Report on the study and optimisation of chest and lumbar spine X-ray imaging

Transcription

1 Institutionen för medicin och vård Avdelningen för radiofysik Hälsouniversitetet Report on the study and optimisation of chest and lumbar spine X-ray imaging Michael Sandborg, Graham McVey, David R Dance and Gudrun Alm Carlsson Department of Medicine and Care Radio Physics Faculty of Health Sciences

2 Series: Report / Institutionen för radiologi, Universitetet i Linköping; 91 ISRN: LIU-RAD-R-91 Publishing year: 2 The Author(s)

3 Report 91 February 2 ISRN ULI-RAD-R--91--SE Report on the study and optimisation of chest and lumbar spine X-ray imaging M Sandborg 1, G McVey 2, D R Dance 2, and G Alm Carlsson 1 1. Department of Radiation Physics, IMV, Linköping University 2. Joint Department of Physics, The Royal Marsden NHS Trust Full addresses: 1. Department of Radiation Physics, IMV, Faculty of Health Sciences, Linköping University, SE LINKÖPING, Sweden Fax Michael.Sandborg@raf.liu.se Gudrun.Alm.Carlsson@raf.liu.se 2. Joint Department of Physics, The Royal Marsden NHS Trust and Institute of Cancer Research, Fulham Road, London SW3 6JJ, United Kingdom Fax d.dance@icr.ac.uk 1

4 Table of contents 1 Introduction 3 2 Method Monte Carlo program Voxel phantom 2.3 Image quality and patient dose descriptors Contrast, OD Signal-to-noise ratio, SNRI Dynamic range, DR Patient dose 2.4 Trial and reference system Application of the model Effect on image quality and patient dose of varying the imaging system parameters Optimisation Comparison of physical and clinical image quality scores Chest Lumbar spine 3 Results and discussion Effect of technical parameters on image quality descriptors and patient dose Chest PA Effect of tube potential Effect of tube potential ripple Effect of filter Effect of grid design Effect of screen sensitivity Effect of optical density Effect of film H&D curve Lumbar Spine AP Effect of tube potential Effect of filtration Effect of tube potential ripple Effect of anti-scatter grid: N=4 cm -1 with Al or carbon fibre materials Effect of anti-scatter grid: N=7 cm -1 with 36 and 2 micron lead strips Effect of screen sensitivity Effect of film H&D curve and optical density Lumbar Spine lateral Effect of tube potential Effect of filtration Effect of tube potential ripple Effect of anti-scatter grid: N=4 cm -1 with Al or carbon fibre materials Effect of anti-scatter grid: N=7 cm -1 with 36 and 2 micron lead strips Effect of screen sensitivity Effect of optical density 2

5 3.2. Comparison of physical and clinical image quality scores Chest PA Results of simulating of the 16 imaging systems used in the clinical trial Comparison with the Image Criteria Score and Visual Grading Analysis Score Lumbar spine AP Lumbar spine Lateral Optimisation Chest PA Scatter-rejection technique Screen speed Optical density, OD Type of Film Image Quality Factor, IQF Lumbar spine AP Scatter-rejection technique Screen speed Optical density, OD Type of Film Image Quality Factor, IQF Change of reference system Lumbar spine lateral Scatter-rejection technique Screen speed Optical density, OD Type of Film Image Quality Factor, IQF Effect on optimisation of choice of reference system 4 Summary and Conclusions 59 Acknowledgement 6 References 6 Figures 62 3

6 1 Introduction The optimisation of radiological equipment and equipment parameters is a key stage in the optimisation of image quality and patient dose in diagnostic radiology. It is essential, however, to underpin such optimisation with theoretical modelling which can provide both the direct quantification of the effect on image quality and dose of changes in system parameters and the opportunity for optimisation of equipment selection and use. Our principal contribution to the joint CEC-project 'Predictivity and Optimisation in Diagnostic Radiology' is in modelling of planar chest and lumbar spine radiographic examinations. The results of this work for the chest PA, lumbar spine AP and lumbar spine lateral examinations are presented in this report. Prior to this, several development stages have been completed which include the calibration and validation of our methods by measurements in the clinical environment on patients and patient images. These important aspects are not dealt with in detail here, but are described in separate reports (Sandborg et al. 1997, McVey et al. 1998, 1999, Sandborg et al. 1999). This report focuses on three aspects from our results of using our Monte Carlo model of the patient and imaging equipment: (1) Study of the effects on image quality and patient dose when the imaging parameters are varied; (2) Establishment of imaging parameters and systems that result in as least as good image quality as systems with good performance singled out from results of clinical trials (optimisations); (3) Comparison of the results from the model with the results from clinical trials performed by partners in the joint CEC-project. An objective of the report is to present our results at a level of detail not usually possible in the refereed scientific literature. The report should therefore not be read all at once, but preferably used as a reference library or documentation of all our efforts. There are many interesting results and findings from this collaborative work and these will be submitted for publication to the appropriate journals. 2. Method 2.1. Monte Carlo program The Monte Carlo program (MCP) used in this work is based on programs developed in the past by our group (Sandborg et al. 1994, Dance et al. 1997). Modifications have been made which are outlined below. The main difference is that the patient is now modelled by a 3-dimensional, segmented male anatomy (voxel phantom) originally developed by Zubal et al. (1994). Using this anthropomorphic voxel phantom, we are able to make more realistic estimates of the variation in the image of contrast and noise as well as calculating the effective dose in the patient. Using the collision density estimator (Persliden and Alm Carlsson 1986) we can estimate the energy imparted per unit area to the image receptor at any point in the image plane behind the voxel phantom. The voxel phantom enables us to more realistically include the variations of scattered and primary photons in the image plane as well as to assess the effects of the limited dynamic range of the image receptor. In addition, the MCP includes the x-ray spectra (anode material and angle, peak tube potential and ripple, and added filtration), anti-scatter grid (strip frequency, lead strip width, grid ratio and material in interspaces and covers) or air gap, couch-top or chest stand and the image receptor (cassette front, screen-film system, and H&D curve). The statistical errors in the calculated quantities was estimated ro 3% or better for the chest PA and lumbar spine PA projections and 4% or better for the chest LAT and lumbar spine LAT projections. Figure 2.1 shows the imaging geometry used in the model to simulate a chest PA examination. 4

