathrm{I}\mathrm{m}\mathrm{a}\mathrm{g}\mathrm{e}\mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{a}+(1|\mathrm{P}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{e}\mathrm{n}\mathrm{t})+(1|\mathrm{E}\mathrm{v}\mathrm{a}\mathrm{l}\mathrm{u}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r})}$$ (2)

$${\displaystyle \mathrm{S}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}\sim \mathrm{T}\mathrm{u}\mathrm{b}\mathrm{e}\mathrm{V}\mathrm{o}\mathrm{l}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e}+(1|\mathrm{P}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{e}\mathrm{n}\mathrm{t})+(1|\mathrm{E}\mathrm{v}\mathrm{a}\mathrm{l}\mathrm{u}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r})}$$ (3)

$${\displaystyle \mathrm{S}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}\sim 1+(1|\mathrm{P}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{e}\mathrm{n}\mathrm{t})+(1|\mathrm{E}\mathrm{v}\mathrm{a}\mathrm{l}\mathrm{u}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r})}$$ (4)

A likelihood ratio test using the ANOVA statistical test was carried out in R to compare the likelihood of the full model with each null model in turn, completely independent to one another. A p-value was returned for each test indicating whether removing tube voltage, image criteria, or both in combination from the model had a statistically significant effect on the evaluator score. For the purpose of this work, a p-value of less than 0.05 was considered statistically significant.

Results

It was found that:

• Tube voltage effected LME modelled evaluator score in a statistically significant manner (χ²(69)=916, p < 2.2×10⁻¹⁶).

• Image criteria effected LME modelled evaluator score in a statistically significant manner (χ²(69)=1021, p < 2.2×10⁻¹⁶).

• Tube voltage and image criteria in combination effected LME modelled evaluator score in a statistically significant manner (χ²(75)=1076, p < 2.2×10⁻¹⁶).

The results of the optimization study for 20 average patients are shown graphically in Figure 3.

Figure 3. .

Average image quality scores modelled using a linear mixed effects algorithm across all image criteria. Error bars indicate the standard error of the mean.

It is clear from Figure 3 that the image quality score is dependent on tube voltage and the scoring criteria (Table 1). The overall quality criteria demonstrates poor performance at low and high tube voltages with 60 and 120 kVp scoring −0.60 and −0.42 respectively. However, the scores improve in the mid-voltage range, with 80, 90 and 100 kVp scoring –0.05, –0.01 and −0.08. This demonstrates overall image quality improves with mid-value tube voltages and if one takes the error bars into account, the suggested optimal range being 80–100 kVp. The dense region image criteria (hilar, diaphragm and spine) show a very similar trend, with scores ranging from −0.57 at low tube voltage (60 kVp diaphragm) to −0.27 at high tube voltage (120 kVp hilar) but improving considerably in the mid-voltage range, again indicating 80–100 kVp to be optimal. This may be due to the efficiency of CsI DR phosphors across the diagnostic energy range.

The scores for lung and lung nodule (i.e. less dense criteria), show little variation with tube voltage with lung ranging from +0.03 at 60 kVp to −0.07 at 120 kVp and lung nodule ranging from −0.01 to +0.02. This relatively flat response may be because air is largely independent of photoelectric absorption across the tube voltage range.

Furthermore, the scores for rib show superiority at low tube voltage with a gradual decrease as tube voltage increases, with scores ranging from +0.08 at 60 kVp to −0.08 at 120 kVp. Given rib has a relatively high density and atomic number compared to its background (typically air in lung); the photoelectric effect is expected to have a relatively strong influence with tube voltage.

Finally, it is clear from Figure 3 that the scores for the 90 kVp images across all image criteria are very close to zero (average less than 0.01) indicating each evaluator gave the 90 kVp test image a score at or very close to zero, as would be expected since they were comparing identical images.

Discussion

It is essential that the medical physicist has an appropriate knowledge of optimized exposure technique to ensure they can advise radiographic operators accordingly. The results of this study will therefore be useful to the medical physicist charged with the task of advising radiology departments on how to optimize tube voltage for chest imaging, given that the results here have shown that tube voltage and image criteria affects evaluator score in a statistically significant manner. This has been demonstrated through expert grading of computer simulated images.

Overall image quality, hilar, spine and diaphragm criteria demonstrated poor performance at low and high tube voltages, peaking in the mid-voltage range (80–100 kVp). This may be explained by the efficiency of CsI DR phosphors across the diagnostic energy range given that CsI has a peak efficiency at 70–90 kVp with a drop-off at lower and higher tube voltages, as shown in Figure 4. The results of this virtual clinical trial for these dense regions follows this physical response of CsI phosphor to X-ray beams attenuated with 18.5 cm PMMA, i.e. a relatively dense material (18.5 cm PMMA has been shown by Martin et al⁴⁶ to provide equivalent attenuation to 20 cm of water at tube voltages that span the diagnostic range (60–120 kVp), which in turn approximates a trunk AP view).

Figure 4. .

