Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

DOI: 10.1148/radiol.2262011812
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Xu, T.
Right arrow Articles by Molloi, S.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Xu, T.
Right arrow Articles by Molloi, S.
(Radiology 2003;226:585-592.)
© RSNA, 2003

Technical Developments

Patient-specific Region-of-Interest Fluoroscopy Device for X-ray Dose Reduction1

Tong Xu, PhD, Huy Quang Le, BS and Sabee Molloi, PhD

1 From the Department of Radiological Sciences, University of California, Medical Sciences I, B-140, Irvine, CA 92697. Received November 12, 2001; revision requested January 2, 2002; final revision received May 2; accepted May 22. Supported in part by grant R01 HL57338 awarded by the National Heart, Lung, and Blood Institute and the United States Department of Health and Human Services. Address correspondence to S.M. (e-mail: symolloi@uci.edu).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 Materials and Methods
Results
Discussion
Appendix
REFERENCES
 
A region-of-interest (ROI) fluoroscopy device that provides an automatically generated ROI filter with an arbitrary shape, as well as digitally compensated images, was built and evaluated. ROI filters were generated by using a deformable attenuation material. Images were compensated by using a compensation ratio and a running average interpolation method. Image compensation parameters were predicted on the basis of the x-ray tube potential used. The image quality with and without an ROI filter was evaluated. This ROI fluoroscopic technique was shown to substantially reduce patient and operator radiation exposure without degrading image quality within the ROI.

© RSNA, 2003

Index terms: Fluoroscopy, technology • Images, processing • Phantoms • Radiations, exposure to patients and personnel • Radiography, technology • Test objects


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
Materials and Methods
Results
Discussion
Appendix
REFERENCES
 
Patients and operators may receive large doses of radiation during diagnostic and interventional x-ray fluoroscopic procedures. In fact, researchers at the National Council on Radiation Protection and Measurements have estimated that one-half of the radiation dose to the United States population resulting from radiologic procedures resulted from x-ray fluoroscopy (1). Complex and time-consuming procedures such as angioplasty and the placement of biliary stents contribute to the high dosage. Many methods have been investigated to reduce patient and operator radiation exposures, including increasing the video camera aperture size (2–4), removing the x-ray grid (2), and using k-edge beam filtration (5,6), pulse fluoroscopy, and/or last-frame hold (7–9). However, these methods are limited by increased image noise, loss of image contrast, increased x-ray tube loading, and image lag.

Region-of-interest (ROI) fluoroscopy has previously been proposed as a method to reduce patient and operator x-ray dose without degrading image quality in the region of the image that is of the most interest (10–15). This technique is based on the assumption that image quality in the periphery of the ROI is relatively unimportant in terms of its contribution to the diagnostic process. At the expense of information about the periphery of the ROI, incident photons on the outside of the ROI are attenuated before reaching the patient. Inside the ROI, the x-ray beam is kept at high intensity. This difference in radiation intensity results in good image quality inside the ROI, but decreased image quality and reduced image gray level outside the ROI. The reduced gray level in the image periphery can be compensated for by using image processing techniques.

Previously reported (10–12,14,16) ROI fluoroscopy techniques differ in the mechanism of x-ray modulation and image compensation. For static modulation, different filtering materials such as brass (10,11) and screens containing gadolinium (12) have been used for x-ray beam attenuation. Optimization of the filtering material improves beam quality after modulation and consequently improves image quality in the periphery (16). For dynamic modulation, a moving segment ROI attenuator made of lead has been used (14); this eliminates the beam-hardening problem resulting from use of static filters.

Two primary methods are used for image compensation to correct for the reduced gray level in the periphery of the image. The mask image subtraction technique requires that an image of the mask be acquired within a uniform field with an ROI filter in the beam; this image is then subtracted from the raw ROI images (10,15). In another image compensation method, arithmetic calculations are performed for each pixel by multiplying the value of the pixel by a certain factor (11) or by passing the peripheral pixels through a lookup table (12). The mask image subtraction method does not require complex image processing techniques such as ROI edge finding, which results in edge artifacts at the ROI boundary. However, it does require that additional mask images be acquired for different beam energies.

In the arithmetic method, the image processor must identify the edge of the ROI by using an edge detection algorithm (11,15) or by thresholding a mask image from a uniform thickness phantom (12). These techniques also require correction for artifacts at the transition zone between the ROI and the periphery. A common limitation of the above ROI fluoroscopy methods is that the shape and size of the ROI are fixed. To vary the size and position of the ROI, Fletcher et al (15) introduced a ROI fluoroscopy system that provides the operator with a series of filters with different ROI sizes. In this technique, the operator can choose the size and position of the ROI. However, more than 100 mask images must be acquired and stored to compensate for the use of filters with ROIs of different sizes, positions, and x-ray beam energies. Furthermore, only a circular ROI is used in the ROI edge detection technique.