7 Voxel-phantom Grid Cassette and screenfilm system Filter Collimators Focalspot Cheststand Figure 2.1. The geometry of the Monte Carlo model including the components of the imaging system. In this example, a chest PA examination with a grid is shown. A detailed description of the calibration and validation of the MCP is given in two separate reports (McVey et al. 1998, 1999) and is not discussed further here. The validation comprises three steps: (1) calibration of the sensitivity of the image receptor; (2) validation of the model of the imaging system components; (3) demonstrating that the voxel phantom is representative of patients in the clinic 2.2. Voxel phantom Anatomical details were included in the voxel phantom. These were used for image quality analysis and were selected on the basis of quality criteria published by the European Commission (CEC 1996). Radiologists in Linköping and London were consulted to help with this selection. The anatomical details included are listed in tables 2.2a-c. For the chest examination, blood vessels of different size and location (central lung, retro-cardiac area, costo-phrenic angle area) were used as well as calcifications in the lung apices. For the lumbar spine, the transverse (AP) and spinous (lateral) processes were used as well as trabecular structures in the L1, L3 and L5 vertebra. A more detailed description of the selection and measurement of anatomical details is given in Sandborg et al. (1997). It should be noted that all details listed in table 2.2 are part of the normal anatomy. Table 2.2a. The anatomical details included in the voxel phantom in a chest PA examination. Chest PA 1.8 mm blood vessel in Central Right Lung (CRL) 1.8 mm blood vessel in Costo-Phrenic Angle area (CPA) 3. mm blood vessel in Retro-Cardiac Area (RCA).5 mm calcification in the Left Lung Apex (LLA).5 mm calcification in the Right Lung Apex (RLA) 5

8 Table 2.2b. The anatomical details included in the voxel phantom in a lumbar spine AP examination. Lumbar Spine AP 2., 3.5, 5. mm bone simulating L1, L3 and L5 transverse process (L1T, L3T, L5T) 1. mm trabecular structure (bone marrow) in L1, L3 and L5 vertebra (L1D, L3D, L5D) Table 2.2c. The anatomical details included in the voxel phantom in a lumbar spine lateral examination. Lumbar Spine Lateral 5., 5.5, 6. mm bone simulating L1, L3 and L5 spinous process (L1S, L3S, L5S) 1. mm trabecular structure (bone marrow) in L1 and L5 vertebra (L1F, L5F, L5B) 2.3 Image quality and patient dose descriptors Two measures of image quality were calculated for the anatomical details in table 2.2: the contrast (optical density difference, OD) and the ideal observer signal-to-noise ratio (SNRI) Contrast, OD In calculating the OD, the effects of film gradient and imaging system unsharpness were considered. The film gradient (γ) was obtained from measurements of the H&D curve using the ISO-standard (ISO 1993). The effect on OD of unsharpness was calculated in the following way. The Fourier transform of the detail was calculated and multiplied by the imaging system total MTF(?), including both receptor (screen), geometric (focal spot size and magnification) and motion unsharpness. Then the inverse Fourier transform of the product was then calculated. Finally, the degradation in contrast was expressed as the reduction in peak amplitude in the centre of the MTF-modified profile of the detail compared to the original, unmodified amplitude. For small details (calcifications and trabecular details), this degradation is significant whereas for large details the degradation is small. A more detailed description of the implementation of unsharpness is given in a separate report (Sandborg et al. 1999) Signal-to-noise ratio, SNRI The SNRI was calculated in two steps. First the SNRMC due to quantum noise only was calculated using the MCP. This is based on the energy imparted per unit area to image elements beside and behind the detail using the methodology in Sandborg et al. (1994) and Tapiovaara and Sandborg (1995). The SNRMC overestimates the SNRI since it assumes a sharp system and does not include the Swank factor (Swank 1973) IOPD=.8 due to the light transport to the film. These effects and also the effect of other noise sources (film noise) were included by applying multiplicative correction factors to SNR 2 MC. These factors were dependent on the optical density. The methods of Nishikawa and Yaffe (199) were used in deriving the correction factors. A complete description is given in Sandborg et al. (1999) Dynamic range, DR In addition to OD and SNRI, the dynamic range (DR) of the image was also studied. Dynamic range is important: even though the object contrast may be large, the contrast on the film may be low due to low film contrast (gradient) in some parts of the film characteristic curve. The dynamic range was here defined as the percentage of the pixels where the gradient γ(od) exceeds a pre-set value. This value was set between Patient dose In order to perform optimisation, estimates of the radiation risk need to be made. The radiation risk was here estimated by the effective dose (E) (ICRP 1991) or entrance air kerma (EAK) without backscatter. All organ or tissue doses used in the derivation of effective dose were calculated by summing the 6

