| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Technical Developments |
1 From the Department of Radiology, Massachusetts General Hospital, Harvard Medical School, White 270-E, 55 Fruit St, Boston, MA 02114 (M.K.K., M.M.M., D.V.S., M.A.B., P.F.H., E.H., S.S.); and GE Medical Systems, Milwaukee, Wis (G.B.A., T.L.T.). Received June 10, 2002; revision requested August 20; revision received September 5; accepted October 24. Supported in part by a grant from GE Medical Systems. Address correspondence to S.S. (e-mail: ssaini@partners.org).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
© RSNA, 2003
Index terms: Abdomen, CT, 70.12112, 76.12112 • Computed tomography (CT), image processing • Computed tomography (CT), image quality • Computed tomography (CT), radiation exposure • Radiations, exposure to patients and personnel
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
|
Materials and Methods |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
Patients were then rescanned in the equilibrium phase to obtain data for two sets of four additional images centered at the upper pole of right kidney. Scanning parameters for the first image data set, including kilovolt peak and tube current settings, were identical to those used for the diagnostic CT examination in the dynamic phase. For the second set of images, so that we could study the impact of noise reduction filters on images acquired with a substantially low radiation dose, tube current was reduced by 50% (to 120–150 mA) with all of the other scanning parameters kept constant. Weighted CT dose index measurements were obtained from the CT console at the time of scanning of the patients.
Noise Reduction Filter Processing
Both sets of these baseline images were postprocessed with six noise reduction filters (GE Medical Systems) to reduce image noise while preserving the qualitative appearance of the noise without perceptible loss of definition of anatomic structures. This was accomplished by using gradient analysis methods to separate the image into structured and nonstructured regions. A threshold parameter controlled the segmentation process. The nonstructured regions were isotropically filtered with a low-pass filter. The structured regions were directionally filtered with a smoothing filter operating parallel to the edges and with an enhancing filter operating perpendicular to the edges. A blending parameter regulated the recombination of the structured and nonstructured segments (Fig 1). The six filters use different combinations of segmentation and blending. These filters were classified by GE Medical Systems as follows: filter A, normal-low; filter B, normal-medium; filter C, normal-high; filter D, special-low; filter E, special-medium; and filter F, special-high.
|
Image Analysis
To facilitate blinded evaluation, images presented to the radiologists did not include any information regarding patient demographics or the scanning protocol that was used (specifically, the kilovolt peak, milliamperage, and reconstruction algorithm used). Qualitative and quantitative analyses were performed. Three subspecialty radiologists (M.M.M., D.V.S., M.A.B.) with expertise in abdominal imaging independently evaluated randomized images at a workstation (Advantage 3.0 Windows; GE Medical Systems, Waukesha, Wis). For qualitative analysis, images were assessed separately for noise, soft-tissue contrast, quality, and sharpness in the liver, adrenal glands, and pancreas (four parameters assessed at each of three sites, for a total of 12 parameters). Another two image parameters—noise and contrast with respect to adjoining soft tissues—were assessed in the abdominal fat.
Thus, 14 parameters were assessed in 392 images (for a total of 5,488 qualitative parameters) by each of three readers, resulting in a total of 16,464 qualitative parameter ratings. The radiologists assessed each parameter according to their individual perceptions of acceptability by using a five-point scale (1 = not acceptable, 2 = substandard, 3 = acceptable, 4 = above average, 5 = excellent). An overall score of subjective image quality represented the sum of the noise, sharpness, contrast, and image quality scores.
Additionally, quantitative measurements of attenuation values (in Hounsfield units) and image noise (in SDs of attenuation coefficients; 1,568 image noise and attenuation coefficient measurements) were obtained for all 392 images in the right hepatic lobe and the abdominal fat (in the anterior abdominal wall) at a constant position with a region of interest of constant size and shape (drawn by M.K.K. at a workstation). The contrast-to-noise ratio (CNR) of the liver with respect to abdominal fat was calculated for each image with the following formula, in which all attenuation and noise measurements are in Hounsfield units: CNR = (AVL - AVF)/AVB, where AVL is the attenuation value of the liver, AVF is the attenuation value of abdominal fat, and AVB is the background abdominal fat noise.
Statistical Analysis
For each subset of CT images (baseline images and those postprocessed with noise reduction filters), qualitative image quality scores for all 14 parameters were reported as means ± SDs. Qualitative image quality scores for low–radiation-dose CT images postprocessed with noise reduction filters were compared with those of baseline CT images acquired with standard tube current. Individual subjective image quality scores for each parameter were compared by using the Wilcoxon signed rank test (SAS/STAT software; SAS Institute, Cary, NC).