In this article, we introduce an ROI fluoroscopy technique that enables the use of ROIs of arbitrary shape and position by enabling the generation of patient-specific attenuating filters automatically and in real time (17,18). The purpose of our study was to evaluate various parameters of this technique.


   Materials and Methods
TOP
 ABSTRACT
 INTRODUCTION
 Materials and Methods
Results
Discussion
Appendix
REFERENCES
 
Overview of the System
The imaging system consists of an x-ray tube, an image intensifier, a charge-coupled device camera, an analog-to-digital conversion interface, a computer, a filter generation device, and a filter composed of an attenuating material. The images were acquired by using a conventional x-ray tube (Dynamax 79-45/120; Machlett Laboratories, Stamford, Conn), a constant x-ray generator (Optimus M200; Philips Medical Systems, Shelton, Conn), a cesium iodide image intensifier with 15- and 23-cm modes, a focused grid (8:1 grid ratio, 36 lines per centimeter), and a charge-coupled device camera (Multicam MC-1134GN; Texas Instruments, Dallas, Tex).

A Matrox Pulsar frame grabber card (Matrox Electronics Systems, Dorval, Quebec, Canada) and a personal computer (Pentium III processor; Compaq, Houston, Tex) were used to digitize the video signal linearly to 640 x 480 x 8-bit precision. The image intensifier mode was set to 23 cm. The images were corrected for pincushion distortion (19). The x-ray tube was set to large (1.2-mm nominal) focal spot. All images in this study were acquired in radiographic mode so that we could have manual control over x-ray tube potential (ie, kilovolt peak) and current (ie, milliamperes).

This ROI fluoroscopy technique can be divided into four essential steps: acquisition of a preexposure image, determination of the shape and location of the ROI by the operator, filter generation, and image compensation. Figure 1 shows a block diagram of this system. The first step is the acquisition of an image of a phantom without the filter in the x-ray beam. After the image has been digitized, the computer displays it, enabling the operator to determine the ROI. With graphic manipulation software, the operator chooses the shape and location of the ROI by drawing an outline on the image. The computer processes this information and translates it to the shape of the filter (Fig 2). The filter generation device, which consists of a 16 x 16 piston array driven by step motors, automatically shapes a deformable attenuation material into the required ROI filter. The attenuating filter is then placed in the x-ray beam, 14 cm from the focal spot. Details of the filter generation procedure are provided in the Appendix.



View larger version (23K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1. Flowchart of the system used for ROI fluoroscopy shows its three components: the x-ray and image acquisition system, the ROI fluoroscopy system, and the user interface.

 


View larger version (41K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2. Left graph shows an ROI as an operator would draw it. The image is segmented into a 16 x 16 matrix of identical squares. Right graph shows the ROI defined by the operator as translated by the computer to the shape of the filter. The straight line, representing the shape of the filter as created by the computer, approximates the dotted line, which represents the ROI hand-drawn by the operator. Each square in the graph corresponds to a filter-forming piston.

 
On images acquired with the ROI filter, there are areas of reduced gray level outside the ROI; this has to be corrected. For linearly acquired images, the relationship between the gray scale values inside and outside the ROI can be expressed by

where GR and Gp are the gray levels within the ROI and the periphery of the image, respectively. CR is the compensation ratio that increases the brightness in the periphery to the same level as that in the ROI. The constant b accounts for the dark current of the charge-coupled device camera, x-ray scatter, and veiling glare. To bring the gray level of the periphery up to that of the ROI, the value b is subtracted from each peripheral pixel and then multiplied by the CR.

The mismatch between the multiplication mask and the raw image causes artifacts in the transition zone. These artifacts can be seen in Figure 3a, which is an image of the humanoid chest phantom described below. Such artifacts were corrected by using a running average interpolation technique to generate a transition zone correction mask. Figure 4 illustrates the compensation procedure. Details of the running average interpolation algorithm are provided in the Appendix.



View larger version (146K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3. (a) A compensated anteroposterior radiograph without transition zone correction obtained in the "midthorax" of a humanoid chest phantom. Artifacts (arrows) are visible at the edge of the ROI. (b) A diagrammatic view of an ROI-periphery transition zone in an image demonstrates the running average interpolation algorithm. x Is the pixel for which the value is to be corrected; the line connecting a and b is the profile across the edge of the ROI; Lb, Lx, and La are the three profiles parallel to the edge of the ROI; and T is a constant number of pixels that can be set by the operator.

 


View larger version (18K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3. (a) A compensated anteroposterior radiograph without transition zone correction obtained in the "midthorax" of a humanoid chest phantom. Artifacts (arrows) are visible at the edge of the ROI. (b) A diagrammatic view of an ROI-periphery transition zone in an image demonstrates the running average interpolation algorithm. x Is the pixel for which the value is to be corrected; the line connecting a and b is the profile across the edge of the ROI; Lb, Lx, and La are the three profiles parallel to the edge of the ROI; and T is a constant number of pixels that can be set by the operator.

