Evaluation of Apparent Noise on CT Images Using Moving Average Filters.

This study aims to devise a simple method for evaluating the magnitude of texture noise (apparent noise) observed on computed tomography (CT) images scanned at a low radiation dose and reconstructed using iterative reconstruction (IR) and deep learning reconstruction (DLR) algorithms, and to evaluate the apparent noise in CT images reconstructed using the filtered back projection (FBP), IR, and two types of DLR (AiCE Body and AiCE Body Sharp) algorithms. We set a square region of interest (ROI) on CT images of standard- and obese-sized low-contrast phantoms, slid different-sized moving average filters in the ROI vertically and horizontally in steps of 1 pixel, and calculated the standard deviation (SD) of the mean CT values for each filter size. The SD of the mean CT values was fitted with a curve inversely proportional to the filter size, and an apparent noise index was determined from the curve-fitting formula. The apparent noise index of AiCE Body Sharp images for a given mAs value was approximately 58, 23, and 18% lower than that of the FBP, AIDR 3D, and AiCE Body images, respectively. The apparent noise index was considered to reflect noise power spectrum values at lower spatial frequency. Moreover, the apparent noise index was inversely proportional to the square roots of the mAs values. Thus, the apparent noise index could be a useful indicator to quantify and compare texture noise on CT images obtained with different scan parameters and reconstruction algorithms.

[Quantitative Evaluation of Metal Artifacts on CT Images on the Basis of Statistics of Extremes].

It is well-known that metal artifacts have a harmful effect on the image quality of computed tomography (CT) images. However, the physical property remains still unknown. In this study, we investigated the relationship between metal artifacts and tube currents using statistics of extremes. A commercially available phantom for measuring CT dose index 160 mm in diameter was prepared and a brass rod 13 mm in diameter was placed at the centerline of the phantom. This phantom was used as a target object to evaluate metal artifacts and was scanned using an area detector CT scanner with various tube currents under a constant tube voltage of 120 kV. Sixty parallel line segments with a length of 100 pixels were placed to cross metal artifacts on CT images and the largest difference between two adjacent CT values in each of 60 CT value profiles of these line segments was employed as a feature variable for measuring metal artifacts; these feature variables were analyzed on the basis of extreme value theory. The CT value variation induced by metal artifacts was statistically characterized by Gumbel distribution, which was one of the extreme value distributions; namely, metal artifacts have the same statistical characteristic as streak artifacts. Therefore, Gumbel evaluation method makes it possible to analyze not only streak artifacts but also metal artifacts. Furthermore, the location parameter in Gumbel distribution was shown to be in inverse proportion to the square root of a tube current. This result suggested that metal artifacts have the same dose dependence as image noises.

Development and validation of the effective CNR analysis method for evaluating the contrast resolution of CT images.

Contrast resolution is an important index for evaluating the signal detectability of computed tomographic (CT) images. Recently, various noise reduction algorithms, such as iterative reconstruction (IR) and deep learning reconstruction (DLR), have been proposed to reduce the image noise in CT images. However, these algorithms cause changes in the image noise texture and blurred image signals in CT images. Furthermore, the contrast-to-noise ratio (CNR) cannot be accurately evaluated in CT images reconstructed using noise reduction methods. Therefore, in this study, we devised a new method, namely, "effective CNR analysis," for evaluating the contrast resolution of CT images. We verified whether the proposed algorithm could evaluate the effective contrast resolution based on the signal detectability of CT images. The findings showed that the effective CNR values obtained using the proposed method correlated well with the subjective visual impressions of CT images. To investigate whether signal detectability was appropriately evaluated using effective CNR analysis, the conventional CNR analysis method was compared with the proposed method. The CNRs of the IR and DLR images calculated using conventional CNR analysis were 13.2 and 10.7, respectively. By contrast, those calculated using the effective CNR analysis were estimated to be 0.7 and 1.1, respectively. Considering that the signal visibility of DLR images was superior to that of IR images, our proposed effective CNR analysis was shown to be appropriate for evaluating the contrast resolution of CT images.

