| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Gastrointestinal Imaging |
1 From the Division of Abdominal Imaging and Intervention, Massachusetts General Hospital and Harvard Medical School, 55 Fruit St, White 270E, Boston, MA 02114 (M.K.K., M.M.M., M.A.B., B.C.L., E.F.H., S.S.); and GE Medical Systems, Waukesha, Wis (K.K., T.L.T., G.A.). Received September 26, 2003; revision requested December 5; revision received December 23; accepted January 30, 2004. 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 RESULTS DISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Low-dose CT images of abdominal lesions in 19 consecutive patients (11 women, eight men; age range, 32–78 years) were obtained at reduced tube currents (120–144 mAs). These baseline low-dose CT images were postprocessed with six noise reduction filters; the resulting postprocessed images were then randomly assorted with baseline images. Three radiologists performed independent evaluation of randomized images for presence, number, margins, attenuation, conspicuity, calcification, and enhancement of lesions, as well as image noise. Side-by-side comparison of baseline images with postprocessed images was performed by using a five-point scale for assessing lesion conspicuity and margins, image noise, beam hardening, and diagnostic acceptability. Quantitative noise and contrast-to-noise ratio were obtained for all liver lesions. Statistical analysis was performed by using the Wilcoxon signed rank test, Student t test, and 
test of agreement.
RESULTS: Significant reduction of noise was observed in images postprocessed with filter F compared with the noise in baseline nonfiltered images (P = .004). Although the number of lesions seen on baseline images and that seen on postprocessed images were identical, lesions were less conspicuous on postprocessed images than on baseline images. A decrease in quantitative image noise and contrast-to-noise ratio for liver lesions was noted with all noise reduction filters. There was good interobserver agreement ( = 0.7).
CONCLUSION: Although the use of currently available noise reduction filters improves image noise and ameliorates beam-hardening artifacts at low-dose CT, such filters are limited by a compromise in lesion conspicuity and appearance in comparison with lesion conspicuity and appearance on baseline low-dose CT images.
© RSNA, 2004
Index terms: Abdomen, CT, 78.12114 • Computed tomography (CT), image processing • Computed tomography (CT), image quality • Computed tomography (CT), radiation exposure, 78.12114 • Radiations, exposure to patients and personnel
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Results of many patient-based studies performed in pursuit of the ALARA, or as low as reasonably achievable, principle for CT radiation dose reduction have shown that the radiation dose from CT scanning can be reduced (4,5). A major concern for reducing radiation dose by adjusting scanning parameters is an increase in the image noise content, which can affect the diagnostic acceptability of images. CT scanner technology has to improve further to increase scanner efficiency and enhance image quality at reduced radiation exposures (6–12). Accordingly, to facilitate acceptance of low-dose CT images in clinical practice, noise reduction filters have been designed to decrease noise in images acquired with low-radiation-dose CT scanning. Results of recent studies have demonstrated the value of noise reduction filters in reducing noise content in images acquired with reduced tube current (6,7). Thus, the purpose of our study was to assess the effect of noise reduction filters on detection and characterization of lesions on low-dose abdominal CT images.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
All examinations were performed with a four-channel multi–detector row CT scanner (LightSpeed QX/I; GE Medical Systems, Waukesha, Wis). A diagnostic abdominal CT examination was performed with oral (Readi-Cat 2; E-Z-Em, Westbury, NY) and intravenous (iopromide, Ultravist; Berlex Laboratories, Wayne, NJ) contrast material in the portal venous phase. Selected image parameters included 140 kVp, 220–280 mA, a detector configuration of 2.5 mm, a beam pitch of 1.5:1, a table speed of 15 mm per gantry rotation (at a 0.8-second gantry rotation time), and the acquisition of 5-mm images reconstructed at 5-mm intervals.
The precontrast or equilibrium phase images were acquired with low radiation dose at reduced tube currents. For the purposes of this study, low-radiation-dose CT images of the abdomen that were acquired at reduced tube currents of 120–144 mAs through the lesions in the precontrast phase (in four patients) or in the equilibrium phase (in 15 patients) were used as baseline image data. The remaining scanning parameters used to obtain these baseline low-dose images were kept identical to those used for dynamic contrast-enhanced scanning.