 


View larger version (35K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4. A diagrammatic view of the image compensation process shows the different masks and operations applied to the raw ROI image. The operations shown are convolution ({otimes}), subtraction (-), and multiplication (x). The final compensation mask (MCR) is recalculated every few seconds to account for patient motion. Mb is a convoluted mask with value b in the periphery and zero in the ROI. MTZ is the mask that corrects for the artifacts in the transition zone.

 
Measurement of Compensation Parameters
The CR depends on the thickness of the attenuation material and the kilovolt peak used. To determine the CR, an aluminum step phantom was used to simulate the large range of gray levels in the image. The aluminum step phantom consisted of seven steps with thicknesses of 5–23 mm. Initially, a 4.0- or 5.6-mm-thick filter with a square ROI consisting of 6 x 6 pistons at the center of the field of view was generated. The step phantom was placed over different thicknesses of Lucite (7, 14, 20, and 30 cm), and images were acquired with each step positioned across the ROI edge. The above measurements were repeated for different beam energies in the range of 60–120 kVp. Milliamperage was kept fixed for each beam energy. Average gray levels inside and outside the ROI for different phantom steps were plotted to determine the CR and the constant b.

Transmission and Dose Measurement
The radiation exposure to the patient resulting from use of this technique was measured in milliroentgens by using a 6-cm3 ionization chamber (model 20X6-6; Radical, Monrovia, Calif). The exposure was measured by calculating the average exposure over 10 frames. The x-ray transmission of the ROI filter was measured by placing the ionization chamber at the center of the x-ray beam, beneath the image intensifier. Transmission measurements were made for 4.0- and 5.6-mm thicknesses of the attenuating material. Each of the above measurements was repeated with 65-, 80-, and 100-kVp beams.

The entrance exposure to the patient resulting from use of this technique was measured by placing the ionization chamber 57 cm away from the x-ray tube focal spot and 30 cm away from the image intensifier. To avoid x-ray backscatter, no humanoid chest phantom was used. Measurements were recorded in the center of the ROI and at four points outside the ROI (top, bottom, right, and left). Two 6 x 6-piston-square ROI filters with thicknesses of 4.0 and 5.6 mm were used for measurements. The measurements were repeated with 65-, 80-, and 100-kVp beams. For each beam energy, exposure measurements were compared by using different filter types. The x-ray tube current was adjusted to maintain the same maximum gray level for the images.

Exposure to the operator was measured in microroentgens with the humanoid phantom in the beam by placing the chamber at the same height as that used to measure radiation exposure to the patient, but at a 0.5-m distance from the center of the x-ray beam. Operator exposure was averaged from four sample positions along this circle of 0.5 m in radius.

Contrast-to-Noise Ratio Measurement
A contrast phantom (Fluoro-Test Tool Model 07-645; Victoreen, Carle Place, NY) (20) was used to investigate the effect of the attenuating filter on contrast and noise in the image. The contrast phantom consisted of a 6.1-mm-thick aluminum plate with 11-mm–diameter holes of various depths. The measurements were made by using the 5% contrast hole. The contrast phantom was placed over a uniform 14-cm Lucite phantom. Measurements of contrast percentage were made by using 4.0- and 5.6-mm ROI filters. Contrast measurements were also made without the ROI filter in the x-ray beam. The contrast (C) was calculated by using the following equation (21):

where G0 is the average gray level of the target and Gb is the average gray level of the background. The contrast measurements were averaged across 10 frames. Image noise was determined as the SD of gray level fluctuation in the subtracted image. The result was divided by the square root of 2 to obtain the noise of the unsubtracted images. The contrast-to-noise ratio (CNR) was then calculated with the following equation:

where {sigma}b is the noise in the image.

Scatter and Veiling Glare Measurements
Scatter and veiling glare was measured for images of the 14-cm Lucite phantom that were obtained with and without the filter by using a beam stop technique (22). Two-millimeter-thick disks of lead were used as beam stops. Measurements were made by using beam stops with diameters of 2.5, 3.6, and 5.0 mm. The beam stops were placed underneath the Lucite phantom. The measured scatter glare fraction was plotted with respect to beam stop size. The scatter and veiling glare fraction, which represents the contribution of scatter and veiling glare to the intensity of the image, was obtained by extrapolating the plotted graph to the beam stop size of zero. The measurements performed with the filter and those performed without the filter were kept consistent by means of adjusting the x-ray tube current so that the same maximum gray level was obtained in both cases.

