| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Technical Developments |
1 From the Department of Radiology, Founders 202, Massachusetts General Hospital and Harvard Medical School, 32 Fruit St, Boston, MA 02114 (M.K.K., C.W., M.M.M., A.S., E.H., S.S., J.S.), and GE Medical Systems, Waukesha, Wis (G.B.A., K.K., T.L.T.). Supported in part by a grant from GE Medical Systems. Received May 23, 2002; revision requested July 16; final revision received October 11; accepted October 23. Address correspondence to J.S. (e-mail: jshepard@partners.org).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMaterials and Methods Results DiscussionAPPENDIXREFERENCES |
|---|
© RSNA, 2003
Index terms: Computed tomography (CT), radiation exposure • Filters, radiographic • Thorax, CT, 60.12112, 60.12115
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionAPPENDIXREFERENCES |
|---|
Currently, there is an emerging consensus for a reduction of the radiation dose associated with CT scanning. However, modulation of scanning parameters, such as tube current and tube voltage, to decrease radiation exposure is limited by compromised image quality. Thus, the purpose of our study was to assess the effect of noise reduction filters on chest CT images acquired with 50% radiation dose reduction.
|
Materials and Methods |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionAPPENDIXREFERENCES |
|---|
Both sets of these standard- and reduced-dose baseline images were postprocessed with six noise reduction filters (GE Medical Systems, Milwaukee, Wis) to reduce image noise while preserving the qualitative appearance of the noise without perceptible loss of anatomic structure. In comparison with tube attenuation filters, these noise reduction filters represent image processing filter algorithms that are applied to postreconstructed images. According to coding of the manufacturer, these filters were designated as follows: filter A, normal-low; filter B, normal-medium; filter C, normal-high; filter D, special-low; filter E, special-medium; filter F, special-high. The Appendix includes the mechanism of noise reduction provided by the filters.
For each of the four patients, two sets (without and with 50% reduced dose) of four baseline images were generated (n = 2 x 4 x 4 = 32). These baseline images (n = 32) were postprocessed with six noise reduction filters (n = 192). Subsequently, the postprocessed images were combined with the baseline images (n = 192 + 32 = 224) and randomized. Thus, 224 images were evaluated by four experienced chest radiologists (C.W., M.M.M., A.S., J.S.) who were unaware of the image order and the parameters.
Qualitative Analysis
To facilitate blinded evaluation, images presented to the chest radiologists did not include patient demographics and scanning protocol information (ie, details of kilovolt peak, milliampere, and reconstruction algorithm used for scanning). Four subspecialty radiologists (C.W., M.M.M., A.S., J.S.) with expertise in thoracic imaging compared the low–radiation-dose postprocessed images with the baseline standard-dose images obtained in the same patient at a similar level in a side-by-side manner by using a digital picture-archiving and communication system diagnostic workstation (Impax RS 3000 1K; AGFA Technical Imaging Systems, Richfield Park, NJ). All four radiologists independently reviewed the images at a constant window width and window level to simulate both lung window settings (window width, 1,500 HU; window level, -600 HU) and soft-tissue window settings for mediastinum and chest wall (window width, 350 HU; window level, 50 HU). Images were assessed for lung noise, lung contrast, sharpness of central lung vessels and airways, sharpness of peripheral lung vessels and airways (ie, within 2 cm of the parietal pleura), mediastinal noise, mediastinal contrast, mediastinal sharpness, chest wall noise, chest wall contrast, and chest wall sharpness (n = 10 factors). The graininess of the image was the main factor in deciding the degree of noise in lung fields, mediastinum, or chest wall. In comparison with the corresponding baseline images, image noise was graded as better than, equal to, or worse than that on the baseline images. Contrast was scored on the basis of relative ability to discern various anatomic structures with differential densities. Sharpness of the lung vessels, mediastinal structures, and chest wall was assessed on the basis of visually sharp reproduction of these structures on the given image. Hence, 10 factors in 224 images (n = 10 x 224 = 2,240 qualitative factors) were reviewed by four radiologists, which resulted in 8,960 qualitative rating factors (n = 2,240 x 4 = 8,960). Each factor was assessed by using a three-point scale (score 1, worse than that of the corresponding standard-dose CT image; score 2, equal to that of the corresponding standard-dose CT image; score 3, better than that of the corresponding standard-dose CT image). In addition, conspicuity of vascular structures (ie, identical visualization of tiny peripheral vascular structures) in the peripheral 2 cm of the lungs was also evaluated with the same three-point scale. The standard-dose and postprocessed images were identically magnified for this purpose.