Efficiency (energy absorbed per number of incident photons) of CsI phosphor as a function of tube voltage for X-ray beams attenuated with 18.5 cm PMMA. CsI, caesium-iodide; PMMA, poly-methyl methacrylate.

Lung and lung nodule demonstrated a flat response with tube voltage. For both criteria this is expected; lung is largely made up of air which has a very low density and atomic number and so is relatively independent of photoelectric absorption, and lung nodules are a mass of tissue contained within this air and so are easy to see irrespective of the tube voltage used; the appearance of both should therefore be "flat" across the tube voltage range. Rib showed superior performance at low tube voltage due to strong photoelectric dependence with low energies (i.e. low tube voltage) as rib has a relatively high density and atomic number compared to its background (air). Furthermore, these results show that for matched effective dose, image quality is dependent not just on tube voltage but on image criteria as well, i.e. region of the chest, so a sensible compromise would be to use 80–100 kVp for DR chest radiography of average adults, given image quality scores for overall and dense region criteria peaks in this range and the less dense criteria are not affected by tube voltage. Rib image quality is of minimal interest in chest imaging so using low tube voltages which gave superior rib quality is rarely, if ever, justified.

Using a tube voltage in this range runs contrary to recent advice given to our radiology department by applications specialists of some X-ray installation companies, as well as the established but out of date CEC guidance. Typically, their advice is to use a high tube voltage, such as 125 kVp, but the findings of this study have demonstrated that this is not optimal. Based on results of this study, in 2017 our radiology department adopted a standard operating protocol for average chest exposures of 80 kVp, 2 mAs. In the 2 years this exposure protocol has been in place, there have been no complaints about image quality. Although this is a relatively mid-range tube voltage, large-scale patient dose audits, following the method by Wood et al,⁴⁷ have consistently shown median dose–area product for standard-sized patients are 0.07 Gycm² (this has been set as our LDRL). This is also lower than the national DRL of 0.10 Gycm² for PA chest exposures. Furthermore, high tube voltage techniques such as that recommended by applications specialists are still used in modern radiological imaging departments, primarily to: (1) subdue the appearance of ribs and (2) optimize skin entrance dose, but the natural conclusion of this work is that chest DR imaging can be performed at lower, mid-range tube voltages. Nevertheless, one must not forget that to maintain an appropriate detector air kerma at the mid-range tube voltages recommended here, tube current-time (i.e. mAs) must be increased, thus increasing patient skin dose. However, the maximum air kermas values at the skin encountered with chest radiography (typical values < 0.1 mGy at 80 kVp with our current clinical protocol) will always by sub mGy, so it should be effective dose (and therefore risk), that is the limiting factor in determining optimum tube voltage for chest radiography with digital imaging systems. Furthermore, years ago when film-screen radiography was all that was available, high tube voltages such as 125 kVp were required to overcome the limitations of the small dynamic range of film, but this is not a concern with digital detectors. Due to the results presented in this manuscript, the use of mid-range tube voltages for DR chest radiography is justified.

Limitations

There are numerous limitations associated with this work. Firstly, there are relatively large errors in the LME evaluator scores due to the relatively limited simulated patient information set (n = 20). Secondly, only "average" patients have been investigated so the optimal tube voltage range recommended here can only be used for patients of this size. However, the IR(ME)R regulations mandate the establishment of written exposure protocols for every type of standard radiological practice; the tube voltage range derived in this study can be used for this purpose. The expert operator is subsequently expected to modify exposure parameters depending on patient size. Thirdly, the acquired level of change one may expect to see in clinical image quality with different tube voltages, for a given change in computer simulated image quality, may not always be observed in reality. The results reported here must therefore be interpreted in the context of this limitation. Fourthly, the use of an antiscatter grid has not been investigated in this paper. It is acknowledged that many of radiology departments use grids routinely for chest imaging, so one must be careful in transferring the results derived here to DR chest imaging with grids. However, at the very least, those departments that do use grids can use the tube voltage range recommended here as a starting point for optimization, and adjust exposure factors accordingly. A similar exercise with images simulated with an antiscatter grid would be worthy of future study. Fifth, due to time constraints and workloads of the image evaluators, repeated measurements with the same simulated images to assess within-subject variability was not carried out. Finally, it should be remembered that the findings in this paper are specific to Agfa CsI-based DR receptor technology. However, given that most DR phosphors are based on CsI phosphors, it is likely that the conclusions of this work are transferable to other manufacturers, but further studies would be worthwhile, especially with Gadolinium Oxy-Sulphide detectors.

Conclusions

Contrary to traditional high tube voltage convention, it has been demonstrated with simulated images containing realistic clinical content (normal and abnormal), that a tube voltage mid-range of 80–100 kVp is optimal for DR imaging of the chest. There is no justification to increase tube voltage to levels recommended by CEC guidance or applications specialists. This will allow the medical physicist, in conjunction with clinical imaging experts, to optimize chest imaging by utilizing appropriate tube voltage range, if this is deemed clinically appropriate at their institution.