General Image Quality
A humanoid chest phantom (Radiology Support Devices, Long Beach, Calif) was used to evaluate the image compensation technique. A catheter and an angioplasty wire were placed over the chest phantom. An irregular ROI filter of 5.6-mm thickness in the periphery was created for the area around the catheter. Images with and without the ROI filter were acquired at 65 kVp. To evaluate image quality, the contrast of the catheter, the contrast of the angioplasty wire, and the noise on the periphery were compared between ordinary fluoroscopy images and compensated ROI fluoroscopy images.

Statistical Analysis
The determinations of compensation parameters (CR and b) and the scatter and veiling glare fraction require linear regression analysis. A statistical software package (SPSS version 11.0; SPSS, Chicago, Ill) was used for this purpose. The compensation parameters and their standard errors were obtained. A quadratic curve regression was used to predict the relationship between the CR and tube potential. Linear regression was also used to predict the intercept b. The correlation coefficients (r) were used as the benchmark of the predictions. All regressions were performed with floating intercept. A P value of less than .05 was considered to represent a statistically significant difference.


   Results
TOP
 ABSTRACT
 INTRODUCTION
 Materials and Methods
Results
Discussion
Appendix
REFERENCES
 
CR Measurements
Figure 5 shows a plot of the average gray level inside the ROI and in the periphery. For each combination of filter type and x-ray tube parameters, linear regression was used to determine the slope and the intercept, which correspond to the CR and the constant b, respectively. Table 1 shows the measured CRs and intercepts (b) for different thicknesses of Lucite. The CR varied within 8% for the Lucite thickness range of 7–30 cm. The intercept b also changed with the Lucite thickness. With transition zone interpolation, these changes do not result in noticeable artifacts at the edge of the ROI. However, owing to the transmission changes of the filter material for different beam energies, the CR showed substantial changes when different kilovolt peaks were used. Figure 6 shows the changes in CR and the intercept b with respect to kilovolt peak for ROI filter thicknesses of 4.0 and 5.6 mm. Figure 6 shows that use of a quadratic regression and a linear regression can result in good prediction of the CR (r = 0.99, P < .001) and the intercept b (r = 0.98, P < .001), respectively.



View larger version (23K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 5. Scatterplot shows relationship between the average gray level inside the ROI and that in the periphery. The data points represent the measurements obtained with different steps of the step phantom. Linear regression (r = 1.0, P < .001), which is represented by the solid line, yielded the CR and the intercept b.

 

View this table:
[in this window]
[in a new window]

  		TABLE 1. CRs and Intercepts (b) for Different Thicknesses of Lucite when a 65-kVp Beam Is Used

 


View larger version (39K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 6a. Scatterplots show the results of plotting (a) CR and (b) the compensation intercept b against x-ray tube potential for the 5.6-mm ({circ}) and 4.0-mm ({blacksquare}) filters. The CRs were fitted by using a binomial function. Linear regression analyses were applied to the compensation intercept data. The solid and dashed lines represent the fitted results. These scatterplots show that use of a quadratic regression and a linear regression can provide good prediction of the CR (r = 0.99, P < .001) and the intercept b (r = 0.98, P < .001), respectively.

 


View larger version (42K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 6b. Scatterplots show the results of plotting (a) CR and (b) the compensation intercept b against x-ray tube potential for the 5.6-mm ({circ}) and 4.0-mm ({blacksquare}) filters. The CRs were fitted by using a binomial function. Linear regression analyses were applied to the compensation intercept data. The solid and dashed lines represent the fitted results. These scatterplots show that use of a quadratic regression and a linear regression can provide good prediction of the CR (r = 0.99, P < .001) and the intercept b (r = 0.98, P < .001), respectively.

 
Transmission and Dose Measurements
Transmission measurements for the uniform thickness filters are shown in Table 2. The measurements show that use of a 5.6-mm-thick filter in a 65-kVp beam, compared with the use of no filter in the beam, resulted in attenuation of the x-ray photons by 86%.


View this table:
[in this window]
[in a new window]

  		TABLE 2. Transmission Reduction for Different Filter Thicknesses Relative to Value Measured without a Filter in Beam

 
The changes in patient radiation exposure inside and outside the ROI resulting from the use of various filter types and x-ray tube potentials are summarized in Table 3. With a 5.6-mm-thick ROI filter, the exposure to the patient in the periphery of the ROI was reduced by 84%, 81%, and 78% for 65-, 80-, and 100-kVp beams, respectively. There was also a slight reduction in exposure inside the ROI. The reduction in exposure inside the ROI was possibly caused by reduction of x-ray scatter in the ROI with the filter in the beam.


View this table:
[in this window]
[in a new window]

  		TABLE 3. Patient Exposure Reductions with Use of 4.0-mm and 5.6-mm Filters

 
Table 4 shows the results of the operator exposure measurements. Operator exposure reduction ranged from 47% to 68% for the various filters and x-ray tube potentials. Reductions observed with use of the 5.6-mm filter were 8%–18% higher than those observed with use of the 4.0-mm filter.