In addition, standard-dose images postprocessed with noise reduction filters were also assessed in an identical manner.
Quantitative Analysis
One author (M.K.K.) obtained the quantitative measurements of attenuation values (in Hounsfield units) and image noise (SD of attenuation coefficients, n = 2 x 2 x 224 = 896 measurements) in the descending thoracic aorta and chest wall at a constant position with a region of interest of a constant size (45 square pixels) and shape on all 224 images. Background image noise was also measured on all images (n = 896 + 224 = 1,120) for calculation of the contrast-to-noise ratio (CNR) of the descending thoracic aorta with respect to chest wall muscles for each image. The CNR was calculated by subtracting the attenuation value of the chest wall muscles from the attenuation value of the thoracic aorta.
Background Image Noise
The effect of noise reduction filters on the spatial resolution of images was determined for baseline images and postprocessed images. The modulation transfer function (MTF) was used to mathematically quantify the influence of the various filters on image spatial resolution. The MTF was computed as the angular average of the two-dimensional Fourier transform of the point spread function measured from a CT image of a phantom that comprised a 50-µm tungsten wire centered in a 2-inch hole through a Plexiglas block. The images were acquired with a multi–detector row CT scanner, as noted previously, at 200 mA and 120 kV and were reconstructed at a section thickness of 2.5 mm. In addition, image noise was measured objectively as the SD of the attenuation value from the original phantom CT image, as well as from postprocessed images.
Statistical Analysis
For each subset of baseline and postprocessed images, qualitative image noise, sharpness, and contrast scores for all 10 factors were reported as the mean ± standard error of the mean. Values for qualitative image factors of low–radiation-dose CT images postprocessed with noise reduction filters were compared with those of baseline images acquired at a standard tube current. Individual scores of qualitative image factors were compared by using the Wilcoxon signed rank test and statistical software (SAS/STAT; SAS, Cary, NC). Similarly, objective image noise of low–radiation-dose CT images postprocessed with noise reduction filters was compared with that of standard-dose baseline images. Statistical differences between these two groups were determined by using the paired t test and software (Excel; Microsoft, Redmond, Wash).
The correlation between subjective image noise and quantitative image noise was determined for all four readers by using the Spearman correlation test. Significant correlation was defined as a difference with a two-sided P value of less than .05. The Cohen 
test was used to assess the degree of interobserver agreement between the readers. P values were considered exploratory in nature, and therefore no Bonferroni correction was made (12). The coefficient values for interobserver agreement were considered as follows: slight, with a value less than 0.20; fair, with a value between 0.21 and 0.40; moderate, with a value between 0.41 and 0.60; substantial, with a value between 0.61 and 0.80; or almost perfect, with a value between 0.81 and 1.00 (12).
|
Results |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionAPPENDIXREFERENCES |
|---|
|
|
|
|
|
|
Quantitative Data
Mean objective image noise and CNR for the corresponding subsets of images are presented in Table 2. The objective data showed a statistically significant difference in objective image noise between the standard-dose and low–radiation-dose CT images (P < .05), with standard-dose images being less noisy than were reduced-dose images. A statistically significant reduction in objective image noise was noted with filters C and F (P < .05). Images of the line-wire phantom also demonstrated reduction of image noise with all filters, with respect to the baseline standard-dose CT images (Table 2, Fig 3). In comparison with the baseline standard-dose images, a statistically significant improvement of CNR in the reduced-dose postprocessed CT images was noted with four of the six noise reduction filters (P < .05).
