| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Full Paper |
1 Medical Physics Department, Guy's and St Thomas' Hospital, London SE1 7EH, 2 King's Centre for Assessment of Radiological Equipment (KCARE), King's Hospital, London, 3 Medical Physics Department, Princess Margaret Hospital, Swindon and 4 Formerly Medical Physics Department, Guy's and St Thomas' Hospital Trust, London SE1 7EH, UK
 |
Abstract |
|---|
 Top Abstract IntroductionTheoryMethodsResultsDiscussionConclusionsReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionTheoryMethodsResultsDiscussionConclusionsReferences |
|---|
A series of such test objects were produced by the FAXIL group at the University of Leeds [5] and we will consider the use of the TO10 test object primarily designed for image intensifier TV fluoroscopy systems. An accepted protocol for the use of this test object in serial testing has been in widespread use over a long period [6–9] and is recommended in IPEM 77 [10].
Two contrast threshold curves obtained with the TO10 test object were published in report STB/7/82 [6]. One curve was obtained from measurements on new systems (we shall term STB1) and the other for systems of average age 5 years (we shall term STB2). Both curves were for fields of view (FOV) in the region of 23–25 cm and II input plane air kerma rates of 260 nGy·s-1. These two curves have been used for many years as standard reference data with which to compare image intensifier threshold contrast detail detectability (TCDD) performance. The measured II system data were normally visually compared with the STB/7/82 data on a log-log graph plot of contrast threshold versus detail diameter.
FAXIL published further data in the 1992 Test Object Users Manual [9] in which a curve for the 25 cm FOV was produced (we shall term F25). Also, curves relating to image intensifier FOVs in the ranges 30/36 cm (we shall term F33) and 15/17 cm (we shall term F16) were produced in the users manual [9]. The contrast threshold data in the 1992 Users Manual was converted to a new quantity however using the concept of a threshold detection index HT(A), which takes into account the detail area and is given in Equation 1
. Furthermore, the threshold detection index is plotted against square root of detail area as opposed to detail diameter.
|
|
where A is detail area and CT is threshold contrast.
From Equation 1
it can be seen that the smaller the contrast threshold CT, the greater the value of the threshold detection index, corresponding to a higher image quality score. For all three data curves (F16, F25 and F33) the II input kerma rate was not specified.
Launders et al [11] updated the viewing protocol for FAXIL TCDD test objects to include a variable distance protocol (VDP) measurement. This enables the observer to vary their viewing distance from the display monitor to optimize the image viewing conditions for both large and small test object details and hence tends to result in slightly higher image scores [11].
From the measurement data, TCDD curves are drawn and are useful in quantifying the image quality of the system. The TCDD curves are not intuitively simple and the production of a single quality image factor to describe image quality is useful. McRobbie et al [12] devised a single figure of merit F for image intensifier systems.
|
|
n=noise (using a Leeds N3 test object at one fixed detail size);
Kair=II input kerma rate (nGy·s-1); and
k=1.48 mm·nGy1/2·s-1/2.
The constant k was chosen to normalize F to unity for values n=3%, Kair=0.35 µGys-1, and R=1.2 line pairs per mm (lp·mm-1). These values were chosen to reflect a typical level of performance. The F index has been shown to be a useful measure, but may suffer from the following potential limitations. First, the figure of merit F is proportional to the limiting high contrast spatial resolution, which was measured without attenuation and with the system operating under automatic exposure factors. However, II input kerma rate and threshold contrast are measured under different conditions with a copper attenuator. Second, the limiting high contrast resolution is primarily designed to ensure that the system is optimally focused [9]. Defocusing is readily adjustable and the high contrast spatial resolution does not deteriorate significantly over time and so does not contribute to image quality factors. Clinically low contrast small details are more significant than high contrast small objects. The McRobbie quality factor only uses one detail size (1.1 cm diameter) and so cannot indicate a true image quality factor.
We believe that TCDD curves measured with the TO10 test object are a more useful basis for evaluating overall image quality. The height of the curve is closely related to the overall system SNR and provides a wide range of detail: the authors suggest our measure may be more sensitive to system performance over a range of detail sizes.
There are confounding factors in attempting to make a comparative index of performance. Problems with the use of subjective visual assessment test phantoms have been noted [13], and other performance measurements of a more objective nature such as the measurement of detective quantum efficiency (DQE) will remain the benchmarks for verifiable assessment of imaging systems. However, these types of measurement may not be practical on a routine basis. The availability of a field measurement protocol which aids serial testing and comparative analysis of systems would be a useful resource for the physicist performing both routine quality control (QC) and commissioning field tests.