Exploring Indicators of Subcutaneous Tissue Fluid Accumulation in Breast Cancer-Related Lymphedema Patients Using Fractal Analysis with Virtual Volume.

Background: Breast cancer treatment sometimes causes a chronic swelling of the arm called breast cancer-related lymphedema (BCRL). Its progression is believed to be irreversible and is accompanied by tissue fibrosis and lipidosis, so preventing lymphedema from progressing by appropriate intervention at the site of fluid accumulation at an early stage is crucial. The tissue structure can be evaluated in real time by ultrasonography, and this study aims at assessing the ability of fractal analysis using virtual volume in detecting fluid accumulation within BCRL subcutaneous tissue via ultrasound imaging. Methods and Results: We worked with 21 women who developed BCRL (International Society of Lymphology stage II) after unilateral breast cancer treatment. Their subcutaneous tissues were scanned with an ultrasound system (Sonosite Edge II; Sonosite, Inc., FUJIFILM) using a 6- to 15-MHz linear transducer. Then, a 3-Tesla MR system was used to confirm fluid accumulation in the corresponding area of the ultrasound system. Significant differences in both H + 2 and complexity were observed among the three groups (with hyperintense area, without hyperintense area, and unaffected side) (p < 0.05). Post hoc analysis (Mann-Whitney U test; Bonferroni correction p < 0.0167) revealed a significant difference for "complexity." The evaluation of the distribution in Euclidean space showed that the variation of the distribution decreased in the order of unaffected, without hyperintense area, and with hyperintense area. Conclusion: The "complexity" of the fractal using virtual volume seems to be an effective indicator of the presence or absence of subcutaneous tissue fluid accumulation in BCRL.

A new evaluation method for image noise reduction and usefulness of the spatially adaptive wavelet thresholding method for CT images.

We have proposed a direct evaluation method concerning preservation of noise-free components for image noise reduction. This evaluation method is to graphically estimate how well a noise-reduction method will preserve noise-free image components by using the normal probability plot of the image pixel value difference between an original image and its noise-reduced image; this difference is equivalent to the "method noise" which was defined by Buades et al. Further, by comparing the linearity of a normal probability plot for two different noise reduction methods, one can graphically assess which method will be more able to preserve the noise-free component than the other. As an illustrative example of this evaluation method, we have evaluated the effectiveness of the spatially-adaptive BayesShrink noise-reduced method devised by Chang et al., when applied to chest phantom CT images. The evaluation results of our proposed method were consistent with the visual impressions for the CT images processed in this study. The results of this study also indicate that the spatially-adaptive BayesShrink algorithm devised by Chang et al. will work well on the chest phantom CT images, although the assumption for this method is often violated in CT images, and the assumption postulated for the spatially-adaptive BayesShrink method is expected to have sufficient robustness for CT images.

Relationship between Dilution Magnification of Non-Ionic Iodinated Contrast Medium and Amplification Effect of Radiation Dose.

Objective: Neuroendovascular treatments are less invasive than surgical clipping. However, the number of fluoroscopy runs may be greater when a contrast medium is used than when routine angiography is performed. Several recent studies have suggested that an iodinated contrast medium causes an increase in the radiation dose. Therefore, it is clinically important to identify physical factors causing amplification of the radiation dose. The purpose of this study was to investigate how dilution of a contrast medium with water influences the amplification effect of the radiation dose using simulation analysis. Methods: Three different types of commercially available contrast media, namely, iopamidol, iohexol, and iodixanol, were diluted 1.7-3.3 times with water and placed in the left brain parenchyma of a numerical brain phantom. Using the Monte Carlo simulation method, the phantom was exposed to X-ray beams under constant exposure conditions, and the energy absorbed in the entire region of the left brain parenchyma was estimated. At the same time, the content and volume of a contrast medium in the cerebral vessels were predicted on the basis of pharmacokinetic and fractal analyses. Results: The increase in absorbed energy was attributed to secondary electrons emitted from the contrast medium and varied depending on its content and volume. Interestingly, the amount of energy absorbed increased with increasing dilution of the contrast medium. Furthermore, the amplification effect of the radiation dose varied according to the type of contrast medium used. Conclusion: These results suggest that the amplification effect of the radiation dose is closely related to an increase in the cross-sectional area in which the X-rays interact with the contrast medium, which is caused by increased distribution of contrast medium in the cerebral vessels. When the contrast medium is diluted with water, its spread in the cerebral vessels plays a more important role than its content in the amplification effect of the radiation dose.