|
|
|
|
test of agreement among the four radiologists was statistically significant (moderate interobserver agreement, simple coefficient, 0.57; two-sided P value, <.05). Additionally we found a significant correlation between the subjective image noise assessment of four readers and the quantitative image noise with the Spearman correlation test (Spearman correlation coefficient, 0.6; P < .05). |
Discussion |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionAPPENDIXREFERENCES |
|---|
Findings in CT radiation dose studies have indicated promising results for a reduction in radiation exposure from a chest CT examination (16–19). The recommended stratagem includes limitation of CT examinations to carefully justified indications, avoidance of needless multiphase protocols, judicious use of repeat or follow-up examinations, and appropriate adjustment of technical scanning parameters on the basis of a patient’s attributes (20,21). Findings of several studies (22–24) about use of chest CT for cancer screening indicated that there was no significant difference in nodule detection at low–radiation-dose chest CT. Mayo et al (24) recommended that a twofold reduction in tube current (ie, 400–140 mAs) and resultant radiation dose do not cause a significant change in subjective image quality or in detection of mediastinal or lung abnormalities with conventional chest CT.
In addition, various technical advances to decrease radiation dose from CT have been developed or are in an experimental stage (25–29). Kachelriess et al (29) investigated the use of multidimensional generalized adaptive filters for CT image noise reduction and reduction in the radiation dose to the patient. They reported a 30%–60% noise reduction in image noise, typically along the direction of the highest attenuation in noncylindrical body regions, such as the shoulder. A novel technique for noise reduction with use of a nonlinear wavelet filter in which the filter thresholds are calculated individually from the "measured" projection data has been described (30). Alvarez and Stonestrom (31) evaluated spatial resolution and noise properties of CT images altered by two-dimensional linear filtering of the initial image. They documented that the use of their filter functions could reduce the noise variance by 17% in comparison with the reduction with conventional filters. Use of nonlinear image processing techniques, in particular smoothing that is based on the understanding of the image, has been reported for creating CT images of good quality by using less radiation (32). These results have shown that newer nonlinear image processing techniques, in particular smoothing that is based on the understanding of the image, may help to create CT images of good quality by using less radiation.
A fundamental objective of this study was to determine whether use of noise reduction filters could improve the image noise with 50% reduction in CT tube current. In addition, we also aimed to assess whether these filters have sufficient promise in facilitating radiation dose reduction to allow their use in CT scanners in the general medical community. To our knowledge, our assessment of 8,960 qualitative and 1,120 objective factors (total factors evaluated, 10,080) of image noise, sharpness, and contrast represents the largest study of low–radiation-dose CT image quality reported in the chest radiology literature. Three of the six filters (filters C, E, and F) analyzed in the present study showed improvement in image noise on low–radiation-dose CT images in comparison with image noise on images acquired at the standard current of 240–280 mA and reconstructed by using standard reconstruction algorithms. There was a significant decrease in image sharpness and contrast with the filters that caused the most significant decrease in image noise, that is, filters C and F. We compared the images acquired with a reduced radiation dose with those acquired with a standard radiation dose, and we noted improvement in the image noise with all noise reduction filters. With the exception of filters C, E, and F, there was no significant decrease in image contrast and sharpness with other filters. The quantitative analysis supported subjective data and demonstrated that the MTF of filtered images is not considerably altered from the original image. From such analysis, it is notable that resolution of structured objects is preserved and, in most cases, improved while noise reduction in nonstructured regions is achieved.
Interestingly, although filter F resulted in maximum improvement in image noise, it also caused a decrease in the conspicuity of small vessels in the peripheral lung fields. A possible explanation for this phenomenon may be that this filter may have "filtered out" the pixels representing these small vessels, because to the filter they may have represented noise. Alternatively, a decrease in image contrast and sharpness associated with this particular filter may have been responsible for decreased conspicuity. Regardless, this observation raises concerns about the possibility of nonvisualization of subtle or low-contrast lesions with filter F. Overall, the readers preferred filters C and F to view the soft tissues, because these filters caused maximum reduction of noise. Findings in the present study suggest that filters C and F were most effective in reducing mediastinal and chest wall noise and could be useful in "very noisy" images. For less noisy images, filters B and E may be used for reducing noise with less compromise in sharpness and contrast. The use of filters C and F resulted in reduced conspicuity of small peripheral vessels; thus, their use in general practice may result in failure to appreciate small nodules. Further clinical trials with lesion detection will be essential to address this issue and to establish the validity and actual application of noise reduction filters for radiation reduction.