9 energy imparted to all voxels in that organ or tissue and dividing by the appropriate mass. For further information, see McVey et al. (1999). 2.4 Trials and reference system A clinical trial (Almén et al. 1998) was performed by our research partners (Malmö and Göteborg University, Sweden and GSF Munich, Germany) as a part of our joint CEC research project. The patient images were evaluated by a group of expert radiologists. The evaluations made by our partners were based on Image Criteria Score (ICS) or Visual Grading Analysis Score (VGAS). In the first evaluation, the radiologists were asked if the EU Quality Criteria (CEC 1996) were fulfilled. In the visual grading analysis, the radiologists viewed two images, one being a reference image, and were asked if the visibility of the given structure was clearly inferior, slightly inferior, equal to, slightly better or clearly better than the same structure in the reference image. In analysing their results, our partners calculated the observer and image average of the ICS and VGAS for each imaging system and ranked the systems in descending order according to the score. For the chest PA, sixteen different imaging conditions were formed by varying four imaging parameters in two steps, namely 12 kv/141 kv, Lanex 16/Lanex 32 screens, grid/air gap, and ODmax=1.3/ODmax=1.8. In the lumbar spine examination, four different systems were formed using two types of screen and two tube potentials: Lanex Regular (4)/Lanex Fast (6) screens, and 72 kv/9 kv. In our optimisation studies we made use of the imaging systems yielding the highest ICS and VGAS. We called these systems the reference systems. They are listed in table 2.4 for the three examinations studied. Table 2.4. The reference imaging systems selected from the results of the clinical trial (Almén et al. 1998). ODmax is the maximum optical density in the image. ODmed is the median optical density in the image. Examination and view Reference system configuration Chest PA 141 kv, Lanex 16, Air gap and ODmax=1.8 Lumbar spine AP 72 kv, Lanex Regular (Grid, ODmed=1.36) Lumbar spine lateral 77 kv, Lanex Fast (Grid, ODmed=1.36) 2.5 Application of the model The Monte Carlo model was applied for three tasks: (1) Exploration of the effects of varying the imaging parameters on the SNRI, OD, DR and patient dose. (2) Development of an optimisation strategy to explore how much the patient dose could be reduced while maintaining the image quality at (or close to) the level of the reference system. (3) Correlation of the results obtained from the model calculations with the scores from the European clinical trial (as evaluated by ICS and VGAS). This required a weighting of the calculated values of OD, SNRI and DR to form a Physical Image Quality Score (PIQS). 7

10 2.5.1 Effect on image quality and patient dose of varying the imaging system parameters The following system parameters were varied to explore their effect on image quality and patient dose: tube potential tube potential ripple filtration (Al, Cu, Nd, and Gd) anti-scatter device (grid design (ratio, line frequency, interspace and cover material) and air gap) screen-film sensitivity optical density and type of film The values investigated for each parameter depended on the examination and are listed in table 2.5. Table 2.5. Parameter values studied in the Monte Carlo model. System parameter Chest Lumbar spine Tube potential (kv) 9, 1, 11, 13, 141, 15 AP: 6, 7, 72, 8, 9, 11 LAT: 7, 77, 79, 8, 9, 95, 11 Ripple (%), 5, 2, 5, 5, 2, 5 Filtration (mm Al) 2.5, 3.5, 5.7, , 3.5, 4.7 Filtration (mm Cu),.1,.3 - Filtration (mm Nd),.5,.15 - Filtration (mm Gd),.6,.19 - Grid ratio 8, 12, 16 8, 12, 16 Grid strip frequency (mm -1 ) 4, 7 4, 7 Lead strip width (µm) 2, 36 2, 36 Interspace and cover material Al, carbon fibre (CF) Al, carbon fibre (CF) Air gap 28 cm 28 cm Screen sensitivity (speed class) 16, 32 32, 4, 6 Maximum optical density (ODU) Type of film H Il2, Il1, TML, MED, TMG, UGP TML, MED, IL2, UGP H The films are listed in order of increasing film gradient according to Herrmann (1997). The SNRI and OD of the anatomical details (table 2.2), the dynamic range of the image and the incident air kerma and the effective dose (E) were calculated and plotted as a function of the imaging parameters listed in table 2.5. Absolute values are presented (section 3.1) as well as values normalised to the corresponding values obtained with the reference system (table 2.4). A range of film H&D curves (Herrmann 1997) was used in the analysis. Examples of such curves with different gradients are shown in figure

11 Chest films Lumbar Spine films 4, 4, Optical Density (ODU) 3,5 3, 2,5 2, 1,5 1,,5 TML IL1 IL2 MED G Optical Density (ODU) 3,5 3, 2,5 2, 1,5 1,,5 TML IL1 IL2 MED TMG UG UGP, , Log kerma Log Kerma Figure H&D curves for the films used in the trials (Herrmann 1997) Optimisation The optimisation strategy was as follows. Firstly, the requirement on image quality was decided. It was assumed that it was not necessary to exactly match the image quality of the reference system, but that for each detail, a maximum of 1% reduction in the quality measures SNRI and OD would be acceptable. This requirement defines the so-called image quality factor, IQF, which in this case takes the value IQF=.9. The IQF-value was subsequently varied (IQF=.85-1.) in order to explore the dependence of the results on this factor. For a given combination of scatter-rejection technique, screen-film speed, optical density and type of film (H&D), SNRI and OD were calculated for each detail as function of tube potential. The tube potential required to achieve IQF=.9 was determined and the critical measure (SNRI or OD) in deriving this value identified. The lowest of the tube potentials required for any of the details to be reproduced with sufficient quality limits the performance of the system and determines the highest tube potential consistent with the requirement of IQF=.9 for all the details. This tube potential corresponds to the lowest effective dose 1 achievable without compromising image quality. The value of E was noted and normalised to that obtained with the reference system. This procedure was repeated for other combinations of scatter-rejection technique, screen-film speed, optical density and film type and those imaging conditions noted that resulted in a dose saving (relative E<1) Comparison of physical and clinical image quality scores In the above section, imaging conditions fulfilling the requirement of IQF=.9 were identified. These all give acceptable image quality. It is noted, however, that while the SNRI and OD of all details are no less than.9, some will have values exceeding.9, possibly even exceeding those of the reference system (IQF>1.). To take this into account and to allow ranking of the systems regards image quality, a single-valued figure-of-merit, the physical image quality score, PIQS, needs to be defined. The 1 For chest PA, the effective dose does not decrease monotonically with increasing tube potential, but a local minimum is found (see section ). Hence it would in some situations be possible to use an even lower tube potential and achieve a slightly higher contrast and SNR with a small dose-reduction. The local dose-minima are, however, very shallow and this complication has therefore been ignored. 9