View this table:
[in this window]
[in a new window]

  		TABLE 4. Operator Exposure Reduction as Measured 0.5 m from Center of Beam

 
Contrast-to-Noise Ratio
The contrast improvement factor for a target inside the ROI was measured for use of a 65-kVp beam. The maximum gray level was kept fixed for different filters. The improvement factors were calculated relative to the contrasts measured without a filter in the beam. The measured contrasts of the target were 0.042, 0.051, and 0.052 with no filter, with the 4.0-mm-thick filter, and with the 5.6-mm-thick filter, respectively. These values indicate contrast improvement factors of 21% and 24% with use of the 4.0- and 5.6-mm filters, respectively.

The measured contrast-to-noise ratios inside the ROI were 3.1, 3.4, and 3.7 for images acquired with no filter, images acquired with the 4.0-mm filter, and images acquired with the 5.6-mm filters, respectively. Contrast-to-noise ratio improvements were 10% and 19% for the 4.0- and 5.6-mm filters, respectively.

Scatter and Veiling Glare
Figure 7 shows the scatter and veiling glare fraction for beam stop diameters inside the ROI. Figure 7 indicates that the contributions of scatter and veiling glare to image intensity were 35%, 25%, and 21% with no filter, with the 4.0-mm filter, and with the 5.6-mm filter, respectively.



View larger version (31K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 7. Scatterplot shows results of linear regression analyses of beam stop size versus scatter glare fraction. For different types of ROIs, the linear correlation coefficient r varies from 0.98 to 0.99, and P varies from .10 to .12. The scatter glare fraction was measured without an ROI filter ({triangleup}) and with ROI filters of 4.0 mm ({blacksquare}) and 5.6 mm ({circ}) in peripheral thickness. The extrapolation of the graph to the beam stop size of zero indicates that the contributions of scatter and veiling glare to the intensity of the image were 35%, 25%, and 21% with no ROI filter, with the 4.0-mm ROI filter, and with the 5.6-mm ROI filter, respectively.

 
General Image Quality
Figure 8a shows an image of the humanoid chest phantom that was acquired without a dose-reduction ROI filter. The raw image acquired with a 5.6-mm ROI filter is shown in Figure 8b. The final compensated image is shown in Figure 8c. Figure 8d shows the image of the compensation mask. Figure 8c shows that the contrast within the ROI is improved with use of an ROI filter. For example, the contrast measured in Figure 8c improved with use of a filter versus use of no filter (Fig 8a)—from 0.16 to 0.19 for the catheter and from 0.10 to 0.13 for the angioplasty wire. On the other hand, the relative noise level in the periphery increased from 1.6% to 3.7% with use of the filter. However, the anatomical structures shown in the periphery of Figure 8c are still sufficiently recognizable to provide a reference for the region of clinical interest. Moreover, the ROI transition zone is not noticeable.



View larger version (134K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 8a. Images of a humanoid chest phantom acquired (a) without and (b) with a 5.6-mm-thick filter in the beam. (b) The original ROI fluoroscopy image shows reduced gray level in the periphery and improved contrast inside the ROI. (c) The same image as in b after application of our compensation technique. With this technique, the peripheral gray level was restored and the image quality inside the ROI was preserved. Although there is a higher level of noise in the periphery because of reduced x-ray exposure, the anatomic structures shown in the periphery are still sufficiently recognizable. (d) Image of the compensation mask itself.

 


View larger version (86K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 8b. Images of a humanoid chest phantom acquired (a) without and (b) with a 5.6-mm-thick filter in the beam. (b) The original ROI fluoroscopy image shows reduced gray level in the periphery and improved contrast inside the ROI. (c) The same image as in b after application of our compensation technique. With this technique, the peripheral gray level was restored and the image quality inside the ROI was preserved. Although there is a higher level of noise in the periphery because of reduced x-ray exposure, the anatomic structures shown in the periphery are still sufficiently recognizable. (d) Image of the compensation mask itself.

 


View larger version (145K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 8c. Images of a humanoid chest phantom acquired (a) without and (b) with a 5.6-mm-thick filter in the beam. (b) The original ROI fluoroscopy image shows reduced gray level in the periphery and improved contrast inside the ROI. (c) The same image as in b after application of our compensation technique. With this technique, the peripheral gray level was restored and the image quality inside the ROI was preserved. Although there is a higher level of noise in the periphery because of reduced x-ray exposure, the anatomic structures shown in the periphery are still sufficiently recognizable. (d) Image of the compensation mask itself.

 


View larger version (25K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 8d. Images of a humanoid chest phantom acquired (a) without and (b) with a 5.6-mm-thick filter in the beam. (b) The original ROI fluoroscopy image shows reduced gray level in the periphery and improved contrast inside the ROI. (c) The same image as in b after application of our compensation technique. With this technique, the peripheral gray level was restored and the image quality inside the ROI was preserved. Although there is a higher level of noise in the periphery because of reduced x-ray exposure, the anatomic structures shown in the periphery are still sufficiently recognizable. (d) Image of the compensation mask itself.