There were several limitations in our study. The issue of potential compromise in diagnostic sensitivity of postprocessed CT images acquired with a reduced radiation dose for detection and characterization of lesions was not addressed. However, as an initial pilot study, our objective was to obtain preliminary data about the effectiveness of noise reduction filters in reducing image noise in images obtained with substantial radiation dose reduction and to ascertain the effect of the filters on image sharpness and contrast with low–radiation-dose CT. An important limitation of our study was that a small number of patients were included, and there existed a consequent interdependency of data and statistical analysis, which resulted from a large number of images being obtained from a small patient cohort. However, this interdependence was accounted for by the use of the Wilcoxon signed rank test (ie, the nonparametric equivalent of the paired t test), which explicitly incorporates the interdependence of the test values to obtain a more powerful test. As noted in our methods, the interdependence of the test results in which the low–radiation-dose postprocessed filtered images were compared with the baseline images was not corrected with Bonferroni adjustment, because the P value was exploratory. A small study size is an issue when the study results are "negative," because they may reflect inadequate power, due to too small a sample size. However, a small sample size does not make positive results more likely. On the contrary, larger differences are required for a positive result in a small study than are required for a positive result in a large study. Radiation safety concerns and consequent procedural difficulties in obtaining consent for acquisition of extra images, along with elaborate labor-intensive postprocessing with noise reduction filters, contributed substantially to the small patient numbers, and we could not fully explore the entire range of biovariability of the entire patient population.
Objective noise and CNR data obtained on all images, as well as noise and spatial resolution (ie, MTF) obtained with the wire phantom, supported the qualitative data. The qualitative improvement of image noise with two of the six filters (filters C and F) correlated with the objective measurement of noise on the images obtained in the patients and on the images of the line-wire phantom that also demonstrated maximum reduction of image noise with filters C and F.
Our initial findings with reduced-dose CT images processed with noise reduction filters are encouraging. Noise reduction filters successfully reduced image noise on low–radiation-dose chest CT images, with some compromise in image sharpness and contrast assessed qualitatively, and slightly altered MTF at higher spatial frequencies. Furthermore, these results can be confirmed with those of studies for lesion detection in a larger cohort of patients with a wider modulation of tube current when noise reduction filters become commercially available for clinical and research purposes.
In conclusion, postprocessing of reduced-dose CT images with noise reduction filters resulted in improved qualitative image noise in the lung fields, mediastinum, and chest wall with all six filters. All the reduced-dose postprocessed images showed better CNR and signal-to-noise ratio in comparison with those of the baseline low–radiation-dose CT images. However, there was some compromise in image sharpness and contrast.
|
APPENDIX |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionAPPENDIXREFERENCES |
|---|
On an image, a group of structural pixels representative of structures of interest and a group of nonstructural pixels representative of nonstructural regions on the image are present. The structural pixels are identified by determining the gradient values for each pixel and by identifying the pixels that have a desired relationship to the gradient threshold value. In the present study, the noise reduction filter technique involves isotropic filtering of nonstructured regions with a low-pass filter and directional filtering of the structured regions with a smoothing filter, which operates parallel to the edges and with an enhancing filter that operates perpendicular to the edges. A blending parameter regulates the recombination of the structured and nonstructured segments (Fig A1). The six filters (filters A-F) analyzed in this study were designed to achieve varying levels of segmentation, blending, and sharpening to provide a range of varying visual effects in noise reduction and structure enhancement. These may be separated into two groups: Filters A, B, C are included in one group and filters D, E, F are included in the other. The former filters apply 33% more aggressive sharpening to the smoothed-structure pixels than do the latter filters. In general, the three filters in each group were designed to incorporate different levels of smoothing. Filters A and D are parametrically twice smoother than are filters B and E and four times smoother than are filters C and F. Filters A and D are the least aggressive in terms of level of filtering, whereas filters C and F are the most aggressive.