12 ultimate aim is to find a physical measure that ranks the systems in the same order as radiologists would do. To achieve this, the results from the clinical trial and the radiologists ranking of the systems were used in the following way Chest The 16 systems used in the trial for chest PA imaging were simulated and values of SNRI and OD of all the details calculated together with values of DR and E. All values were normalised to the highest value of the respective parameter found with any of the systems. Since it is not likely that the value of SNRI or OD of a single detail (or a single value of DR) will correlate well with ICS or VGAS, combinations of details and measures were tested. For each imaging system, the normalised values of DR, OD and SNRI were summed as appropriate. For example, in one case, the values of the SNRI of all the details were included in the sum; in another case, the values of both SNRI and OD were considered. In all cases, the DR was included in the sum. This sum gives the value of the PIQS for the system. The systems were then ranked in descending order of PIQS: that achieving the highest value was given rank-value RPIQS=1 and that with the lowest rank value RPIQS=16. Next, the five imaging systems with the highest and those with the lowest ICS and VGAS were identified and given rank-values 1-5 and 12-16, respectively, in descending order of their ICS and VGAS. For these systems, the sums of their corresponding RPIQS rank-values were calculated. With a perfect match of the rankings using ICS (VGAS) and PIQS, these sums should take the values 15 and 7, respectively. In reality, values higher than 15 and lower than 7 were obtained indicating that perfect agreement was not achieved. The absolute value of the differences ( ) 15 and R PIQS ( R PIQS ) 7 for the five systems with the highest and lowest ICS and VGAS, respectively, were calculated. The preferred combination of measures for the PIQS minimises these differences Lumbar spine In the lumbar spine examination, only four different imaging conditions were evaluated in the trial. The SNRI of the trabecular structures, the OD of the transverse processes in the AP projection or the OD of the spinous processes in the LAT projection and the DR (γ>2.25) were used in defining the PIQS. The rank-value RPIQS was then compared directly to the corresponding rank-values using ICS and VGAS for both the AP and LAT projections. 1

13 3. Results and discussion 3.1 Effect of technical parameters on image quality descriptors and patient dose Chest PA Effect of tube potential The effects of tube potential on the image quality descriptors, the signal-to-noise ratio (SNR) and optical density difference ( OD) of blood vessels and on patient effective dose and entrance air kerma were investigated separately for the air gap and grid techniques. Air gap technique Patient Dose Figures a and c show the variations of the effective dose and the entrance air kerma (without backscatter), respectively, with x-ray tube potential between 9 and 15 kv. Figures b and d show the same variation normalised to the values derived for the reference system operating at 141 kv. An air gap technique was first simulated and a total filtration being 5.7 mm Al. The reference system gives an effective dose (E) of 23.8 µsv and an entrance air kerma (Kair) of 8.7 µgy. At the lowest tube potential (9 kv), the effective dose is the same but the air kerma is 1.37 times higher than at 141 kv. Contrary to the air kerma, the lowest value of the effective dose is not found at the highest tube potential but at a lower setting, approximately 11 kv. However, the minimum is shallow and the difference between the highest and lowest effective dose is less than 1% between 9-15 kv. Image Quality SNR of small details in the lung apices Figure e shows the variations of SNR for a.5 mm calcification in the left (LLA) and right (RLA) lung apex with x-ray tube potential between 9 and 15 kv. The difference between the SNR of the two details is because of the different OD beside the calcification. The range of OD beside the calcifications is and.4-.9 for the LLA and RLA, respectively. Figure f shows the variations normalised to the SNR with the reference system, operating at 141 kv. The reference system gives a SNR of 3.7 (LLA) and 1.9 (RLA). The reduction in SNR is significant in changing the tube potential from 9 kv to 15 kv. At 9 kv the SNR is 1.44 and 1.3 times larger than at 141 kv for the LLA and RLA details, respectively. Contrast of blood vessels in the lung Figure g shows the variations of the OD of blood vessels in the central right lung (CRL), retrocardiac area (RCA) and in the costo-phrenic angle area (CPA) with tube potential between 9 kv and 15 kv. Figure h shows the variation with OD normalised to that of the reference system, operating at 141 kv, for the blood vessel in the CRL. Approximately the same reduction in OD with increasing tube potential is obtained for the blood vessels in the other regions (RCA, CPA). For the reference system, the difference between the optical density behind the blood vessel and the background lung tissue is: -.67 for the CRL, -.68 for the RCA, and.64 for the CPA regions, respectively. There is typically a small reduction in OD with increasing tube potential. The OD in the CRL, RCA, and CPA regions at 9 kv are only factors of 1.11, 1.5, and 1.8 larger, respectively, than at 141 kv. Dynamic range Figure i shows the variation with tube potential of the percentage of the calculated pixels where the film gradient exceeds Figure j shows the same quantity normalised to that of the reference system, operating at 141 kv. 11