Similarly, the objective image quality (image noise and CNR measurements) of low-dose CT images postprocessed with noise reduction filters was compared with that of baseline CT images obtained with standard tube current. Differences between these two groups were determined by using the Student t test (Excel; Microsoft, Redmond, Wash). For comparison of overall performance of low-dose CT images postprocessed with noise reduction filters in terms of multiple objective parameters (ie, noise in liver and abdominal fat and CNR), a one-way analysis of variance with the Dunnett t test was performed (SAS/STAT software; SAS Institute).
Linear correlation between subjective image noise and contrast scores and quantitative image noise measurements and CNRs was determined for all three readers with the Spearman correlation test. P < .05 was considered to indicate a statistically significant difference. The Cohen 
test was used to assess the degree of agreement between the readers. To decrease the interdependency of the data, we performed Bonferroni correction of the P values to determine the significance of the differences (9). For coefficients, < 0.20 was considered to represent slight interobserver agreement; = 0.21–0.40, fair agreement; = 0.41–0.60, moderate agreement; = 0.61–0.80, substantial agreement; and = 0.81–1.00, almost perfect agreement (9).
|
Results |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
|
|
After postprocessing of the baseline low-dose CT images with noise reduction filters, all three readers noted an overall decrease in image noise and an improvement in overall image quality compared with baseline low-dose CT images (Table 1). Although all six filters yielded reduced image noise, there was a statistically significant improvement in subjective image noise scores with filters C, E, and F (P < .05). The differences in image sharpness and contrast between the baseline low-dose CT images and the postprocessed low-dose CT images were not statistically significant (P > .05).
With respect to baseline standard-dose CT images, reduction in image noise of low-dose CT images with use of noise reduction filters was noted at subjective evaluation. There were no significant differences in image sharpness, contrast, and overall image quality in the low-dose CT images postprocessed with noise reduction filters when they were compared with the standard-dose CT images (P > .05). There was agreement among all three readers that low-dose CT images postprocessed with filter F showed a significant decrease in image noise in comparison with baseline standard-dose CT images (P < .05). In addition, improvement in subjective image noise scores with use of filter C (according to readers D.V.S. and M.M.M.) and filter E (according to readers M.A.B. and M.M.M.) was also documented when images postprocessed with these filters were compared with baseline standard-dose CT images. With respect to baseline standard-dose CT images, a similar trend of image noise reduction (as assessed subjectively) was noted in standard-dose CT images postprocessed with filters C, E, and F.
Quantitative reduction in image noise was observed on CT images acquired with reduced tube current and postprocessed with all six noise reduction filters when they were compared with baseline reduced-dose CT images. This was statistically significant with three noise reduction filters (filters C, E, and F). Two subsets of low-dose CT images postprocessed with filters C and F showed significant reduction in image noise with respect to corresponding baseline standard-dose CT images (Table 2) (Fig 2). There was no significant difference in objective image quality measurements (image noise in liver and fat and CNR) between baseline images acquired with standard tube current and low-dose CT images postprocessed with noise reduction filters C and E (P > .05) (Table 2). There was no significant change in the CT numbers for liver and abdominal fat following postprocessing of the baseline low-dose CT images with noise reduction filters.
|
|
|
|
test of agreement ( = 0.6, two-sided P < .05). In addition, a statistically significant correlation between the results of subjective image noise assessment by the three readers and quantitative measurements of image noise was noted when the Spearman correlation test was applied (Spearman correlation coefficient = -0.6, two-sided P < .05). The weighted CT dose index at standard-dose and 50%-reduced–dose abdominal CT ranged between 17.8 and 22.2 mSv and 8.4 and 10.5 mSv, respectively. |
Discussion |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
In addition, various technical advances designed to result in a decrease in radiation dose at CT have been developed or are in an experimental stage. These include pre-patient collimation of x-ray beams, online tube current modulation, and postprocessing of images (5,10–12). Toth et al (10) reported radiation dose reductions of up to 40% when a tracking system was used to stabilize and collimate an x-ray beam in a multisection CT scanner. On the basis of results of studies with phantoms and patients, Kalender and colleagues (11) observed that online tube current modulation enables a considerable reduction in radiation dose when noncylindric areas of the body are scanned. Recently, Mastora et al (12) recommended the use of anatomically adapted tube current modulation in CT angiography for thoracic outlet syndrome as a strategy that permits substantial dose reduction. Itoh and colleagues (13) documented a 17% reduction in radiation dose in CT of the chest with use of an aluminum filter.