 

   Discussion
TOP
 ABSTRACT
 INTRODUCTION
 Materials and Methods
Results
Discussion
Appendix
REFERENCES
 
Image Compensation
The compensation technique requires a priori knowledge of the CR and offset values. The CR value changes for different beam energies. However, it is relatively constant for different patient thicknesses. The relationship between CR and kilovolt peak can be used as a lookup table for generation of the compensation mask. An offset is also needed to correctly adjust the brightness of the periphery of the image to the same level as that within the ROI. This offset is also a function of tube potential. This relationship can also be used in the lookup table for generation of the compensation mask. These relationships were used to successfully compensate images acquired with different beam energies. This technique, which can automatically compensate images acquired with different beam energies, addresses the limitation of previous techniques that required the operator to subjectively choose the CR (11).

A linear relationship between the average gray level inside the ROI and that in the periphery was observed in this study. This linear relationship indicates that the effect of beam hardening with use of the ROI filter is minimal and of no importance. It also ensures that a single CR and offset can be used for an image that contains a large range of attenuation due to the patient’s anatomy.

It is desirable to have a smooth transition in the image between the ROI and the periphery. A running average interpolation technique was used to compensate the edge of the ROI. However, residual artifacts can be detected in cases in which the image contains high-contrast objects across the edge of the ROI. These infrequent edge artifacts do not interfere with image quality within the ROI, which is normally the main focus of the interventional procedure.

Another potential limitation is patient motion during the fluoroscopic procedure. Use of the transition zone correction mask will not be adequate in cases in which patient motion is in the same range as the size of the transition zone. Continued use of the old mask would result in misregistration artifacts along the edge of the ROI. To maintain image quality, a new transition zone correction mask must be generated. The transition zone mask can be updated every few seconds without interrupting the fluoroscopic procedure. Image quality would be degraded for a few seconds immediately after the instance of patient motion. However, image quality would be restored as soon as the new mask was applied to the images.

Dose Reduction
Our results show that use of the ROI filter can reduce patient exposure by up to 85%. With this technique, the operator can choose the ROI filter thickness, which controls the extent of exposure reduction. The trade-off for increased exposure reduction is increased image noise in the periphery. Furthermore, with this technique, the operator has the flexibility to determine the size, shape, and location of the ROI. Optimizing the ROI for each particular procedure can maximize exposure reduction. It is also possible to use different attenuating materials to optimize the spectral shape of the beam (5,6,16).

Our results also show that use of the ROI filter reduced operator exposure by up to 70%. This reduction in exposure is particularly important for complicated interventional procedures requiring long exposure times. This technique is particularly helpful due to the fact that the reduced exposure is not at the cost of degraded image quality in the ROI.

Use of the described ROI fluoroscopy technique can result in a reduction of the x-ray dose to both the patient and the operator because the technique provides an automatically generated ROI filter with an arbitrary shape, as well as digitally compensated images.

Area Beam Equalization
To optimize image quality, in addition to the use of a simple ROI filter, in which there is a fixed thickness of attenuator in the periphery of the ROI, use of the filter fabrication device described in this report can be helpful. Image quality can be optimized by equalizing the beam intensity reaching the image intensifier from within the ROI and from within the periphery (18). This is particularly true in cases in which there is a large dynamic range in the original image. The ROI filter in this situation will also include image-dependent attenuator thickness. Application of area beam equalization technique should further improve image quality in this situation.

Automation
To be clinically applicable, this ROI fluoroscopy technique needs to be automated, so that as little user intervention as possible is necessary. The ROI filter generation process has already been automated. The image processing steps can easily be automated by using currently available real-time image processors. In our current system, compensation mask calculation and filter generation take approximately 15 seconds. This means that normal fluoroscopy, without a ROI filter, is used for the first 15 seconds after ROI selection. Compensation mask calculation and ROI filter generation is performed "in the background" without any operator intervention.

It is also possible to automatically recalculate a new compensation mask if a zoom mode of the image intensifier is chosen. Therefore, it is not necessary to generate a new ROI filter until the interventional procedure requires a differently shaped or positioned ROI. Implementation of techniques to reduce the required time for compensation mask calculation and filter generation is the subject of current research.