14 The percentage of pixels fulfilling this criterion for the reference system is 91%. The percentage decreases with decreasing tube potential and at 9 kv is only 82%. Grid technique Figures k-v contain the same information as figures a-j but the performance of an antiscatter grid has been included for comparison. Similar variations with tube potential are found with the grid but some differences are noted and described below. The grid strip frequency is 4 cm -1, the ratio 12 and the lead strip width 4 µm. The material in the interspaces and covers is aluminium. The performance of grids with fibre materials can be found in section Patient Dose Figures k-n show the variations of E and Kair with tube potential. The effective dose is between times larger than with the reference system (141 kv, air gap). A minimum in effective dose is located at 13 kv. The increase in air kerma compared to that of the reference system is a factor of between , the lower value at the highest tube potential (15 kv). Image Quality SNR of small calcification details in the lung apices Figures o-p show variations of the SNR with tube potential. The SNR using the grid is between 7-12% lower than with the reference system using an air gap. The decrease in SNR with the grid is probably due to the beam-hardening and absorption of the grid interspace/covers and lead strips, respectively. Contrast of blood vessels in the lung Figures q-t show the variation of the OD with tube potential. The differences between the grid and the air gap technique are small. Some advantages with either technique are noted in certain regions of the lung. For example in the retro-cardiac area (RCA), there is a small increase (5%) in OD with the grid at a low tube potential compared to the air gap at the same tube potential. Dynamic range It is noted (figures u-v), that for the air gap a significantly larger percentage of the pixels correspond to values of γ>1.25 than is the case for the grid. The difference is between %. Also, there is a tendency for this percentage to increase with increasing tube potential Effect of tube potential ripple Patient Dose Figures a and c show the effective dose and entrance air kerma as a function of the percentage ripple of the tube potential at two peak tube potentials and total filtrations (12 kv / 3.7 mm Al, 141 kv / 5.7 mm Al). The air gap technique is used. Figures b and d show the same variations with E and Kair normalised to their values with the reference system (air gap, 141 kv, % ripple). The air kerma increases with increasing ripple and the increase is 7% and 18%, respectively for 141 kv and 12 kv at 5% ripple. The effective dose increases less with increasing ripple than the air kerma. At 141 kv the difference is less than 2% and within our statistical computational error. At 12 kv the increase in effective dose is 6% as the ripple increases from to 5%. Image Quality SNR of small calcification details in the lung apices 12

15 Figures e and f show the variations of the SNR with increasing ripple of the tube potential. The increase in SNR with increasing ripple is 12% and 9% for the two peak tube potentials 12 kv and 141 kv, respectively. The SNR is larger using the lower tube potential (see section ). With increasing ripple, the mean photon energy and the corresponding half value layer (HVL) decrease (see table ), hence increasing the SNR. Table Mean photon energy and half value layer, HVL, for different % ripple of the tube potential. 12 kv / 3.7 mm Al 12 kv / 3.7 mm Al 141 kv / 5.7 mm Al 141 kv / 5.7 mm Al Ripple (%) Mean energy (kev) HVL (mm Al) Mean energy (kev) HVL (mm Al) Contrast of blood vessels in the lung Figures g and l show the variation of OD of the blood vessels in the lung with increasing ripple. The increase in OD with increasing ripple is small at both tube potentials with a maximum increase of 4% at 5% ripple. Dynamic range Figures m and n show the corresponding variation with tube potential ripple of the percentage of the calculated pixels where the film gradient exceeds The variation of this quantity with increasing ripple is small and lies within our statistical uncertainty Effect of filtration The effects of filtration on the image quality descriptors and on patient dose were investigated for added filters of Al, Cu and for two K-edge filters (Nb and Gd). The results for each type of filter are reported separately in the following sections. Al filtration Patient Dose Figures a and c show the variation of the effective dose and air kerma with increasing Al filtration at 12 and 141 kv. Figures b and d show the same variations with E and Kair normalised to their values with the reference system, operating at 141kV, 5.7 mm Al (air gap technique). The air gap system operating at 141 kv/2.5 mm Al gives an E= 24.3 µsv and Kair= 11. µgy. For a tube potential of 12 kv/2.5 mm Al, the E is 23.8 µsv and the Kair µgy. At 141 kv with the lowest filtration (2.5 mm Al), the effective dose is 1.2 times larger than with the filtration of the reference system (5.7 mm Al). The corresponding air kerma is 1.25 times larger. For a tube potential of 12 kv, the effective dose changes by a factor between compared to the effective dose of the reference system for filtrations of mm Al. The corresponding factors for the air kerma are In conclusion, the effect of added Al-filtration on the effective dose is small, particularly at the higher tube potential, whereas the effect on entrance air kerma is larger, particularly at small thicknesses of added Al-filtration. 13