Kachelriess et al (14) assessed a generalized multidimensional adaptive filtering method that applies nonlinear filters in up to three dimensions of the raw data domain and could reduce image noise by 30%–60% in noncylindric body regions like the shoulders. Iida and colleagues (15) evaluated various shapes of filters used in the convolution integrals of the filtered back-projection procedure. They found that a high-frequency-cut characteristic of the filters significantly reduced the statistical noise. Alvarez and Stonestrom (16) studied the spatial resolution and noise properties of CT images altered by means of two-dimensional linear filtering of the initial image. They observed that such filtering could reduce the noise variance by 17% as compared with conventional filtering. Keselbrener et al (17) evaluated whether nonlinear image processing techniques—in particular, smoothing based on understanding of the image—might result in CT images of good quality being obtained with less radiation. Their results showed that application of nonlinear filters to CT imaging data led to better image quality than did the use of currently available linear filters. They concluded that newer nonlinear image processing techniques—in particular, smoothing—that are based on understanding of the image may result in CT images of good quality being obtained with less radiation.
The major purpose of the noise reduction filters used in the present study was to reduce image noise while preserving the qualitative appearance of the noise without a perceptible loss in definition of anatomic structures. Discrete pixel images are composed of an array or a matrix of pixels that have varying properties and are represented by a signal—typically a digitized value representative of a sensed parameter, such as radiation received within each pixel region. This technique provides a method for identifying salient structures in discrete pixel images that uses gradient data generated for each pixel to automatically determine a gradient threshold that is then used to separate structural features in the resulting image from nonstructural features.
The filtering was accomplished by using gradient analysis methods to separate the image into structured and nonstructured regions. In an image, a group of structural pixels representative of structures of interest and a group of nonstructural pixels representative of nonstructural regions are present. In the filtering technique described in this report, the structural pixels are identified by determining gradient values for each pixel and by identifying pixels having a desired relationship to the gradient threshold value. The gradient threshold value is identified by comparing gradient values for the pixels to a desired value and comparing gradient directions for the pixels to one another. A signal processing circuit is configured to identify structural pixels representative of features of interest and nonstructural pixels on the basis of their gradient values. It also performs orientation smoothing of the structures of interest, homogenization smoothing of the nonstructural regions, orientation sharpening of the structures of interest, and blending of textural data into the nonstructural regions.
In addition, the technique facilitates the separation of small noisy regions from the definition of image structure. Such regions are identified in a computationally efficient manner, and their size may be defined by default values or by values selected by an operator. The resulting structural definition is then further enhanced through smoothing. This study was initiated to determine if the noise reduction filters show sufficient promise for improving image quality at reduced radiation doses that they can be deployed in CT scanners for the benefit of the general medical community.
To our knowledge, the present assessment of subjective and objective image quality, in which 18,032 parameters were evaluated, represents the largest set of low-dose CT image quality data yet reported in the radiology literature. All six noise reduction filters reduced image noise when applied to low-dose CT imaging data. Three of the six filters used in our study yielded a reduction in image noise on low-dose CT images when such images were compared with images that were acquired at the standard tube current of 240–300 mA that is currently being used in CT protocols and were reconstructed with standard reconstruction algorithms. All three readers consistently scored noise on low-dose CT images postprocessed with filter F as being lower than that on CT images obtained with the standard tube current. This finding was substantiated by the results of quantitative measurement of image noise and CNR (P < .05).
We did not notice significant alterations in image contrast and sharpness on 50%-reduced–dose CT images postprocessed with noise reduction filters. However, an edge-enhancement phenomenon that was noted in some of the images postprocessed with filters C and F (Fig 2c, 2d) may potentially adversely affect the perception of subtle lesions in clinical studies. This may be a result of variable visual effects of noise reduction and structure enhancement, because among the six noise reduction filters evaluated in this study, these two filters incorporate the most aggressive level of filtering. In concordance with the results of previous studies that were conducted with phantoms and patients with smaller body sizes (7), our results show that these filters have a potential to result in a reduction of radiation dose of at least 50% from standard levels.