   Appendix
TOP
 ABSTRACT
 INTRODUCTION
 Materials and Methods
Results
Discussion
Appendix
REFERENCES
 
Filter Fabrication Device
The filter fabrication device consists of a 16 x 16 matrix of square pistons, an attenuating mold, and a set of stepper motors. Each of the 256 pistons has a dimension of 1.6 x 1.6 mm2. These pistons push into the deformable attenuating material to form the filter. The filter material is composed of a mixture of silicone rubber (Silicone Rubber Packing; Depco, Hauppauge, NY) and cerium dioxide powder. This soft attenuation material is confined within a 3.8 x 3.8 x 1.9-cm3 container. Stepper motors, controlled by the computer, are used to force the pistons into the mold. Sixteen stepper motors control a row of pistons simultaneously. The device is mounted on the C arm beside the x-ray tube, and the filter is positioned in the x-ray beam by a stepper motor. This device is an upgraded version of a system that has been described in detail elsewhere (18).

Filter Fabrication Mechanism
After the operator has drawn an ROI, the image is segmented into 16 x 16 squares in the same area (Fig 2). After piston calibration, each subregion of the image corresponds to a specific piston. A segmented area is considered to be inside the ROI if its center is contained within the ROI. When the square areas approximate the ROI, the shape and location of the ROI is known in terms of which pistons will be pushed into the attenuating material. The attenuating material is initially flattened with a uniform thickness of 8.4 mm. When the pistons push into the attenuating mold, the excess material flows to the periphery of the container. The attenuating mold is solid enough to retain its shape after the pistons have been separated from it. After this separation, the horizontal stepper motor extends the filter by a predetermined distance. The filter is then placed directly in the x-ray beam, 14 cm from the focal spot.

Compensation Procedure
Equation (1) in the body of this article represents the two major steps of the compensation procedure. This procedure is also illustrated in Figure 4. First, a mask image (Mb) with the value of b in the periphery and zero in the ROI is subtracted from the image to be compensated. The resulting image is multiplied with another mask (MCR) with the value of CR in the periphery and 1.0 within the ROI. Due to the focal spot blurring and the finite thickness of the filter, the change in brightness across the edge of the ROI is gradual. The area along the edge of the ROI constitutes the transition zone. A pixel is considered to be inside the transition zone if it is within ± T pixels from the edge of the ROI (ie, 15–20 pixels). The ROI edges of the mask must also have a gradual change to account for the intensity gradient across the ROI transition zone. This is accomplished by blurring Mb and MCR by means of convolution with a chosen kernel. However, the compensated images still contain residual artifacts in the transition zone along the ROI edges. An additional mask (MTZ) is required to correct for these artifacts. As described in the next section, this mask is obtained by using running average interpolation. The MCR and MTZ are then multiplied together to produce the final compensation mask. Convolution and transition zone correction takes about 5 seconds for a 512 x 480 x 8-bit image with a Pentium III 500-MHz computer. However, this procedure will need to be performed only once, at the beginning of the fluoroscopic procedure. After Mb and MCR are calculated, image-processing hardware can be used to easily perform the compensation and display the fluoroscopic images in real time.

Running Average Interpolation
The mismatch between the MCR multiplication mask and the raw image causes artifacts in the transition zone. These artifacts can be corrected by using a transition zone correction mask (MTZ). The mask is generated by using a running average interpolation technique. For each pixel (eg, pixel x in Fig 3) in the transition zone, a profile normal to and centered at the ROI edge is found. At the corner of the ROI the profile is allowed to rotate with the ROI edge. This profile ends at two points (eg, a and b in Fig 3) at the edge of the transition zone. The length of the profile is approximately 2T pixels, which is equal to the width of transition zone. Centered at a, b, and x are three profiles parallel to the ROI edge (eg, La, Lb, and Lx in Fig 3). The length of these profiles is L pixels. The average pixel values of the three parallel profiles are ma, mb, and mx. To produce an image without transition zone artifacts, the mean pixel values should follow the linear relationship

where Dab is the distance between pixels a and b, and is the expected mean value of profile Lx, which is the linear interpolation between ma and mb. The quotient between mx and , rx = /mx, yields the correction ratio of pixel x.

The above procedure is repeated for every pixel in the transition zone to produce MTZ. This correction produces a gradual gradient, in terms of average pixel value, across the edge. This algorithm differs from that of Labbe et al (11) in the method used to calculate ma, mb, and mx. In their technique, the parallel profiles are fixed segments. However, our profiles are moving with the pixel x; this produces the running averages ma, mb, and mx. With fixed segments, artifacts are present at their junctures. On the other hand, the running average approach eliminates these artifacts. For parameters T and L, we used 15 and 50 pixels, respectively. These parameters can be changed to optimize the compensation.


   FOOTNOTES
 
Abbreviations: CR = compensation ratio, ROI = region of interest

Author contributions: Guarantor of integrity of entire study, S.M.; study concepts, T.X., H.Q.L., S.M.; study design, T.X., S.M.; literature research, S.M., T.X.; experimental studies, T.X., H.Q.L., S.M.; data acquisition, T.X., H.Q.L.; data analysis/interpretation, T.X., H.Q.L., S.M.; statistical analysis, T.X.; manuscript preparation, T.X., H.Q.L.; manuscript editing, T.X., H.Q.L., S.M.; manuscript revision/review and final revision approval, S.M., T.X.