After being reconstructed with a standard algorithm, the baseline low-radiation-dose images were postprocessed with each of six noise reduction filters (GE Medical Systems, Milwaukee, Wis) to generate the postprocessed low-dose images. The purpose of the noise reduction filters was to manipulate the low-dose CT images to reduce image noise while preserving the qualitative appearance of the noise without a perceptible loss of anatomic structure delineation.
As coded by GE Medical Systems, the noise reduction filters were designated as follows: Filter A was normal-low; filter B, normal-medium; filter C, normal-high; filter D, special-low; filter E, special-medium; and filter F, special-high. The postprocessed images were then randomized with the baseline images to generate a randomized image data set. Thus, for each of the 19 patients, seven sets of images comprising a baseline set and six postprocessed image sets (postprocessed with filters A–F) were generated.
Image noise or mottle represents the major obstacle in CT radiation dose reduction. A decrease in CT radiation dose leads to an increase in image noise, which at least theoretically may adversely affect the diagnostic acceptability of an examination by obscuring lesions that are visible on CT scans acquired with a higher radiation dose. Noise reduction filters have been designed to decrease image noise in scans acquired with reduced radiation dose. A two-dimensional linear image filtering process that alters the noise properties of CT images on the basis of a knowledge of imaging properties and the noise of the system has been reported (13). The use of nonlinear image processing techniques—in particular, smoothing—has also been reported for creating CT images of good quality with less radiation (14).
Discrete pixel images are composed of a pixel matrix with varying properties and are represented by a digitized value representative of a sensed parameter, such as radiation received within each pixel region. In an image, a group of structural pixels representative of structures of interest and a group of nonstructural pixels representative of nonstructural regions in the image are present. The technique used in development of the presently available noise reduction filters evaluated in this study provides a method for identifying important structures in discrete pixel images. This method makes use of gradient data generated for each pixel to determine a gradient threshold that is then used to separate structural and nonstructural features in the resulting image.
With this technique, 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 matching gradient values for the pixels to a desired value and by comparing the gradient directions for the pixels among one another. A processing circuit is configured to identify structural pixels (which represent features of interest) and nonstructural pixels on the basis of gradient values. Noise reduction filters also perform 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 structures. 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. The noise reduction filters A–F evaluated in the present study were designed to achieve varying levels of segmentation, blending, and sharpening to provide a range of variable visual effects in noise reduction and structure enhancement. Filters A, B, and C apply 33% more aggressive sharpening to the smoothed structure pixels than do filters D, E, and F. Filters A and D are parametrically two times "smoother" than filters B and E and four times smoother than filters C and F. Filters C and F are the most "aggressive" in terms of the level of noise filtering, while filters A and D employ the least aggressive level of filtering.
Qualitative Image Analysis
For qualitative analysis, three subspecialty radiologists with expertise in abdominal imaging (B.C.L. and M.M.M., with 5 years of experience, and M.A.B., with 7 years of experience) independently evaluated randomized images at a workstation (Advantage 3.0 Windows; GE Medical Systems, Waukesha, Wis). To ensure blinded evaluation, images presented to the radiologists did not include patient demographics or protocol information (ie, details regarding the peak kilovoltage and millamperage used at the CT examination and which noise reduction filter had been used in postprocessing the image).
An independent evaluation, as well as a direct side-by-side comparison of baseline and postprocessed images, was performed. For independent evaluation of the randomized images, the presence, number, and location of lesions and their most likely diagnosis were assessed. The standard of reference for the presence of lesions and their diagnoses were findings on standard-dose CT images, which were acquired along with low-dose images in each patient. Lesion attenuation and the presence of contrast enhancement, associated calcification, lymphadenopathy, and/or changes in adjoining tissues (such as fat stranding and altered attenuation) were also recorded. Lesion margins were graded as well-defined, ill-defined, or intermediate.
