PurposeThe average (fav) or peak (fpeak) noise power spectrum (NPS) frequency is often used as a one-parameter descriptor of the CT noise texture. Our study develops a more complete two-parameter model of the CT NPS and investigates the sensitivity of human observers to changes in it.ApproachA model of CT NPS was created based on its fpeak and a half-Gaussian fit (σ) to the downslope. Two-alternative forced-choice staircase studies were used to determine perceptual thresholds for noise texture, defined as parameter differences with a predetermined level of discrimination performance (80% correct). Five imaging scientist observers performed the forced-choice studies for eight directions in the fpeak/σ-space, for two reference NPSs (corresponding to body and lung kernels). The experiment was repeated with 32 radiologists, each evaluating a single direction in the fpeak/σ-space. NPS differences were quantified by the noise texture contrast (Ctexture), the integral of the absolute NPS difference.ResultsThe two-parameter NPS model was found to be a good representation of various clinical CT reconstructions. Perception thresholds for fpeak alone are 0.2 lp/cm for body and 0.4 lp/cm for lung NPSs. For σ, these values are 0.15 and 2 lp/cm, respectively. Thresholds change if the other parameter also changes. Different NPSs with the same fpeak or fav can be discriminated. Nonradiologist observers did not need more Ctexture than radiologists.Conclusionsfpeak or fav is insufficient to describe noise texture completely. The discrimination of noise texture changes depending on its frequency content. Radiologists do not discriminate noise texture changes better than nonradiologists.
KEYWORDS: Modulation transfer functions, CT reconstruction, Computed tomography, Image quality, Computer simulations, Data acquisition, Deep learning, Sensors, Medical image reconstruction
As Deep Learning Reconstruction (DLR) begins to dominate computed tomography (CT) reconstruction, performance evaluation via conventional phantoms with uniform backgrounds and specific sizes may benefit from augmentation with simulated controlled test objects inserted into anatomical backgrounds. The purpose of this study is to validate a simulation tool with physics-based image quality metrics both in phantom and in patient data. An analytic forward projection tool, based on detector and source geometry with beam spectra was designed to match the specifications of Canon Medical’s Aquilion ONE Prism. The CatphanTM 500 and two water phantoms, 24cm and 32cm in diameter, were scanned with Aquilion ONE Prism at various mA levels and reconstructed with FBP. Corresponding simulated images were generated. The CT number, noise power spectrum (NPS) and modulation transfer function (MTF) were evaluated and compared between the simulated images and actual images. Simulated projection data of a CatphanTM sensitometry cylinder was also combined with a patient sinogram and reconstructed with a variety of kernels. The MTF of three different contrast rods were measured and compared with the MTF measured from CatphanTM. The CT numbers were equivalent between the simulated data and the image acquired from the actual CT system. The MTF measured from the simulated data of both phantom and patient image matched with the MTF from CatphanTM. The noise properties of the simulated data also aligned with the NPS of the 24cm and 32cm water phantom image. The simulation tool was able to generate images with image quality equivalent to the images scanned and reconstructed from the actual CT system. With this validation study, the simulation will be utilized to further evaluate the performance of deep learning reconstructions (DLRs).
Physicists generally use the Noise Power Spectra (NPS) and Standard Deviation (SD) to characterize noise properties associated with non-linear reconstruction algorithms. However, these metrics capture only first and second order statistics. The purpose of this work is to characterize the impact of the higher order statistics, commonly associated with non-linear reconstruction, on noise texture. Images of a 32cm water phantom were acquired on the Aquilion ONE Genesis Computed Tomography (CT) system and reconstructed with deep learning reconstruction (DLR), model-based iterative reconstruction (MBIR), hybrid iterative reconstruction (AIDR), and filtered backprojection (FBP). Regions of interest (ROIs) of 100x100pixels were extracted from the center of the images. Pure Gaussian noise counterpart image datasets with the same mean, SD, and NPS as each acquired data condition were also generated by convolving random white noise with the root-NPS of the acquired data. Nine naïve observers were tasked with distinguishing the acquired noise image from its pure Gaussian counterpart via a two-alternative forced choice experiment. Results showed the FBP images appeared indistinguishable from their pure Gaussian counterparts (Percent Correct=54%), while MBIR images were readily distinguishable from Gaussian ones (Percent Correct=98-100%). DLR and AIDR images were more difficult to distinguish from their pure Gaussian counterparts (Percent Correct=58-88%), than MBIR, which indicates that it is more similar in perceived texture to Gaussian noise. This work demonstrates the appearance of CT noise texture may be dependent on higher orders statistics not captured by the NPS; noise textures with identical NPS and SD can be distinguished based on non-Gaussian properties.