|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Author contributions: Guarantors of integrity of entire study, J.S., S.S.; study concepts, M.K.K., T.L.T., G.B.A., C.W.; study design, M.K.K., C.W.; literature research, M.K.K., G.B.A.; clinical studies, M.K.K., J.S., C.W., M.M.M., A.S.; experimental studies, K.K., G.B.A., T.L.T.; data acquisition, M.K.K.; data analysis/interpretation, E.H., M.K.K., M.M.M.; statistical analysis, E.H., M.K.K., M.M.M.; manuscript preparation, M.K.K., M.M.M., C.W., G.B.A., K.K., S.S.; manuscript definition of intellectual content, M.K.K., T.L.T.; manuscript editing, M.K.K., M.M.M., C.W., G.B.A., K.K.; manuscript revision/review and final version approval, M.K.K., M.M.M., C.W., K.K.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionAPPENDIXREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  T. Kubo, P.-J. P. Lin, W. Stiller, M. Takahashi, H.-U. Kauczor, Y. Ohno, and H. Hatabu Radiation Dose Reduction in Chest CT: A Review Am. J. Roentgenol., February 1, 2008; 190(2): 335 - 343. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K Imai, M Ikeda, S Wada, Y Enchi, and T Niimi Analysis of streak artefacts on CT images using statistics of extremes Br. J. Radiol., November 1, 2007; 80(959): 911 - 918. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A.B. Smith, W.P. Dillon, R. Gould, and M. Wintermark Radiation Dose-Reduction Strategies for Neuroradiology CT Protocols AJNR Am. J. Neuroradiol., October 1, 2007; 28(9): 1628 - 1632. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Y. Nakayama, K. Awai, Y. Funama, M. Hatemura, M. Imuta, T. Nakaura, D. Ryu, S. Morishita, S. Sultana, N. Sato, et al. Abdominal CT with Low Tube Voltage: Preliminary Observations about Radiation Dose, Contrast Enhancement, Image Quality, and Noise Radiology, December 1, 2005; 237(3): 945 - 951. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M A Lewis and S Edyvean Patient dose reduction in CT Br. J. Radiol., October 1, 2005; 78(934): 880 - 883. [Full Text] [PDF]
|
|
|
|
|
|
M. K. Kalra, S. Rizzo, M. M. Maher, E. F. Halpern, T. L. Toth, J.-A. O. Shepard, and S. L. Aquino Chest CT Performed with Z-Axis Modulation: Scanning Protocol and Radiation Dose Radiology, October 1, 2005; 237(1): 303 - 308. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. M. R. Rizzo, M. K. Kalra, B. Schmidt, R. Raupach, M. M. Maher, M. A. Blake, and S. Saini CT Images of Abdomen and Pelvis: Effect of Nonlinear Three-dimensional Optimized Reconstruction Algorithm on Image Quality and Lesion Characteristics Radiology, October 1, 2005; 237(1): 309 - 315. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. K. Kalra, M. M. Maher, R. S. Kamath, T. Horiuchi, T. L. Toth, E. F. Halpern, and S. Saini Sixteen-Detector Row CT of Abdomen and Pelvis: Study for Optimization of Z-Axis Modulation Technique Performed in 153 Patients Radiology, October 1, 2004; 233(1): 241 - 249. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. K. Kalra, M. M. Maher, M. A. Blake, B. C. Lucey, K. Karau, T. L. Toth, G. Avinash, E. F. Halpern, and S. Saini Detection and Characterization of Lesions on Low-Radiation-Dose Abdominal CT Images Postprocessed with Noise Reduction Filters Radiology, September 1, 2004; 232(3): 791 - 797. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. K. Kalra, M. M. Maher, T. L. Toth, R. S. Kamath, E. F. Halpern, and S. Saini Radiation from "Extra" Images Acquired with Abdominal and/or Pelvic CT: Effect of Automatic Tube Current Modulation Radiology, August 1, 2004; 232(2): 409 - 414. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. D. Cody, D. M. Moxley, K. T. Krugh, J. C. O'Daniel, L. K. Wagner, and F. Eftekhari Strategies for Formulating Appropriate MDCT Techniques When Imaging the Chest, Abdomen, and Pelvis in Pediatric Patients Am. J. Roentgenol., April 1, 2004; 182(4): 849 - 859. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. K. Kalra, M. M. Maher, T. L. Toth, L. M. Hamberg, M. A. Blake, J.-A. Shepard, and S. Saini Strategies for CT Radiation Dose Optimization Radiology, March 1, 2004; 230(3): 619 - 628. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||
) and 50% (
) MTF for original images and images postprocessed with filters A-F. lp/cm = line pairs per centimeter.