Images were graded for lesion conspicuity by using a five-point scale, in which a score of 1 indicated the definite presence of an artifact that mimicked a lesion; a score of 2, the presence of a suspicious lesion or perhaps an artifact that mimicked a lesion; a score of 3, the presence of a subtly seen lesion with ill-defined margins; a score of 4, the presence of a well-seen lesion with poorly visualized margins; and a score of 5, the presence of a well-seen lesion with well-visualized margins.
In addition, confidence in making the most likely diagnosis considering the image quality was evaluated by using a three-point scale, in which a score of 3 indicated a confident diagnosis; a score of 2, a reasonable diagnosis; and a score of 1, a possible diagnosis. Side-by-side comparison of postprocessed images with baseline images was performed to evaluate four parameters: image sharpness, image noise, beam-hardening artifacts, and diagnostic acceptability.
Image sharpness was defined as the sharpness of abdominal visceral structures such as the liver, kidneys, adrenal glands, and spleen and was quantified on a five-point scale in which a score of 5 represented the sharpest and a score of 1 represented the most blurred image. Image noise, defined as "graininess" in the image, was also evaluated by using a five-point scale, in which a score of 5 indicated unacceptable noise; a score of 4, above-average increased noise; a score of 3, average noise in an acceptable image; a score of 2, less-than-average noise; and a score of 1, minimum or no image noise.
Beam-hardening artifacts were defined as streak artifacts and were quantified as absent, present but not affecting interpretation, or present and affecting image interpretation. Diagnostic acceptability was graded with a five-point scale, in which a score of 5 indicated superior; a score of 4, above average; a score of 3, average; a score of 2, suboptimal; and a score of 1, unacceptable diagnostic acceptability on the basis of the radiologist’s confidence in making a reasonable diagnosis from an image.
Quantitative Image Analysis
After the qualitative analysis, quantitative attenuation, noise, and contrast-to-noise ratio measurements were obtained for all image sets in which liver lesions were depicted (13 patients). To quantify attenuation and image noise, circular regions of interest of constant size (30 square pixels) were drawn (M.K.K.) in normal liver parenchyma and liver lesions to measure their attenuation values (in Hounsfield units) and image noise (as standard deviations of attenuation coefficients). Lesion-to-liver contrast-to-noise ratio (CNR) was determined with respect to background noise by using the following formula: CNR = (LSAV – LVAV)/BN, where LSAV is the attenuation value of the lesion, LVAV is the attenuation value of the liver, BN is the background noise, and all values are in Hounsfield units.
Statistical Analysis
Statistical analysis was performed by using the Wilcoxon signed rank test, the Student t test, and the test of agreement, where appropriate. Qualitative parameters of postprocessed images were compared with those of the baseline images acquired at reduced tube current. Individual subjective findings were compared by using the Wilcoxon signed rank test (SAS/STAT software; SAS, Cary, NC). The Cohen test was used to determine the degree of agreement between the readers. Quantitative image quality parameters (ie, image noise and contrast-to-noise ratio) of postprocessed images were compared with those of baseline low-dose images by using the Student t test (Excel; Microsoft, Redmond, Wash). Significant statistical correlation was defined as that represented by a P value of less than .05.
If no Bonferroni correction is applied, we would have a chance of 0.2649 (26.49%) of finding one or more significant differences by chance alone in comparing images postprocessed with the six noise reduction filters with the baseline low-dose CT images. To compensate for the increased probability of a chance occurrence of an event in multiple comparisons, the Bonferroni correction (15) was applied to redefine the level of confidence (ie, the 
level or P value). The for each test was lowered to .0085 to bring the level overall back to .05 for the multiple comparisons performed in the present study.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
All 82 lesions were seen in each data set of low-dose images postprocessed with the noise reduction filters. Localization of lesions, their most likely diagnosis, and radiologist confidence in making the diagnosis were not substantially different when the postprocessed images were compared with the baseline low-dose images (P > .5). Concordance between baseline low-dose images and all postprocessed images was noted for lesion characteristics such as attenuation, margins, contrast enhancement, associated calcification (noted in three patients), lymphadenopathy, and changes in adjoining tissues.