As detailed in the update on the recommended viewing protocol by Launders et al [11], care does need to be exercised in making comparative measurements between systems because of differences in beam quality, II input kerma rate, II size and FOV selected for the measurement. Any image processing differences, e.g. noise reduction, will also need to be considered so as not to confound the validity of any such comparison.
In this paper we demonstrate a simple method for making a comparison between systems using the threshold contrast detail Leeds Test object TO10 using the parallel property of TCDD curves. Corrections of II input kerma rate, beam quality, field size and pulsed fluoroscopy can be incorporated.
|
Theory |
|---|
|
TopAbstractIntroductionTheoryMethodsResultsDiscussionConclusionsReferences |
|---|
F25, STB1 and STB2 curves were based upon measured data from several systems [6, 9] and therefore their overall shape is generally representative. All the curves follow a characteristic peaked shape (Figure 1). The use of scaling (normalization) factors was investigated to test if a single characteristic curve, with appropriate normalization factors could be used as a standard reference curve. Accordingly, STB1 and STB2 were scaled and best fitted to F25 by using the mean ratio of threshold detection index for all respective detail sizes to F25 to estimate the scaling factors. Figure 2 shows the very close correlation between F25, STB1 and STB2, which is not surprising as all relate to a 25 cm image intensifier. This justifies the use of a single quality index using the parallel property.
|
|
. The ratio of the measured detection index HT(Ai) for a given detail area Ai against the corresponding value for the F25 curve HTref(Ai) is taken (Table 1). The mean value of these ratios, summed over all available detectable data points, will be the quality index.
|
|
|
From the literature [6, 9, 11] all these curves were based on a range of systems, with different image intensifiers, operating at different efficiencies and with different TV camera lag characteristics. All reference curve data were at 70 kVp and 1 mm copper beam attenuation.
Input kerma rate correction
The image quality of a system will depend on the II input kerma rate measured at the II entrance plane. As the dose rate is increased the quantum noise will be reduced, allowing easier visualization of low contrast details. In order to compare systems, or even the same system, operating at different II input kerma rates, the quality index Q needs to be modified by a dose normalizing factor. The dose normalization is based on the quantum noise which follows Poisson statistics (FAXIL [11], McRobbie [12]). The noise will therefore be inversely proportional to the square root of the number of detected photons, or their input rate [14]. The threshold detection index is itself proportional to the SNR of the system and hence modifying factors applied to this index must reflect the effect of quantum statistics on noise level [11]. Hence, the modifying factor applied to the threshold detection index values will be the square root of the ratio of the actual measured input kerma rate against a standard reference value of input kerma rate. Other sources of noise in the imaging chain will be present such as electronic noise or intensifier structural noise, however relative to quantum noise these are negligible in fluoroscopy [15].
The quality index will be scaled to that which would be obtained when running the system at a fixed reference value (Table 2). Dose normalized quality index Qn is first defined in Equation 4.
|
|
applied to the known Q value of the STB1 curve (0.78) at the stated input kerma rate of 261 nGy·s-1 for this data [6]. Km is the measured input kerma rate for a given system under consideration.
|
Further corrections are required owing to variations in inherent subject contrast of the test object with different tube potential (kVp) and beam attenuation (in mm of copper), and variation of energy absorption in the II input phosphor with different beam qualities must also be taken into account.
Correction of contrast differences in the test object with different beam qualities
In older protocols [1, 2, 5, 6] the specified test conditions were 70 kVp with 1 mm copper beam filtration for the TO10 test object. However, in the 4th edition of the Leeds test objects manual [9] threshold contrast values were published between 65 kVp and 80 kVp at 5 kVp intervals at 1 mm, 1.5 mm and 2 mm levels of copper beam filtration for all the rows and details sizes in the TO10 test object. Measurements may then be made on systems for which standard factors of 70 kVp with 1 mm copper cannot be obtained. The true threshold contrasts can be measured taking into account the differences in radiation beam quality. These corrected threshold contrasts are used to calculate normalized quality index Qn.