16 Image Quality SNR of small details in the lung apices Figures e and f show the variation of the SNR with increasing Al filtration. The SNR at the lowest filtration is 4.1 and 5.3 at 141 kv and 12 kv, respectively. The SNR decreases with increasing added Al-filtration (141 kv, LLA-detail) from 1.11 to.97 times that of the reference system. The corresponding factors at 12 kv are , for mm Al, respectively. Contrast of blood vessels in the lung Figures g and h show the variation of the OD with increasing Al filtration. The reduction in OD with increasing Al-filtration is small (<5%), particularly at the higher tube potential. Dynamic range Figures i and j show the variation with added Al-filtration of the percentage of the calculated pixels where the film gradient exceeds There is a small (4%) increase in this percentage with increasing Al filtration. Cu filtration Patient Dose Figures k and m show the variations of the effective dose and air kerma with increasing Cu filtration at 12 kv and 141 kv. The Cu filtration is added to a base filtration of 2.5 mm Al. Figures l and n show the same variations with E and K normalised to their values with the reference system, operating at 141 kv, 5.7 mm Al. The air-gap technique was used in this part of the study. At 141 kv, the effective dose is reduced from 1.2 to.97 times the value for the reference system as the Cu filtration increases between -.3 mm. At 12 kv, the corresponding reduction is for Cu filtrations of -.3 mm. For the air kerma, the reduction with increasing Cu filtration is much larger. At 141 kv, the air kerma is times that of the reference system. At 12 kv, the corresponding range of values is even larger, Image Quality SNR of small details in the lung apices Figures o and p show the variation of the SNR with increasing Cu filtration. The SNR without Cu filtration (2.5 mm Al) is 4.1 and 5.3 at 141 kv and 12 kv (not shown in the figures), respectively. The SNR decreases with increasing added Cu-filtration between -.3 mm (141 kv, LLA-detail) from 1.11 to.87 times that of the reference system (5.7 mm Al). The corresponding values at 12 kv are Contrast of blood vessels in the lung Figures q and r show the variation of the OD (CRL-detail) with increasing Cu filtration. The reduction in OD with increasing filtration is small (<9%), particularly at the higher tube potential and in regions behind the heart ( 2%). Dynamic range Figures s and t show the variation with added Cu-filtration of the percentage of the calculated pixels where the film gradient exceeds There is an 8% increase in this percentage with increasing Cu filtration (from to.3 mm Cu) at 12 kv. The corresponding value at 141 kv is 5%. Nb and Gd (K-edge) filtration Patient Dose Figures u and x show the variation of the effective dose and air kerma with increasing Nb filtration for the 12 kv and 141 kv tube potentials, respectively. Figures v and y show the same 14

17 variation of E and K normalised to their values with the reference system, operating at 141 kv, 5.7 mm Al. Corresponding figures for the Gd-filter are shown in figures ?-t. The air-gap technique was used in this part of the study. At 141 kv, the effective dose decreases with increasing Nb-filtration (-.15 mm Nb). The E with.15 mm Nb is.96 times that of the reference system. Using a Gd-filter, E instead increases with increasing filtration (1.11 times that of the reference system with.2 mm Gd). However, the air kerma decreases with increasing filtration for both the Nb- and Gd-filters; the reduction being larger with the Nb-filter. For example, with the.15 mm Nb filter, K is.78 times that of the reference system but with.19 mm Gd the corresponding value is only.96. Clearly, the Nb filter is not a 'true' K-edge filter but acts more like a 'high-pass' filter similar to Cu. Corresponding values using the 12 kv spectrum are for the effective dose:.88 (Nb) and 1.3 (Gd). For the air kerma, relative values are:.82 (Nb) and 1.2 (Gd). Image Quality SNR of small details in the lung apices Figures z-α and κ-λ show the variation of the SNR with increasing Nb and Gd filtration. With increasing filtration, the two filters reduce the SNR approximately by the same factor compared to the reference system. At 141 kv and the thickest filters, the SNR is.8 and.81 times the SNR for the reference system. The situation is different for the 12 kv spectrum. The use of the.19 mm Gd filter increases the SNR to 1.23 times that of the reference system. The corresponding factor is only 1.3 for the.15 mm Nb filter. Contrast of blood vessels in the lung Figures β-γand µ-ν show the variation of the OD with increasing Nb and Gd filtration. All OD -values decrease slightly with increasing Nb and Gd filtration. The reduction is, however, small and less than 8%. Dynamic range Figures ε and π-ρ show the variation with added Nb and Gd-filtration of the percentage of the calculated pixels where the film gradient exceeds As with added filters of Cu and Al, the percentage of pixels fulfilling the criterion increases slightly with increasing filtration. At 141 kv, the maximum increase, relative to the reference system, is 1.4 and 1.7 for the thickest Nb (.15 mm) and Gd (.19 mm) filters, respectively. At 12 kv, the corresponding factors are 1. and Effect of grid design The effects of the grid design on the image quality descriptors and on the patient dose were investigated. The grid ratio, r, was varied in three steps between r=8-16. Firstly, the effect of Al and fibre material in the grid covers and interspaces are reported and secondly, the effects of changing the width of the lead strips for a grid with 73 strips/cm is discussed. All values are for a tube potential of 141 kv and total filtration of 5.7 mm Al. Al or carbon fibre cover and interspaces Patient Dose Figures a and c show the variation of effective dose and air kerma with increasing grid ratio for grids with 4 lead strips/cm, 4 µm lead strip width and for grids with either aluminium (Al-grid) or carbon fibre materials (CF-grid) in covers and interspaces. Figures b and d show the same variations of E and K normalised to the corresponding values of the reference system. 15