All three readers consistently graded lesion conspicuity on images postprocessed with filter F lower than lesion conspicuity on the baseline low-dose images (P = .001). Although readers did document decreased conspicuity on images obtained with the remaining five filters compared with conspicuity on baseline low-dose images, no statistically significant differences were found (P > .05). In a patient with recurrent renal cell carcinoma, beam-hardening artifacts caused by metallic clips in the nephrectomy bed were noted in both baseline low-dose images and images postprocessed with each of the six noise reduction filters.
Image Noise
At side-by-side comparison of postprocessed images with baseline images, all three radiologists noted decreased subjective image noise on postprocessed images compared with the noise on baseline low-dose images, with a maximum decrease in subjective noise on images that had been postprocessed with filters C and F (P = .02 and P = .3, respectively [not significant according to results of Bonferroni correction]). No significant difference in image sharpness or beam-hardening artifacts (seen on images obtained in three patients) was noted between baseline low-dose images and postprocessed images (P > .05). Compared with the baseline low-dose images, the images postprocessed with filters C and F had the lowest diagnostic acceptability scores (P = .04 [not significant according to results of Bonferroni correction]). Diagnostic acceptability of the image data sets postprocessed with the remaining noise reduction filters was inferior but not significantly different from the diagnostic acceptability of the corresponding baseline low-dose images (P > .05).
There was no statistically significant difference in the attenuation values (in Hounsfield units) of lesions and liver parenchyma between baseline low-dose images and postprocessed images (P > .4). A significant (P = .004) reduction in quantitative image noise in liver lesions was noted on images postprocessed with filter F when they were compared with the baseline low-dose images. Although reduced quantitative image noise in liver lesions was also seen on images postprocessed with filters A, D, B, and E (in ascending order of the magnitude of the reduction), the differences were not statistically significant (P = .09–.50). All images postprocessed with noise reduction filters had reduced quantitative image noise compared with the quantitative image noise on corresponding baseline images (Figure).
|
|
|
|
|
|
|
There was substantial concordance between the three radiologists, as determined with a coefficient of 0.7 (two-sided P < .05).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Broadly speaking, the technologic improvements address the issue of radiation reduction by improving scanner efficiency or image quality at low-radiation-dose scanning. Techniques that improve scanner efficiency minimize the unused portion of the x-ray beam during scanning or automatically reduce the beam energy in regions or planes of the body that can be efficiently scanned with decreased radiation (10,11). The former technique involves performing prepatient beam collimation by prospectively removing the portion of the beam not incident on the detector rows and improving detector row configuration with the aim of utilizing the maximum portion of the incident beam.
The use of such x-ray filters as bow-tie or beam-shaping filters reduces the surface radiation dose by minimizing radiation exposure in the thinner portions of patient anatomy (11). The automatic tube current modulation technique involves altering the tube current to reduce exposure to portions of the body that can be scanned with reduced tube current without a substantial change in image quality (12). Techniques that aim to improve image quality at low radiation doses include the use of image reconstruction algorithms and postreconstruction noise reduction filters (6,7,11,13–14,16).
The 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 of anatomic structure delineation. Image postprocessing with noise reduction filters does not require raw scan data and is performed directly on DICOM, or digital imaging and communications in medicine, images. Previous pilot studies have revealed that noise reduction filters decrease qualitative and quantitative noise in low-dose abdominal and chest CT images (6,7). The present study was performed to determine the effect of these filters on lesion detection and characterization on abdominal CT images acquired with reduced radiation doses.
As discussed in the preceding section, radiation reduction at CT may result in increased noise that may affect lesion conspicuity, detection, and characteristics at diagnostic examinations. Therefore, it is imperative that any tool designed to improve the quality of images acquired with reduced radiation dose be assessed for its effect on diagnostic quality. Although we have shown that noise reduction filters decrease image noise with some compromise in image contrast and sharpness—particularly on low-dose chest CT images—to our knowledge, the effect of noise reduction filters on lesion detection, conspicuity, and other characteristics has not been investigated (6,7).