Correction of image intensifier input phosphor absorption efficiency with beam quality
The correction factor for input phosphor absorption is necessary because the ratio of mass energy absorption coefficients of the caesium iodide (CsI) II input phosphor to air is not constant over the range of beam qualities used in the Leeds TO10 test object measurements. Hence the energy absorption in the image intensifier input phosphor, for a given input air kerma at the intensifier face, is a function of radiation beam quality. It follows that this also affects the conversion efficiency of the image intensifier as a whole and so a correction should be applied to allow comparison with other systems and various dose level modes on a system.
Field measurements of systems must generally be carried out in automatic mode where the tester has no direct control over system tube potential and current. The fluoroscopy automatic brightness control (ABC) system will control operating factors to ensure a constant luminance from the output phosphor of the image intensifier by following a characteristic kVp/mA curve. Hence, for any given operating mode, the tester must add sufficient copper beam attenuation to attain a certain tube potential range (65–80 kVp) generally attempting to achieve close to 70 kVp. We therefore require a correction to be applied to convert test conditions (kVp and copper thickness) back to standard values of 70 kVp and 1 mm Cu used for the reference curves.
The output luminance of the II will be proportional to the rate at which energy is deposited in the input phosphor. In maintaining a constant luminance by feedback control the ABC effectively regulates a constant energy deposition rate in the input phosphor [16]. In practice only the II input kerma rate is easily measurable. The ratio of energy absorbed in the CsI input phosphor to measured input air kerma will vary depending upon the mean energy of the fluence spectrum [17]. Values of the ratio of mass energy absorption coefficients for CsI and air were calculated from published data [18] at different photon energies in the relevant range and are presented in Table 3. By using the Institute of Physics and Engineering in Medicine (IPEM) Report 78 X-ray spectrum processor code [19, 20] radiation beam qualities for the full range of test conditions were investigated. Mean spectral energy for each kVp and copper filter combination was calculated using the spectrum processor and values are presented in Table 4. Values of the ratio of mass energy coefficients from CsI to air, denoted R(V,X) where V is the tube potential and X the total thickness of copper added to the beam, were estimated from a best fit cubic from values in Table 3. In order to apply these results in a ready form a ratio Rn(V,X)=R(V,X)/R(70,1.0) has been calculated, normalized to the standard conditions (Table 4). The modified form of Equation [4] is shown later in Equation 5 together with a further correction factor still to be discussed.
|
|
Quality index for different FOV sizes
The normalized quality index Qn can be extended to cover all FOVs. Launders [11] details a method to extend the use of the TO10 test object to different FOVs. In this approach the relative sizes of the FOVs are used to correct the apparent detail size between different FOVs and the TCDD curves are plotted on the same square root of detail area axis to enable comparison between performance on different FOV sizes. Appropriate kerma rate normalization factors are employed to modify the overall magnitude of the TCDD curves to account for different input kerma rates that relate to the different FOVs.
The FAXIL data for the 30/36 cm FOV (F33) and 15/17 cm FOV (F16) were used [9] (Table 1). Apparent detail size correction factors were calculated from the ratio of these sizes (33 cm and 16 cm, respectively, for the mid point of each range) to F25. These were then used to correct the square root of area (horizontal axis) ordinate points plotted for each respective curve F33 or F16 so that the apparent detail size will match F25. Curves F16, F25 and F33 are plotted in Figure 3, where the curves all follow the same general shape when corrected by the ratio in apparent detail size. The reference exposure for F16 and F33 can be estimated using the parallel property of these curves to F25. A minimum root mean square (RMS) difference best fit was used to calculate the normalization factors for best fitting the curves. The multiplying factors applied to each curve to obtain the best fit were 1.30 for F33 and 0.78 for F16. These are presented in Figure 4, where the curves show good correlation, justifying that all quality index data can relate back to a single reference curve F25. From these normalization values, the values of Kref relating to each respective FOV were calculated from Equation 4 and are shown in Table 2.
|
|
Extending the quality index to cover pulsed fluoroscopy systems
When pulsed fluoroscopy is used the temporal averaging effect by the observer is reduced. To compensate for this some manufacturers may increase the dose per pulse to reduce the noise level. The overall effect may nevertheless be dose reduction. Aufrichtig et al [23] investigated the effect of pulse rate and dose on perceptual quality of fluoroscopic images. They found that dose reductions were possible for a matched image score (equivalent perception dose) in threshold detection measurements. Mean dose savings of 22%, 38% and 49% for fluoroscopy at 15, 10 and 7.5 pulses per second (pps), respectively, were reported. A pulse rate of 30 pps was treated as continuous by Aufrichtig on the basis that perceptual differences at higher pulse rates than this were small owing to limitation in the human visual system [23]. It is proposed to make allowance for pulsed fluoroscopy modes by extending the definition of the quality index. The approach taken here will be to introduce an equivalent perception dose factor 
(f) calculated from Aufrichtig's results, where f is the pulse frequency in pps (Table 5). Under this definition a system operating at a pulsed fluoroscopy frequency f would need to be operating at a fraction (f) of the input kerma rate in order to show an equal normalized quality factor Qn compared with the continuous fluoroscopy mode of operation. This is justified as the equivalent perception dose at a given pulsed sampling frequency in fluoroscopy should give rise to the same value for normalized quality index Qn when comparing different pulsed frequencies on the same system.