18 The effective dose increases with increasing grid ratio by factors of (Al-grid) and (CF-grid) compared to the reference system for grid ratios between The corresponding increase in air kerma is by factors of and for the Al and CF-grids. Thus, E and K are between 8-18% higher (r=8-16) using the Al-grid compared to the CF-grid, the larger value using the higher grid ratio. One may expect the increase in air kerma and effective dose to be the same with increasing grid ratio. However, the reference system uses the air-gap technique and the conversion factor between entrance air kerma (without backscatter) and effective dose differs by 12% for the grid and air-gap geometries due to the slightly different positions of the x-ray beam on the voxel phantom. Image Quality SNR of small details in the lung apices Figures e-h show the variation of the SNR of the LLA detail with grid ratio. The SNR increases with increasing grid ratio. For the CF-grids, the SNR relative to the reference system increases with a factor from.9 to 1.3 as the grid ratio increases from 8 to 16. The SNR relative to the reference system is only for the Al-grid. The corresponding values for the RLA-detail are (CFgrid) and (Al-grid). Using the CF-grid, the SNR is improved by between 7-1% (LLA) and 2-5% (RLA). The fact that some factors are less than one indicates that the use of the grid is inferior to using the air gap technique for scatter removal and that the primary contrast is reduced with the grid. Contrast of blood vessels in the lung Figures i-n show the variation of the OD with grid ratio. The OD increases with increasing grid ratio. In the CRL, the OD normalised to that of the reference system increases to values between (Al-grid) and (CF-grid). In the RCA region, the corresponding values are (Al-grid) and (CF-grid). The values for the CPA region lie between those for the CRL and RCA regions. In conclusion, the CF-grid improves the OD (2-5%) of blood vessels in all regions compared to the Al-grid. Dynamic range Figures o-p show the variation with grid ratio of the percentage of the calculated pixels where the film gradient exceeds The percentage of pixels fulfilling this criterion decreases with increasing grid ratio, by factors of (r=8-16, Al-grid), compared to the reference system. Similar results were obtained for the CF-grid. Width of the lead strips Patient Dose Figures q and s show the variations of effective dose and air kerma with increasing grid ratio for grids with 73 lead strips/cm, carbon fibre interspaces and covers, and lead strip widths; either 36 µm or 2 µm. Figures r and t show the same variations with E and K normalised to their values with the reference system. For grid ratios between 8-16, the ranges of increases in effective dose relative to the reference system are (36 µm) and (2 µm). The corresponding increases in air kerma are and Thus, E and K are both between 19-22% (r=8-16) higher with 36 µm than with 2 µm lead strips. (An explanation of the different relative increases in K and E is given earlier in this subsection.) 16

19 Image Quality SNR of small details in the lung apices Figures u-v show the variation of the SNR with increasing grid ratio for grids with 36 and 2 µm lead strip widths. For the 36 µm, the SNR of the LLA-detail increases by a factor of (r=8-16) compared to the reference system. For the 2 µm grid, the increase is only by a factor of The corresponding values for the RLA-detail (not shown in the figure) are.8-1. (36 µm grid) and (2 µm grid). Using the 36 µm grid, the SNR is improved by 2-9%. Contrast of blood vessels in the lung Figures x-y show the variation of the OD with increasing grid ratio in the CRL region. The normalised values of OD increase for grids with 73 lines/cm used at 141 kv, by factors of (36 µm grid) and (2 µm grid). In the RCA region (not shown in the figure), the corresponding factors are (36 µm grid) and (2 µm grid). In conclusion, a 36 µm strip width improves the OD by 3-8% compared to a 2 µm grid. Dynamic range Figures z-α show the variation with increasing grid ratio of the percentage of the calculated pixels where the film gradient exceeds The percentage of pixels fulfilling this criterion decreases with increasing lead strip width and increasing grid ratio. Compared to the reference system, the fractional decrease is by factors of (r=8-16), using the 36 µm grid. For the 2 µm grid, the corresponding factors are higher: Effect of screen sensitivity Patient Dose Figures a and c show the effect on effective dose and air kerma when the nominal speed of the screen-film system is increased (Lanex 16/TML (reference system) and Lanex 32/TML). Figures b and d show the same change with E and K normalised to their values with the reference system. The effective dose and air kerma decrease by factors of compared to the reference system using the Lanex 32/TML screen-film system at 141 kv. While the decrease in E is the same at 141 kv and 12 kv, the decrease in air kerma (relative to the reference system) is by only a factor of.6 at 12 kv. Image Quality SNR of small details in the lung apices Figures e-h show the change in the SNR with increasing speed of the screen-film system. The SNR using the Lanex 32 is reduced to of the SNR using the Lanex 16 screen at 141 kv. Contrast of blood vessels in the lung Figures i-l show that there is essentially no change of the OD with speed of the screen-film system, or tube potential (141 kv, 12 kv). Dynamic range Figures m and n show the variation with increasing speed of the screen-film system of the percentage of the calculated pixels where the film gradient exceeds There is no change in this value with increasing speed of the screen-film system. The dependence on dynamic range on tube potential (141 kv, 12 kv) is commented on in section