In the present study, all of the noise reduction filters decreased qualitative image noise and quantitative image noise in liver lesions and liver parenchyma; this effect was statistically significant with filter F (P = .004). Unfortunately, the use of filter F also decreased the conspicuity of most lesions in a consistent manner. However, despite the negative effect on lesion conspicuity, the use of noise reduction filters did not substantially affect lesion detection, localization, attenuation, or enhancement pattern or radiologist confidence in making a possible diagnosis. The effect of noise reduction filters on lesion detection was no different on abdominal images than on pelvic images. No pseudolesions were seen on images postprocessed with the noise reduction filters. Similarly, calcifications were seen with equal ease on baseline low-dose images and on postprocessed images. Beam-hardening artifacts in one patient were noted on both baseline low-dose images and on images postprocessed with each of the six noise reduction filters.
Although overall diagnostic acceptability, compared with that on baseline low-dose images, was compromised on images postprocessed with all noise reduction filters, the reduction in diagnostic acceptability was significant with filter F only. Our study results suggest that although noise reduction filters reduce image noise, they result in compromised diagnostic acceptability and lesion conspicuity. However, the effect of currently available noise reduction filters on lesion conspicuity can also affect the detection of more subtle and smaller lesions of the abdomen not evaluated in the present investigation.
These observations suggest that there are problems in utilizing the current versions of noise reduction filters for reducing radiation dose at abdominal and pelvic CT examinations. On the other hand, the use of noise reduction filters for low-dose CT images acquired in high-contrast settings such as CT colonography, CT for evaluation of kidney stones, and CT urography may be helpful. Noise reduction filters may also be helpful in decreasing noise in CT images that are very "grainy" owing to the inadvertent use of low beam energy (commonly caused by low tube current), which can happen more commonly in large patients.
However, further refinements in currently available noise reduction filters are necessary to minimize their adverse effect on lesion conspicuity and eliminate the probability of missing subtle lesions in low-radiation-dose images with high noise content.
There were limitations in our study. The effect of noise reduction filters on lesion detection and characterization on images acquired with greater degrees of radiation dose reduction (ie, with tube currents of less than 120 mA) was not assessed. All images used in this study were obtained in the precontrast or equilibrium phase and not in the more commonly employed portal venous or arterial phases of contrast enhancement. However, our methods of assessing the images and the effect of using the noise reduction filters would have been identical with dynamic image data. Importantly, although the number of lesions detected on postprocessed images was identical to that detected on the baseline images, lesion conspicuity was compromised.
Another limitation of our study was that because baseline low-dose images and postprocessed images were reviewed in the same reading session, radiologists might have been biased in their assessment of lesion detection and diagnosis because they were asked to look for lesions in different sets of images acquired at the same levels. With a larger range of diseases and lesion sizes, the use of noise reduction filters could theoretically have obscured lesions because of decreased conspicuity. An important limitation of our study included an analysis of a small number of patients and the consequent interdependency of data and statistical analyses that resulted from a large number of images being obtained in a small patient cohort. However, we used the Bonferroni correction to redefine the P value for analysis of significant statistical effects, taking into account the number of comparisons made in data analysis.
In conclusion, a reduction in image noise was associated with a decrease in lesion conspicuity after current versions of noise reduction filters were used in low-dose CT images. Use of the current versions of the filters did not compromise other lesion characteristics such as margins, calcification, and status of surrounding soft tissues. These findings suggest that the current versions of noise reduction filters do not improve lesion conspicuity in low-radiation-dose CT images acquired at routine examinations. However, the use of more aggressive filters may help in making low-radiation-dose CT images of high-contrast regions at CT colonography, CT for evaluation of kidney stones, and CT urography more acceptable.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
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]
|
|
|
|
 |
|
 B. C. Lucey, J. C. Varghese, A. Hochberg, M. A. Blake, and J. A. Soto CT-Guided Intervention with Low Radiation Dose: Feasibility and Experience Am. J. Roentgenol., May 1, 2007; 188(5): 1187 - 1194. [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]
|
|
|
|
|
|
Y. Funama, K. Awai, Y. Nakayama, K. Kakei, N. Nagasue, M. Shimamura, N. Sato, S. Sultana, S. Morishita, and Y. Yamashita Radiation Dose Reduction without Degradation of Low-Contrast Detectability at Abdominal Multisection CT with a Low-Tube Voltage Technique: Phantom Study Radiology, December 1, 2005; 237(3): 905 - 910. [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]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