Quantification of image quality using information theory.

Aims of present study were to examine usefulness of information theory in visual assessment of image quality. We applied first order approximation of the Shannon's information theory to compute information losses (IL). Images of a contrast-detail mammography (CDMAM) phantom were acquired with computed radiographies for various radiation doses. Information content was defined as the entropy Σp( i )log(1/p ( i )), in which detection probabilities p ( i ) were calculated from distribution of detection rate of the CDMAM. IL was defined as the difference between information content and information obtained. IL decreased with increases in the disk diameters (P < 0.0001, ANOVA) and in the radiation doses (P < 0.002, F-test). Sums of IL, which we call total information losses (TIL), were closely correlated with the image quality figures (r = 0.985). TIL was dependent on the distribution of image reading ability of each examinee, even when average reading ratio was the same in the group. TIL was shown to be sensitive to the observers' distribution of image readings and was expected to improve the evaluation of image quality.

Quantitative assessment of the influence of anatomic noise on the detection of subtle lung nodule in digital chest radiography using fractal-feature distance.

PURPOSE: To confirm whether or not the influence of anatomic noise on the detection of nodules in digital chest radiography can be evaluated by the fractal-feature distance. MATERIALS AND METHODS: We used the square images with and without a simulated nodule which were generated in our previous observer performance study; the simulated nodule was located on the upper margin of a rib, the inside of a rib, the lower margin of a rib, or the central region between two adjoining ribs. For the square chest images, fractal analysis was conducted using the virtual volume method. The fractal-feature distances between the considered and the reference images were calculated using the pseudo-fractal dimension and complexity, and the square images without the simulated nodule were employed as the reference images. We compared the fractal-feature distances with the observer's confidence level regarding the presence of a nodule in plain chest radiograph. RESULTS: For all square chest images, the relationships between the length of the square boxes and the mean of the virtual volumes were linear on a log-log scale. For all types of the simulated nodules, the fractal-feature distance was the highest for the simulated nodules located on the central region between two adjoining ribs and was the lowest for those located in the inside of a rib. The fractal-feature distance showed a linear relation to an observer's confidence level. CONCLUSION: The fractal-feature distance would be useful for evaluating the influence of anatomic noise on the detection of nodules in digital chest radiography.

Fractal-feature distance as a substitute for observer performance index in contrast-detail examination.

PURPOSE: To assess whether or not the fractal-feature distance using the box-counting algorithm can be a substitute for observer performance index. METHODS AND MATERIALS: Contrast-detail (C-D) phantom images were obtained at various mAs-values (0.5-4.0 mAs) and 140 kV(p) with a Fuji computed radiography system, and the reference image was acquired at 50 mAs; all cylindrical targets in the C-D phantom were visualized on this image. The C-D images were converted to binary images using the profile curves around the smallest cylindrical target images on the reference images. The fractal analysis was conducted using the box-counting algorithm for these binary images. The fractal-feature distances between the low-dose and reference images were calculated using the fractal dimension and the complexity. Furthermore, we performed the C-D analysis in which ten radiologists participated, and compared the fractal-feature distances with the image quality figures (IQF) derived from the C-D analysis with Markov chain. RESULTS: For all C-D phantom radiographs, the relationship between the length of the square boxes and the number of boxes to cover the positive pixels of the binary image was linear on a log-log scale (r>or=0.999). A strong linear correlation was found between the fractal-feature distance and IQF (r=0.990). CONCLUSION: We have shown that the binary image of C-D phantom can be analyzed by the box-counting algorithm and its fractal-feature distance increases as the radiation dose decreases. Furthermore, we have shown that the fractal-feature distances will be equivalent to IQFs in C-D analysis.