20 Effect of optical density Patient Dose Figures a and c show the effective dose and air kerma as functions of the maximum optical density using the Kodak Lanex 16 screen and the Kodak TML film. Figures c and d show the same variation of E and K normalised to their values with the reference system (optical density, ODmax=1.8). The E and K both increase by the same factor with increasing optical density. The increase is linear up to OD=1.5. At OD>1.5, the increase is more rapid due to the shape of the film characteristic curve. The relative effective dose and air kerma at OD=1.3 are.64. Image Quality SNR of small details in the lung apices Figures e-h show the variation of the SNR with increasing optical density on the film. The SNR increases with OD up to an OD of about 2. where the film noise starts to be a significant part of the noise. The SNR of the detail in the RLA region is less than that of the detail in the LLA region since the OD (and thus the fluence of photons) is lower in the RLA region. Contrast of blood vessels in the lung Figures i-n show the variation of the OD with increasing optical density on the film in the three regions of the lung, CRL, RCA and CPA. The OD of the vessels in the CRL region increases rapidly with increasing OD up to OD max =1.7. At OD max =1.3, the OD is.94 of that at ODmax=1.8. Hence, the contrast in the central right lung is only little affected by the change in the exposure conditions. In the RCA region, however, the OD increases linearly with increasing ODmax and at ODmax=1.3 the OD is only.69 of that at ODmax=1.8. The variation of the OD in the CPA region resembles that in the CRL. There is thus a significant improvement in the OD of blood vessels with proper exposure of the film in certain regions in the image. Dynamic range Figures o-p show the variation with ODmax of the percentage of the calculated pixels where the film gradient exceeds The percentage of pixels fulfilling this criterion increases rapidly with increasing ODmax. At ODmax=1.8, 91% of the pixels fulfil the criterion. The value at ODmax=1.3 is.69 of that at ODmax= Effect of film H&D curve Patient Dose Figures a and c show the effective dose and air kerma as function of the maximum optical density using the Kodak Lanex 16 screen and four films with different gradients. The films were normalised to have the same log kerma at OD=1.. This normalisation is probably not the most realistic one and hence the results on patient dose should be treated with some caution. The four cases simulate films with a high gradient (G) a medium gradient (M), the reference system film (Kodak TML) and a film with a wide latitude and low gradient (IL2). Figures b and d show the same variation of E and K normalised to their values with the reference system (ODmax=1.8, Kodak TML film). The patient dose increases rapidly with decreasing film gradient because of the way the film characteristic curves have been normalised (Herrmann 1997). At OD above 1. the H&D curves of the films start to differ and at higher OD-values the sensitivity of the IL2-film is less than that of the G-film. 18

21 Image Quality SNR of small details in the lung apices Figures e-h show the variation of the SNR with increasing optical density on the film for the four different films. The way the film H&D curves are normalised is also reflected in the values of the SNR since the films that requires the highest dose (low gradient films) also have higher SNR at the higher optical densities. Contrast of blood vessels in the lung Figures i-n show the variation of the OD with increasing maximum optical density on the film for the four films in the three regions of the lung, CRL, RCA and CPA. In all three situations the film with the highest gradient results in the highest OD. This is more pronounced in the CRL and CPA regions and in the RCA region at ODmax>1.5. Dynamic range Figures o-r show the variation with ODmax of the percentage of the calculated pixels where the film gradient exceeds The results for the G and M films are similar but for the TML and IL2 films the variation with ODmax is different. At low ODmax<1., the IL2 film results in a lower percentage of pixels fulfilling the criterion, but at ODmax>1.5, the opposite is true for both this film and the TML film. 19

22 3.1.2 Lumbar Spine AP Effect of Tube Potential Patient Dose Figures a and c show the variation of effective dose and entrance air kerma without backscatter, with x-ray tube potentials between 6 kv and 11 kv. Figures b and d show the same variation with the effective dose and the entrance entrance air kerma normalised to the reference system, operating at 72 kv. The reference system gives an effective dose of µsv and an entrance air kerma of µgy to the patient. At the lowest tube potential, the effective dose is 1.73 times larger than at 72 kv and the entrance air kerma is 2.1 times larger. At the highest tube potential, the effective dose is.47 times lower than at 72 kv and the entrance air kerma is.34 times lower. Image Quality Contrast of Transverse Processes Figure e shows the variation of optical density difference ( OD) for 2 mm, 3.5 mm and 5 mm of cortical bone, representing the transverse processes on the L5, L3 and L1 vertebrae, with x-ray tube potentials between 6 kv and 11 kv. Figure f shows the same variation for the OD normalised to the reference system, operating at 72 kv, for the L3 transverse process. The other transverse processes produce approximately the same reduction in OD for increasing tube potential. The reference system gives the differences between the optical density on the contrast detail and the background soft tissue as: ODU for the L5 process, ODU for the L3 process, and.255 ODU for the L1 process. For the L3 process, the OD is 1.38 times larger at 6 kv than that for the reference system at 72 kv and.59 lower at 11 kv. SNR for Trabecular Structures Figure g shows the variation of SNR for a 1 mm bone marrow cavity within cortical bone, representing trabecular structure on the L5, L3 and L1 vertebrae, with x-ray tube potentials between 6 kv and 11 kv. Figure h shows the same variation for the SNR normalised to the reference system, operating at 72 kv, for the trabecular structure on the L3 vertebra. The trabecular structures on the other vertebra produce approximately the same reduction in SNR for increasing tube potential. The reference system gives the SNR as: 5.16 for the L5 vertebra detail, 4.4 for the L3 vertebra detail, and 4.1 for the L1 vertebra detail. For the L3 vertebra detail, the SNR is 1.33 times larger at 6 kv than that for the reference system at 72 kv and.58 lower at 11 kv. There is a similar response to tube potential for both the contrast and SNR details Effect of Filtration Patient Dose Figures a and c show the variation of effective dose with x-ray tube filtrations between 2.5 mmal and 4.7 mmal for tube potentials of 72 kv and 9 kv. Figures b and d show the same variation with the effective dose and the entrance air kerma normalised to the reference system, operating at 72 kv with a total filtration of 4.7 mmal. The reference system operating at 72 kv gives an effective dose of µsv and an entrance air kerma of µgy to the patient. For a tube potential of 9 kv, the effective dose is 73.9 µsv and the entrance air kerma is µgy. 2