In abdominal magnetic resonance imaging, liver-to-spleen CNR is calculated with respect to background image noise (18–20). However, in abdominal CT imaging, there is no fixed or standard practice for calculating CNR. A determination of optimum sites for measurement of CNR in abdominal CT images would itself involve a major investigation. Hence, in the absence of standard documented guidelines for optimum measurement of CNR, we arbitrarily obtained liver-to-fat CNR in all images. Quantitative abdominal CT image contrast measurements calculated in terms of CNR had good linear correlation with the results of subjective image contrast assessment by the three readers. Ideally, one should measure CNRs between liver parenchyma and liver metastases. However, the dynamic nature of CT scanning for liver metastases makes realistic dual-dose scanning impractical. Because there was no significant change in the CT numbers of liver and abdominal fat following postprocessing of the baseline low-dose CT images with noise reduction filters, we do not anticipate any loss of ability to use the quantitative attenuation values in such postprocessed images.
There were several limitations in our study. We did not evaluate the sensitivity of postprocessed reduced-dose CT images in the detection and characterization of lesions because our objective was to obtain preliminary data on the effect of noise reduction filters on image quality at low-dose CT. Subtle lesions on reduced-dose CT images may be missed due to increased image noise and beam-hardening artifacts. In addition, the CT examinations in this study were performed in the equilibrium phase. However, this should not have affected image interpretation, because the image quality evaluation method for dynamic image data would have been identical. Major procedural difficulties in recruiting the patients and scheduling the protocol contributed to the small number of patients in our study. Consequently, this may have led to interdependency in the statistical results due to the large number of images in a small patient cohort.
However, each of the 392 images was independently evaluated for each parameter by three radiologists who did not know which images were low-dose CT images postprocessed with noise reduction filters. As pilot study results, these findings should be helpful for guiding future studies of lesion detection with reduced-dose CT images processed with noise reduction filters. A further study with a wider modulation of tube current and a larger number of patients with varied anthropometric measurements may be essential before we can substantially reduce the radiation dose while obtaining diagnostic image quality with these promising image filters.
In summary, use of noise reduction filters was found to be effective in reducing noise (as assessed subjectively and objectively) in CT images acquired with a 50% reduction in radiation dose. These filters have the potential to enable a reduction in CT radiation dose without producing appreciable deterioration in image quality.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Author contributions: Guarantor of integrity of entire study, S.S.; study concepts, M.K.K., T.L.T., S.S.; study design, M.K.K., M.A.B., S.S.; literature research, P.F.H., M.K.K., M.M.M.; clinical studies, M.A.B., M.M.M., D.V.S.; experimental studies, T.L.T., G.B.A.; data acquisition, M.K.K., M.M.M.; data analysis/interpretation, M.K.K., M.M.M., E.H.; statistical analysis, E.H., M.K.K., M.M.M.; manuscript preparation, M.K.K., M.M.M., P.F.H.; manuscript definition of intellectual content, M.K.K., T.L.T., G.B.A., P.F.H.; manuscript editing and revision/review, M.K.K., M.M.M., M.A.B., D.V.S., S.S.; manuscript final version approval, M.M.M., M.A.B., D.V.S., S.S.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  D. Marin, R. C. Nelson, E. Samei, E. K. Paulson, L. M. Ho, D. T. Boll, D. M. DeLong, T. T. Yoshizumi, and S. T. Schindera Hypervascular Liver Tumors: Low Tube Voltage, High Tube Current Multidetector CT during Late Hepatic Arterial Phase for Detection--Initial Clinical Experience Radiology, June 1, 2009; 251(3): 771 - 779. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. Kubo, Y. Ohno, S. Gautam, P.-J. P. Lin, H.-U. Kauczor, H. Hatabu, and iLEAD Study Group Use of 3D Adaptive Raw-Data Filter in CT of the Lung: Effect on Radiation Dose Reduction Am. J. Roentgenol., October 1, 2008; 191(4): W167 - W174. [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]
|
|
|
|
 |
|
 A J Van Der Molen, W J H Veldkamp, and J Geleijns 16-slice CT: achievable effective doses of common protocols in comparison with recent CT dose surveys Br. J. Radiol., April 1, 2007; 80(952): 248 - 255. [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]
|
|
|
|
|
|
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]
|
|
|
|
|
|
J. Yee, N. N. Kumar, S. Godara, J. A. Casamina, R. Hom, G. Galdino, P. Dell, and D. Liu Extracolonic Abnormalities Discovered Incidentally at CT Colonography in a Male Population Radiology, August 1, 2005; 236(2): 519 - 526. [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]
|
|
|
|
|
|
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]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