|
is useful for serial testing, i.e. repeat testing on the same equipment under specified conditions. The normalized quality index Qn defined in Equation 5 must be used to compare systems operating on different FOVs, under different II input kerma rates, at different pulsed fluoroscopy rates and under different beam test conditions.
|
|
where Km=measured input kerma rate of the system in nGy·s-1;
Kref=reference input kerma rate of the system in nGy·s-1;
Rn(V,X)=radiation beam quality normalized factor and is a function of V (tube potential) and X (beam attenuation in mm copper); and
(f)=pulsed fluoroscopy correction factor.
|
Methods |
|---|
|
TopAbstractIntroductionTheoryMethodsResultsDiscussionConclusionsReferences |
|---|
The results were reviewed for four hospital fluoroscopy systems A, B, C and D during use of the quality index as part of a QC programme.
The Leeds TO10 contrast detail test object was placed on the cover plate of the image intensifier. Sufficient copper was placed in the beam until a kVp in the range 65 kV to 80 kV was reached. The air kerma rate was measured using an MDH 9010 ion chamber and electrometer system. In accordance with the recommended protocols in the instruction book [9] the data were obtained using a fixed distance viewing protocol (FDP) of approximately four times the diameter of the blanking circle from the system display monitor. Previous values were only examined after the tests had been completed so as not to influence their impression of the image quality of the system. Where possible, both observers were used. The value of Observer 1 was involved in all the measurements.
It is recognized that the presence of measurement errors due to within observer, between observer and between sample variance will be present in these measurements [24, 25]. No attempt has been made to quantify these in this study, hence reflecting field practice measurements and not repeated tests under laboratory conditions. However, regular blind testing comparisons were made to verify overall image score consistency between the observers in this study. It is recognized that the variations in threshold contrasts will be of the order of 16% for single observer and 11% for two observer measurements [25].
Some measurements were carried out using the VDP to investigate this effect on the image quality of systems C and D.
|
Results |
|---|
|
TopAbstractIntroductionTheoryMethodsResultsDiscussionConclusionsReferences |
|---|
shows results from system A, a general purpose over couch tube fluoroscopy system with a 27 cm FOV II, over 5 years (the system was installed in 1989) with results from Observer 1. It can be seen that the Q index falls slightly over time as the system ages. II input kerma rates were increased over time to compensate for conversion loss, increased camera noise and monitor ageing to maintain image quality. The Qn values show a more significant decline, which reflects the increasing input kerma rate at the intensifier with this system. No major TV chain components, e.g. image intensifier, TV camera or display monitor, were changed during the sample period.
|
Table 7 shows results for system B, a new large format II system with pulsed fluoroscopy modes. Even with the large differences in input kerma rate in the different modes of operation the Qn values reflect satisfactory imaging performance according to these indicators. This is also true with the pulsed fluoroscopy modes. It should be noted that these pulsed modes of operation result in a loss of temporally changing information, because of the lower rates of sampling, and this is not reflected in a static test object test. Also, the presence of noise reduction in the digital image processing algorithms will be expected to have an effect on image quality. The presence of online digital image processing and pulsed fluoroscopic acquisition of images may be relevant factors for consideration on data from systems of recent manufacture.
|
. For comparison, VDP data have been used to calculate quality index values although these are based on FDP curves.
|
it can be seen that the system has shown consistently satisfactory performance over the 3 sample survey years. It should be noted that consistent values for Qn were obtained across all three FOVs tested. The VDP gave a greater value of Qn due to enhanced detail scores compared with FDP. This is because at shorter viewing distances visualization of small details is improved, and at larger viewing distances the large detail score tends to be enhanced.
Further measurements of Qn using the FDP were carried out on system D, a general purpose C-arm fluoroscopy system, and are presented in Table 9. These data again show satisfactory performance over the sample years, and a general trend for the VDP measurements to score greater Q index values than the FDP counterparts.
|
|
Discussion |
|---|
|
TopAbstractIntroductionTheoryMethodsResultsDiscussionConclusionsReferences |
|---|
), where a value of 1 is equivalent to the image quality of reference curve F25. This value takes into account II input kerma rate, beam quality, FOV size and pulsed fluoroscopy. The quality index has been used successfully in Guy's and St.Thomas' Hospital for many years for serial testing of equipment and comparing systems during equipment evaluation for purchase.
From many years of experience, image assessment criteria are proposed as an aid to the classification of equipment performance levels in Table 10. These may subsequently be reviewed on the basis of a larger sample data set, perhaps with appropriate modifications for pulsed fluoroscopy and operation of digital noise reduction image processing.
|
While it was necessary to estimate air kerma rates for the F16, F25 and F33 curves, this does not affect the relationship between the curves. Future work may nevertheless produce a new curve and associated II input kerma rate reference level with which to compare fluoroscopy systems.
As recommended in the update on viewing protocol [11] the same approach could be applied to data obtained with VDP after establishing reference curves based on a VDP approach as the shape of the VDP curves may be expected to differ from FDP. It is important to specify when plotting the data obtained and calculating the quality index whether these relate to the FDP or VDP. The production of curves using VDP will be the subject of further work.
The standardization of observer scoring levels between centres and between individual observers remains a potentially problematic area, but this could perhaps be assisted by the introduction of standard TCDD hard copy images with which individuals working within the field could assess their own scores against a standard reference level. At a local level consistency should be verified between observers.
The presence of different digital image processing algorithms, such as noise reduction, should also be taken into account when making any comparative assessment of systems. Systems with high lag, or with digital frame averaging noise reduction techniques will score a higher Qn value. However, these images could suffer from movement artefacts.
Dynamic imaging characteristics may be subject to other tests but it is not intended at this stage to take these into account within the dose normalized quality index Qn.
|
Conclusions |
|---|
|
TopAbstractIntroductionTheoryMethodsResultsDiscussionConclusionsReferences |
|---|
Making comparisons between systems is inherently difficult as there are many factors that would need to be taken into account and not all of them are known. The quality index Qn is a useful comparative indicator which assesses the relative efficiency taking into account differences in II input kerma rate, beam quality and fluoroscopic pulse sampling rate. However, this number cannot be used in isolation and factors such as lag, distortion and high contrast resolution must also be considered.
The use of the quality index approach will facilitate more effective operation of quality control programmes and assessment of criteria for replacement or upgrade of systems on a more objective basis than the comparison of TCDD curve data alone. The system will enable clearer communication with the radiology department. The proposal that most modern systems in good adjustment would be expected to return normalized quality index values Qn of greater than 1.0 has been supported by results obtained in this study.
Received for publication February 22, 2001. Revision received March 18, 2003. Accepted for publication April 10, 2003.
|
References |
|---|
|
TopAbstractIntroductionTheoryMethodsResultsDiscussionConclusionsReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  C. Walsh, D. Gorman, P. Byrne, A. Larkin, A. Dowling, and J. F. Malone QUALITY ASSURANCE OF COMPUTED AND DIGITAL RADIOGRAPHY SYSTEMS Radiat Prot Dosimetry, March 4, 2008; (2008) ncn047v1. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. Kepler and A. Vladimirov Optimisation strategies introduced for CR at health care centres in Estonia Radiat Prot Dosimetry, March 1, 2008; 129(1-3): 127 - 131. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 I D Honey, A Mackenzie, and D S Evans Investigation of optimum energies for chest imaging using film-screen and computed radiography Br. J. Radiol., May 1, 2005; 78(929): 422 - 427. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D S Evans, A Mackenzie, C P Lawinski, and D Smith Threshold contrast detail detectability curves for fluoroscopy and digital acquisition using modern image intensifier systems Br. J. Radiol., September 1, 2004; 77(921): 751 - 758. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| BJR | DMFR | IMAGING | ALL BIR JOURNALS |
=F25, FAXIL Data (F25, 1992); x =STB1, STB/7/82 New Systems Data (DOH, 1982); +=STB2, STB/7/82 Systems of age 5 years (DOH, 1982).
=F33, FAXIL Data (F33, 1992); 
=F16, FAXIL Data (F16, 1992).