Contrast enhancement efficacy of iodinated contrast media: Effect of molecular structure on contrast enhancement.

PURPOSE: To investigate the contrast enhancement in DSA images based on the X-ray absorption characteristics of iodinated contrast media. METHODS: We have derived a new formula of predicting the pixel value ratio of two different contrast media and designate it as "Contrast Enhancement Ratio (CER)". In order to evaluate the accuracy of CER, we have evaluated the relationship between CER and pixel value ratio for all combinations of eleven iodinated contrast media. The non-ionic iodinated contrast media, iopamidol, iomeprol, iopromide, ioversol, iohexol, and iodixanol, were evaluated in this study. Each contrast medium was filled in the simulated blood vessel in our constructed anthropomorphic phantom, and DSA images were obtained using an angiographic imaging system. To evaluate the contrast enhancement of the contrast medium, the mean pixel value was calculated from all pixel values in the vascular image. RESULTS: CER was indicated to agree well with the pixel value ratio of two different contrast medium solutions and showed a good accuracy. CER was also shown to have a good linear relation to the pixel value ratio when the iodine concentration was constant. This means that the molecular structure of the contrast media affects contrast enhancement efficacy. Furthermore, in evaluation of contrast enhancement of iodinated contrast media by using the weight factor (that is a key factor in CER) ratio, Iodixanol, and iopamidol, and iomeprol have the same ability of contrast enhancement in DSA images, and iohexol shows the lowest ability. CONCLUSIONS: We have derived a new formula (CER) of predicting the pixel value ratio of two different contrast medium solutions, and shown that CER agreed well with the pixel value ratio for blood vessel filled with eleven contrast media.

Information loss in visual assessments of medical images.

We applied information theory to quantify information losses in assessing contrast-detail (C-D) analysis. Images of a C-D phantom were acquired with a flat panel detector (FPD) and a computed radiography (CR) by changing surface entrance doses. Six phantom radiographs (FPD: five images; CR: one image) were prepared for visual evaluations. Thirteen radiographers and two radiologists participated in the observation test. Detectability was defined as the shortest length of the cylinders of which border the observers could recognize from the background, and was recorded using row number. Information content was defined as the entropy summation operatorp(i)log(1/p(i)) with detection probabilities p(i), which were calculated from distribution of detection rate of the ith column. Information loss, in unit of bits, was calculated as the difference between information obtained and information content when all the columns were detected. The information losses decreased with the increase in cylinder diameters and with the increase in surface entrance dose. Because the information loss varies depending on distribution of detection rate, this method of using the information theory was expected to be more sensitive in evaluating the C-D image quality than using the averaged values of detectability.

The application of Markov theory to contrast-detail analysis.

RATIONALE AND OBJECTIVES: To improve the accuracy of contrast-detail (C-D) analysis, we have devised a new evaluation method of the detection performance in the C-D analysis by using the Markov process model. This article describes this new evaluation method and the results of applying it to the illustrative examples. MATERIALS AND METHODS: A commercially available C-D phantom was employed as a test object, and 20 phantom radiographs with an average background density of 0.8 were prepared. Thirty-five observers interpreted all phantom radiographs independently of each other with the room lights off. We assume the Markov model in which the selection of the smallest visible contrast for each row of disks with the same diameter depends only on the contrast selected as the smallest visible one for the same row in the just-before reading experiment. By using this Markov model, the contrast detectability for each disk diameter was calculated, and a C-D diagram was constructed based on these results. The conventional C-D diagram was also derived from the averages of contrast detectability obtained from the interpretations of all phantom radiographs. RESULTS: All C-D diagrams obtained from our devised method agreed well with each other and were in good accord with the conventional C-D diagram from the interpretations of all phantom radiographs. CONCLUSIONS: We have devised a new evaluation method of C-D analysis by using the Markov model and have shown this method will be consistent with the conventional evaluation method in the C-D analysis.

A method of clustering observers with different visual characteristics.

Evaluation of observer's image perception in medical images is important, and yet has not been performed because it is difficult to quantify visual characteristics. In the present study, we investigated the observer's image perception by clustering a group of 20 observers. Images of a contrast-detail (C-D) phantom, which had cylinders of 10 rows and 10 columns with different diameters and lengths, were acquired with an X-ray screen-film system with fixed exposure conditions. A group of 10 films were prepared for visual evaluations. Sixteen radiological technicians, three radiologists and one medical physicist participated in the observation test. All observers read the phantom radiographs on a transillumination image viewer with room lights off. The detectability was defined as the shortest length of the cylinders of which border the observers could recognize from the background, and was recorded using the number of columns. The detectability was calculated as the average of 10 readings for each observer, and plotted for different phantom diameter. The unweighted pair-group method using arithmetic averages (UPGMA) was adopted for clustering. The observers were clustered into two groups: one group selected objects with a demarcation from the vicinity, and the other group searched for the objects with their eyes constrained. This study showed a usefulness of the cluster method to select personnels with the similar perceptual predisposition when a C-D phantom was used in image quality control.

Arterial contour detectability in head CT angiography.

PURPOSE: Arterial contour extraction is essential for visualization and analysis of vasculature in CT angiography (CTA). A means for evaluating the detectability of artery contours CTA images is required. We developed and tested a new method for this purpose based on phase information from two-dimensional Fourier transforms of CTA images. The relationship between arterial contour detectability and a patient's ocular lens dose was evaluated in CTA images obtained with various tube voltages and currents. METHODS: A head phantom was designed for use as a target object containing a simulated vascular tree, filled with dilute contrast medium (10 mg iodine/ml). The head phantom was scanned using a 64-multidetector CT scanner with tube voltages of 80-140 kV and tube currents corresponding to volume CT dose index [Formula: see text] ranging from 24.4 to 72.8 mGy. Lens doses were measured using the planar silicon PIN-photodiode system. The quality of artery contours in the CTA source images was assessed using a computed detectability index. RESULTS: Lens dose increased proportionally with tube voltage and current. The use of 80 kV provided the highest contour detectability. However, for each tube voltage, the detectability of artery contours was almost constant across the CTDI(vol) values. These results were mostly consistent with the subjective recognition of artery contours on CTA images. CONCLUSIONS: A CTA protocol using 80 kV and 420 mA can reduce the radiation exposure to ocular lens by approximately 40 %, and improve the artery contour detectability compared with a routine protocol.

[Grouping of observer's visual characteristics on the basis of difference in the criteria of visual detection and selection of the fittest clustering method:verification of observer's group with radiologist-like visual characteristics].

Various clustering methods are used in cluster analyses, with each clustering method demonstrating unique advantages. Therefore, it is important to make the best use of the advantages each method provides. We have recognized that it is necessary in the evaluation of X-ray images to classify observers quantitatively according to visual characteristics (grouping of observers) and have clustered observers using the UPGMA method, which is one of the clustering methods. We found that the observers were clustered into two different groups, one with radiologist-like characteristics and the other with medical physicist-like characteristics. Furthermore, we suggested that the group with radiologist-like characteristics was suitable for QC of X-ray images. However, it is doubtful whether the UPGMA method is most suitable for the grouping of observers. In this work we clustered observers using various clustering methods and examined the most suitable method for the evaluation of X-ray images. The results showed that the ward method was least suitable for the grouping of observers, and they were distinctly grouped into two different categories by using a further method.

[Classification of observer-groups with cluster analysis and the criteria of visual decision for each group].

It is important that the evaluation of X-ray images include the observer's visual characteristics. However, evaluations of X-ray images that include these characteristics are not performed because of the difficulty of quantitatively elucidating visual characteristics. In this study, we classified observers into groups (clusters) by the same criteria of visual decision, using cluster analysis (unweighted Pair-Group method using arithmetic averages), and evaluated X-ray images on the basis of this separation. Clinical application is also discussed. It was found that observer clustering caused a decrease in between-observer variation. Observers were grouped into two different categories: one with the characteristics of radiologists and the other with the characteristics of medical physicists. Our results indicated that the group with the characteristics of radiologists was suitable for the quality control (QC) of X-ray images.

A new automated assessment method for contrast-detail images by applying support vector machine and its robustness to nonlinear image processing.

The automated contrast-detail (C-D) analysis methods developed so-far cannot be expected to work well on images processed with nonlinear methods, such as noise reduction methods. Therefore, we have devised a new automated C-D analysis method by applying support vector machine (SVM), and tested for its robustness to nonlinear image processing. We acquired the CDRAD (a commercially available C-D test object) images at a tube voltage of 120 kV and a milliampere-second product (mAs) of 0.5-5.0. A partial diffusion equation based technique was used as noise reduction method. Three radiologists and three university students participated in the observer performance study. The training data for our SVM method was the classification data scored by the one radiologist for the CDRAD images acquired at 1.6 and 3.2 mAs and their noise-reduced images. We also compared the performance of our SVM method with the CDRAD Analyser algorithm. The mean C-D diagrams (that is a plot of the mean of the smallest visible hole diameter vs. hole depth) obtained from our devised SVM method agreed well with the ones averaged across the six human observers for both original and noise-reduced CDRAD images, whereas the mean C-D diagrams from the CDRAD Analyser algorithm disagreed with the ones from the human observers for both original and noise-reduced CDRAD images. In conclusion, our proposed SVM method for C-D analysis will work well for the images processed with the non-linear noise reduction method as well as for the original radiographic images.

A detection method for streak artifacts and radiological noise in a non-uniform region in a CT image.

By using the CT images obtained by subtracting two CT images acquired under the same conditions and slice locations, we have devised a method for detecting streak artifacts in non-uniform regions and only radiological noise components in CT images. A chest phantom was scanned using 16- and 64-multidetector row helical CT scanners with various mAs values at 120kVp. The upper lung slice image was employed as a target image for evaluating the streak artifacts and radiological noise. One hundred parallel line segments with a length of 80 pixels were placed on the subtracted CT image, and the largest CT value in each CT value profile was employed as a feature variable of the streak artifacts; these feature variables were analyzed with the extreme value theory (Gumbel distribution). To detect only the radiological noise, all CT values contained in the 100 line profile were plotted on normal probability paper and the standard deviation was estimated from the inclination of its fitted line for the CT value plots. The two detection methods devised in this study were able to evaluate the streak artifacts and radiological noise in the CT images with high accuracy.

A method for estimating noise variance of CT image.

Rank et al. have proposed an algorithm for estimating image noise variance composed of the following three steps: the noisy image is first filtered by a difference operator; a histogram of local signal variances is then computed; and, finally the noise variance is estimated from a statistical evaluation of the histogram. We have verified the accuracy of this algorithm on a CT image by indirect methods, and have shown that this method is able to estimate CT image noise variance with reasonable accuracy, regardless of whether or not the noiseless image is uniform. Further, we have proposed a simple alternative method for the last two steps of the Rank et al. method. However, one must pay attention to the fact that the estimated noise variance will be biased when the nearest two pixels are correlated and that this algorithm does not work well if the assumption of stationarity of noise components is violated.

Quantitative assessment of image noise and streak artifact on CT image: comparison of z-axis automatic tube current modulation technique with fixed tube current technique.

The purpose of our study is to quantitatively assess the effects of z-axis automatic tube current modulation technique on image noise and streak artifact, by comparing with fixed tube current technique. Standard deviation of CT-values was employed as a physical index for evaluating image noise, and streak artifact was quantitatively evaluated using our devised Gumbel evaluation method. z-Axis automatic tube current modulation technique will improve image noise and streak artifact, compared with fixed tube current technique, and will make it possible to significantly reduce radiation doses at lung levels while maintaining the same image quality as fixed tube current technique.