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 Materials and Methods
Results
Discussion
Appendix
REFERENCES
 

  1. National Council on Radiation Protection and Measurements. Exposure of the U.S. population from diagnostic medical radiation Bethesda, Md: National Council on Radiation Protection and Measurements, 1989.
  2. Rudin S, Bednarek DR, Miller JA. Dose reduction during fluoroscopic placement of feeding tubes. Radiology 1991; 178:647-651.[Abstract/Free Full Text]
  3. Rossi RP, Wesenberg RL, Hendee WR. A variable aperture fluoroscopic unit for reduced patient exposure. Radiology 1978; 129:799-802.[Abstract]
  4. Kudoh H, Masai N, Ichida T, Kubota T, Nakamura K, Yamada R. Development of variable dose x-ray fluoroscopic system: aimed at reduction of x-ray exposure during interventional radiologic procedure. Radiat Med 1998; 16:229-232.[Medline]
  5. Villagran JE, Hobbs BB, Taylor KW. Reduction of patient exposure by use of heavy elements as radiation filters in diagnostic radiology. Radiology 1978; 127:249-254.[Abstract]
  6. Kohn ML, Gooch AW, Jr, Keller WS. Filters for radiation reduction: a comparison. Radiology 1988; 167:255-257.[Abstract/Free Full Text]
  7. Grollman JH, Jr, Klosterman H, Herman MW, Moler CL, Eber LM, MacAlpin RN. Dose reduction low pulse-rate fluoroscopy. Radiology 1972; 105:293-298.[Medline]
  8. Holmes DR, Jr, Wondrow MA, Gray JE, Vetter RJ, Fellows JL, Julsrud PR. Effect of pulsed progressive fluoroscopy on reduction of radiation dose in the cardiac catheterization laboratory. J Am Coll Cardiol 1990; 15:159-162.[Abstract]
  9. Abdel-Malek A, Yassa F. An adaptive gating approach for x-ray dose reduction during cardiac interventional procedures. IEEE Trans Med Imaging 1994; 13:2-12.[Medline]
  10. Rudin S, Bednarek DR. Region of interest fluoroscopy. Med Phys 1992; 19:1183-1189.[CrossRef][Medline]
  11. Labbe MS, Chiu MY, Rzeszotarski MS, Bani-Hashemi AR, Wilson DL. The x-ray fovea: a device for reducing x-ray dose in fluoroscopy. Med Phys 1994; 21:471-481.[CrossRef][Medline]
  12. Rudin S, Bednarek DR, Yang CY. Real-time equalization of region-of-interest fluoroscopic images using binary masks. Med Phys 1999; 26:1359-1364.[CrossRef][Medline]
  13. Rudin S, Guterman LR, Granger WE, Bednarek DR, Hopkins LN. Application of region-of-interest imaging techniques to neurointerventional radiology. Radiology 1996; 199:870-873.[Abstract/Free Full Text]
  14. Sassi SA, Britten AJ. Moving segments region of interest attenuator for x-ray fluoroscopy. Med Phys 1999; 26:19-26.[CrossRef][Medline]
  15. Fletcher LM, Rudin S, Bednarek DR. Method for image equalization of ROI fluoroscopic images using mask localization, selection and subtraction. Comput Med Imaging Graph 1996; 20:89-103.[CrossRef][Medline]
  16. Massoumzadeh P, Rudin S, Bednarek DR. Filter material selection for region of interest radiologic imaging. Med Phys 1998; 25:161-171.[CrossRef][Medline]
  17. Molloi S, Van Drie A, Wang F. X-ray beam equalization: feasibility and performance of an automated prototype system in a phantom and swine. Radiology 2001; 221:1-8.[Free Full Text]
  18. Molloi S, Tang J, Mather T, Zhou Y. Area x-ray beam equalization for digital angiography. Med Phys 1999; 26:2684-2692.[CrossRef][Medline]
  19. Gronenschild E. The accuracy and reproducibility of a global method to correct for geometric image distortion in the x-ray imaging chain. Med Phys 1997; 24:1875-1888.[CrossRef][Medline]
  20. Wagner AJ, Barnes GT, Wu XZ. Assessing fluoroscopic contrast resolution: a practical and quantitative test tool. Med Phys 1991; 18:894-899.[CrossRef][Medline]
  21. Webb S. The physics of medical imaging Medical Science Series. Bristol, England: Hilger, 1988; 633.
  22. Floyd CE, Jr, Lo JY, Chotas HG, Ravin CE. Quantitative scatter measurement in digital radiography using a photostimulable phosphor imaging system. Med Phys 1991; 18:408-413.[CrossRef][Medline]




This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Xu, T.
Right arrow Articles by Molloi, S.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Xu, T.
Right arrow Articles by Molloi, S.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE