British Journal of Radiology (2004) 77, 675-678
© 2004 British Institute of Radiology
doi: 10.1259/bjr/72726487
Optimization of variable temporal averaging in digital fluoroscopy
C J Kotre, PhD1 and
E Guibelalde, PhD2
1 Regional Medical Physics Department, Newcastle General Hospital, Newcastle-upon-Tyne NE4 6BE, UK and 2 Department of Radiology (Medical Physics), School of Medicine, Complutense University of Madrid, 28040 Madrid, Spain
 |
Abstract
|
|---|
In modern X-ray fluoroscopy systems, the amount of temporal averaging (i.e. persistence) applied to the image is often user selectable. The objective of this work is to quantify the effect of variable temporal averaging on the detection of low contrast test objects moving at a range of known speeds within the digital fluoroscopic image. An image intensifier system with a short-persistence television camera was used to record image sequences of a moving threshold contrast-detail diameter test object onto broadcast-standard U-matic videotape. The image sequences were replayed through an image processing system allowing different amounts of temporal averaging to be applied. The test images were scored by an experienced observer. The temporal averaging time constants produced by the added image processing were measured using a method based on noise correlation. Results are presented showing the trends of threshold contrast with test detail diameter and movement speed. The optimum value of temporal averaging time constant is presented as a function of detail diameter for a range of speeds. By comparison with the limited information available in the literature on organ movement, it is tentatively concluded that for the organ movement speeds expected in the abdomen, the optimum imaging system persistence time constant should be approximately 0.15 s. For the much greater speeds associated with cardiac motion no additional frame averaging, i.e. just the persistence provided by the observer's visual system, appears to be optimal.
|
Introduction
|
|---|
Random noise fluctuations in fluoroscopy arising from both quantum and electronic sources can be reduced by temporal averaging. In older designs of X-ray image intensifier, a fixed degree of temporal averaging is provided by the persistence of the various components of the system such as the input and output phosphors of the image intensifier, the vacuum-tube TV camera and the TV display monitor. The visual system of the observer also includes an element of persistence which is vital for the perception as continuous moving images of TV frames refreshed at 25 frames per second. In modern image intensifier-TV systems, the fluoroscopic image is viewed in a digital format consisting of typically 512 x 512 or 1024 x 1024 pixels, updated at the usual TV refresh rate. The main function of this digitized form of display is to provide a last image hold facility upon the cessation of fluoroscopy, but the presence of a digital frame store in the system also allows the provision of variable persistence by real-time digital weighted frame averaging. For systems incorporating this facility, the persistence of the image intensifier-TV system becomes a user controlled variable which requires some care in its selection. If a long persistence is chosen, a large reduction in image noise can be achieved with a resulting improvement in image quality or reduction in dose-rate required for the same image quality. If the selected persistence is too long, however, the system may be unusable in clinical practice due to unacceptable smearing of moving structures. Conversely, if a very short persistence is chosen, the image noise may be such that low contrast structures cannot be detected without increases in dose-rate. The objective of the present work is to quantify the effect of variable temporal averaging on the detection of low contrast objects moving at a range of known speeds within the digital fluoroscopic image, as a preliminary to optimization in clinical practice. This work differs from recent detailed work on the human perception of temporally averaged images [1], and of the effects of motion blurring [2] which used simulated image sequences, in that the measurements were carried out using real fluoroscopic image sequences to provide a realistic balance of noise sources. In particular, the static noise associated with inhomogeneity of the image intensifier input phosphor, which can dominate at long persistence and/or high dose-rates, is included.
|
Experimental method
|
|---|
The laboratory digital fluoroscopy system consisted of a Thomson CSF (Meudon la Foret, France) TH 9435 E GKV1 image intensifier with a 23 cm diameter field of view, coupled to a Pulnix TM 765 CCD television camera (Sunnyvale, CA). A TV camera based on a charge-coupled device (CCD) was used to provide a low intrinsic persistence for the imaging system [3]. The X-ray generator used was a Picker Vector 70 fitted with a Picker PX401P X-ray tube. To provide fluoroscopic test sequences to which varying amounts of temporal averaging could be applied, the television signal was recorded on a broadcast-standard U-matic video recorder then played back via an Imaging Technology (Woburn, MA) Series 151 image processing unit and finally displayed on a short-persistence monochrome studio television monitor.
The image sequences consisted of recordings of the well-known Leeds TO10 threshold contrast-detail diameter test object being moved back and forth horizontally across the image intensifier field of view by a PC-controlled robot arm which has been described previously [4, 5]. The Leeds TO10 test object contains sets of low-contrast circular test features ranging from 0.25 mm to 11.1 mm in diameter, with the calibrated contrasts for each diameter decreasing in steps of 
2. To find the threshold contrast, the observer counts the disks visible above the background noise at each diameter. The TO10 object is a standard test-tool for quality assurance measurements on image intensifier systems [6]. The test object was irradiated with a 70 kVp X-ray beam with additional filtration of 1 mmCu to give the correct contrast calibration for the test details. The dose-rate at the image intensifier input plane was fixed at 0.5 µGys–1, which is a reasonably typical value for an image intensifier of this diameter field of view (see discussion). The speed of movement was selectable, and nominal speeds of 0 mms–1, 6.25 mms–1, 12.5 mms–1, 20 mms–1, 40 mms–1 and 75 mms–1 were used in the experiment.
The test object sequences were scored by one experienced observer for each of five different amounts of frame averaging applied during video playback by the image processing hardware. The reading of each disk size was carried out as near as possible to mid-sweep, so that the object was known to be moving at its calibrated speed. No time restriction was applied to the observations and where necessary, the observer re-ran the videotape until a firm reading was obtained. The observation distance was maintained at approximately 60 cm. No advantage was found from moving closer to observe the smaller objects.
The real-time frame averaging algorithm can be described as 
where S(W)n is the nth averaged pixel value, S(W)n–1 is the previous averaged pixel value at the same position in the image, Pn is the incoming digitized pixel value, and W is a weighting factor governing the amount of frame averaging applied [7]. The weighting factor W controls the proportion of image information from the current frame which is added into the frame averaged total. The values of W used in the experiment were 1, 2, 4, 8 and 16, where 1 corresponds to no additional frame averaging and 16 corresponds to a highly averaged image with pronounced movement blurring. This algorithm is equivalent to the forms used by previous authors [1, 8], but uses a weighting factor W which more realistically expresses the process as implemented using binary arithmetic in electronic hardware for real-time processing.
A previous experiment was carried out to quantify the amount of temporal frame averaging corresponding to the various values of W. The method employed was based on the correlation between successive image frames [3], which has the advantage that the persistence can be measured without unrealistic large-signal perturbations such as scanning slits or bars. A digital frame store was used to acquire temporally contiguous fluoroscopy images of a 1.5 mm copper filter at a low dose-rate. The correlation of the variance between an initial base TV frame and successive later frames was then measured by the correlation coefficient. Plotted against time, this function can be fitted by a simple exponential decay from which a time constant can be derived. The time constant characterizes the rate at which the initial variance pattern is replaced by incoming quantum noise, and is therefore a measure of the persistence of the system. The persistence due to the TV display and observer are additional to this measured value.
In order to find the optimum persistence for each diameter of test detail at each speed of movement, the observed threshold contrasts were plotted against the persistence time constant corresponding to the frame averaging used. A simple cubic curve-fit was applied in each case and the persistence at which the minimum threshold contrast (corresponding to the maximum signal-to-noise ratio) occurred was recorded.
|
Results
|
|---|
The measured time constants corresponding to the weighting factor values W=1, 2, 4, 8 and 16 are given in Table 1 for the experimental system used here and for two common commercial fluoroscopy systems (data from [3]). There is reasonable correspondence between the time constant values for the various settings, making it possible to interpret the optimum time constant values given below in terms of settings on these units.
View this table:
[in this window]
[in a new window]
|
Table 1. Persistence time constants for the experimental system used in this work compared with the equivalent results for two commercial image intensifier systems
|
|
Figure 1 shows an example of the trend in threshold contrast with increasing frame averaging time constant, for the range of speeds. This example is for the 11.1 mm disk diameter. The illustrative error bar represents a standard deviation of 16% on each observation, based on the work of Marshall et al [9] for static threshold contrast tests in fluoroscopy. It is possible that this underestimates the observation error for the faster moving disks, where the observer is tracking the position of the object as well as making a decision on its visibility.
View larger version (18K):
[in this window]
[in a new window]
|
Figure 1. An example of the threshold contrast results obtained against frame averaging time constant, for a range of speeds. This example is for the 11.1 mm disk diameter. The illustrative error bar represents a standard deviation of 16% on each observation.
|
|
The method for deriving the optimum frame averaging time constant is illustrated in Figure 2. This example is for an 11.1 mm disk moving at 20 mms–1. The curve fit is a cubic, and the optimum frame averaging time constant is defined as that which corresponds to the minimum point on the curve, shown by the arrow in Figure 2. In order to make an estimate of the experimental error on this value, the standard error on the position of the curve fit was taken as 16%/n, where n is the number of experimental points (usually n=5, but where no observations could be made, e.g. small, fast moving disks, n<5). The 1 standard error limits are shown as dashed lines in Figure 2. The estimate of the standard error on the optimum frame averaging time constant was then taken as the range over which the fitted curve could be translated and remain within the 1 standard error limits.
View larger version (14K):
[in this window]
[in a new window]
|
Figure 2. Illustration of the method for deriving the optimum frame averaging time constant. This example is for the 11.1 mm disk diameter moving at 20 mm s–1. The solid arrow shows the minimum of the curve, taken as the optimum time constant. The curved dashed lines indicate an estimate of the standard error about the cubic curve fit. The resulting estimate of standard error on the optimum time constant is shown by the horizontal error bar.
|
|
Figure 3 shows the optimum persistence time constant as a function of disk diameter for speeds of 6.25 mm s–1, 12.5 mm s–1, 20 mm s–1, 40 mm s–1 and 75 mm s–1. The curves shown are freehand fits representing a simple, self-consistent set of functions which pass between the error bars on the points. Although the uncertainties calculated for each point using the method given above were not identical, they were found to be similar in magnitude, and a single representative error bar has been used in the figure for clarity.
View larger version (17K):
[in this window]
[in a new window]
|
Figure 3. The experimentally derived optimum persistence time constant as a function of disk diameter for speeds of (a) 6.25 mm s–1, (b) 12.5 mm s–1, (c) 20 mm s–1, (d) 40 mm s–1 and (e) 75 mm s–1. The illustrative error bar represents an average standard error estimate for each point.
|
|
|
Discussion
|
|---|
The influence of quantum noise would be expected to reduce as the amount of temporal frame averaging, and hence the number of X-ray photons contributing to the image, is increased. For quantum noise alone, the relative noise level might be expected to reduce approximately as the reciprocal of the square root of the temporal averaging time constant. As can be seen for the zero speed curve in Figure 1
, this simple model does not fit the measured results at short time constants, where the threshold contrast, and by implication relative noise level does not drop as quickly as predicted. At short time constants, the effect of the persistence of vision of the observer dominates the frame averaging being provided by the image intensifier imaging chain. This has been described previously in relation to digital fluoroscopy [7].
At long time constants, when the temporally varying quantum noise is reduced sufficiently by frame averaging, the influence of other (static) noise sources becomes dominant over the quantum noise. In the case of image intensifier fluoroscopy, the predominant additional noise source is the fixed pattern of input phosphor structure mottle [3]. When attempting to score the moving test object at long persistence in this experiment, the input phosphor structure mottle was clearly visible (more so than with the static observations), giving the appearance of a "ground glass" screen over which the test details were moving. This source of noise has not been considered in previous simulations of movement in fluoroscopy [1, 2] but is a feature of real systems. The amplitude of this source of static noise would be expected to be reduced at high speeds of test detail movement by the persistence in the visual system of the observer, but it is likely that this advantageous effect is small compared with the increasing difficulty of performing the threshold contrast detection task while tracking a moving object.
The results for the moving test details overall show the expected systematic increase in threshold contrast with speed as the signal amplitude reduces due to smearing it over a larger area of the image, and the difficulty of the observation task increases. For the moving test details, Figure 1 shows an initial lowering of the threshold contrast with time constant, indicating improved signal-to-noise ratio, followed by a rise as the temporal averaging serves to reduce the signal amplitude. For the smallest disk diameters, this rise was more rapid and no disks could be seen at high speeds and long persistence.
Figure 3 shows the optimum persistence time constant as a function of disk diameter and speed. This can be used in conjunction with Table 1 to give a guide to the best temporal averaging selection where the approximate size and speed of the signal is known, e.g. moving catheter tips. The limited published information on organ movement speeds has been summarized by Guibelalde et al [4]. From this, a broad average movement speed in the abdomen (combining results for diaphragm, liver and kidney) can be taken as 10 mm s–1. From Figure 3, the optimum persistence at 12.5 mm s–1 is constant at a value of 0.15 s for object diameters greater than 4 mm. From Table 1 this persistence corresponds to W=4, or AVG2 and NR4 in the Philips and GE terminology, respectively. For cardiac motion, the maximum speed studied, 75 mms–1 is probably the best guide. At this speed, no additional frame averaging (i.e. just the averaging provided by the observer's visual system, W=1) is optimal for small object diameters.
The results presented relate to an intensifier input dose-rate of 0.5 µGys–1. Changing this dose-rate would alter the balance between the quantum and other noise sources, resulting in a change in the magnitude of noise reduction obtained, but the optimum persistence time would be expected to be similar at other dose-rates. The test system comprised an image intensifier with a CsI input phosphor and a TV display system with variable (and calibrated) persistence. The overall contrast-detail performance of the system was consistent with that routinely found in quality assurance tests. Since the phosphor type, relative level of quantum noise, and persistence time are the main features defining the experiment, it would therefore be expected that the optimization trends reported here would be similar for all image intensifier-based digital fluoroscopy systems.
|
Conclusions
|
|---|
The optimum settings for variable temporal averaging in digital fluoroscopy have been investigated for moving low contrast test details. The results presented include all noise sources within the imaging system plus the effects of the object movement on the perception and decision threshold of the observer. By comparison with the limited information available in the literature on organ movement, it is tentatively concluded that for the organ movement speeds expected in the abdomen, the optimum imaging system persistence time constant should be approximately 0.15 s. For the much greater speeds associated with cardiac motion no additional frame averaging, i.e. just the persistence provided by the observer's visual system, appears to be optimal for small objects.
|
Acknowledgments
|
|---|
We wish to thank Dr Nick Marshall for carrying out the persistence measurements on the experimental system.
|
Footnotes
|
|---|
This work was partially supported by the European Union's radiation protection programme (DIMOND III), contract number FIGM-CT-2000-00061. 
Received for publication November 18, 2003.
Revision received February 18, 2004.
Accepted for publication April 23, 2004.
|
References
|
|---|
- Wilson DL, Jabri KN. Perception of temporally filtered x-ray fluoroscopy images. IEEE Trans Med Imaging 1999;18:22–31.[CrossRef][Medline]
- Xue P, Wilson DL. Effects of motion blurring in x-ray fluoroscopy. Med Phys 1998;25:587–99.[Medline]
- Marshall NW, Kotre CJ. Measurement and correction of the effects of lag on contrast-detail test results in fluoroscopy. Phys Med Biol 2002;47:947–60.[Medline]
- Guibelalde E, Vano E, Fernandez JM, Molinero A, Alberdi J, Merillas A. Design of a PC controlled test device for the study of patient motion in X-ray radiology: first applications and results. Br J Radiol 1998;71:1185–91.[Abstract]
- Guibelalde E, Vano E, Kotre CJ, Faulkner K, Fernandez JM, Ten JI, et al. The use of dynamic phantoms in interventional radiology. Rad Prot Dos 2001;94:155–9.
- Hay GA, Clarke OF, Coleman NJ, Cowen AR. A set of X-ray test objects for quality control in television fluoroscopy. Br J Radiol 1985;58:335–44.[Abstract]
- Kotre CJ, Marshall NW, Faulkner K. Contrast-detail testing techniques for modern x-ray image intensifier systems. Rad Prot Dos 1995;57:245–7.
- Kruger RA. A method for time domain filtering using computerized fluoroscopy. Med Phys 1981;8:466–70.[Medline]
- Marshall NW, Faulkner K, Kotre CJ, Robson K. Analysis of variations in contrast-detail measurements performed on image intensifier-television systems. Phys Med Biol 1992;37:2297–302.[CrossRef]
Top
Abstract
Introduction