Low-contrast lesions are difficult to detect in noisy low-dose CT images. Improving CT image quality for this detection task has the potential to improve diagnostic accuracy and patient outcomes. In this work, we use tunable neural networks for CT image restoration with a hyperparameter to control the variance/bias tradeoff. We use clinical images from a super-high-resolution normal-dose CT scan to synthesize low-contrast low-dose CT images for supervised training of deep learning CT reconstruction models. Those models are trained using with multiple noise realizations so that variance and bias can be penalized separately. We use a training loss function with one hyperparameter called the denoising level, which controls the variance/bias tradeoff. Finally, we evaluate the CT image quality to find the optimal denoising level for low-contrast lesion detectability. We evaluate performance using a shallow neural network model classification model to represent a suboptimal image observer. Our results indicate that the optimal networks for low-contrast lesion detectability are those that prioritize bias reduction rather than mean-squared error, which demonstrates the potential clinical benefit of our proposed tunable neural networks.
The Noise Power Spectra (NPS) only characterizes first and second order statistics associated with noise in Computed
Tomography (CT) reconstructions. The purpose of this work is to characterize the impact of the higher order statistics on
perception of noise texture for a variety of reconstruction algorithms. Images of a 32 cm water phantom were acquired on
the Aquilion ONE Genesis CT system and reconstructed with AiCE deep learning reconstruction (DLR), model-based
iterative reconstruction (MBIR), hybrid iterative reconstruction (AIDR), and filtered backprojection (FBP). Regions of
interest (ROIs) of 100x100pixels were extracted from the center of the images and 4th order statistics of each ROI were
assessed via excess kurtosis measurement. Pure Gaussian noise counterpart image datasets with the same mean, standard
deviation (SD), and NPS as each acquired data condition were also generated by convolving random white noise with the
root-NPS of the acquired data. Nine naïve observers were tasked with distinguishing the acquired noise image from the
pure Gaussian counterpart via a two-alternative forced choice experiment. Excess kurtosis in the image ROIs was 0.01
for FBP, 0.74 to 0.85 for FIRST, 0.03 to 0.08 for AIDR, and -0.13 to 0.21 for AiCE. Results showed the FBP images appeared
indistinguishable from their pure Gaussian counterparts with a Percent Correct (PC)=54%, while MBIR images were
readily distinguishable from their pure Gaussian counterparts, PC=98 to 100%. DLR and AIDR images are more difficult
to distinguish from their pure Gaussian counterparts, with the PC ranging from 58% to 88%. The discriminability index
derived from the PCs correlated strongly with excess kurtosis.
Noise texture in CT images, commonly characterized by using the noise power spectrum (NPS), is mainly dictated by the shape of the reconstruction kernel. The peak frequency of the NPS (fpeak) is often used as a one-parameter metric for characterizing noise texture. However, if the downslope of the NPS beyond the fpeak influences noise texture visibly, then fpeak is insufficient as a single descriptor. Therefore, we investigated the human-detectable differences in NPSs having different fpeak and/or downslope parameters. NPSs were estimated using various reconstruction kernels on a commercial CT scanner. To quantify NPS downslope, half of a Gaussian function was fit through the NPS portion that lies beyond fpeak. The σ of this Gaussian was used as the downslope descriptor of the NPS. A two alternative forced choice observer study was performed to determine the just noticeable- differences (JND) in fpeak only, σ only, and both simultaneously. Visibility thresholds for these changes were determined and an elliptical limiting detectability boundary was determined. The JND threshold ellipse is centered on the reference values and has a major and minor radius of 0.47 lp/cm and 0.12 lp/cm, respectively. The major radius makes an angle of 143° with the x-axis. A change in only fpeak of 0.2 lp/cm is below the detection threshold. This number changes if the apodization part of the NPS changes simultaneously. In conclusion, both the peak frequency and the apodization section of the NPS influence the detectability of changes in image noise texture.
The need for table motion in multi-detector CT causes image volumes acquired for whole organ motion and perfusion studies to lack temporal uniformity. The next revolution in clinical CT, dynamic volume CT, mitigates this limitation by providing the ability to acquire an entire organ with isotropic resolution in a single gantry rotation with no table movement. The first dynamic volume CT scanner has recently been introduced and comprises 320 detector rows of
0.5mm channel thickness, covering 16cm of anatomy in one rotation of 0.35sec. This scanner offers many advancements in terms of temporal uniformity, reconstruction, and radiation dose. This system significantly reduces motion artifact and eliminates contrast phase differences within the volume. Because this scanner does not require
helical acquisition for volumetric imaging, it delivers significantly less dose for applications such as CT coronary angiography exams as well as reduced dose in most other applications. Furthermore, by eliminating table motion, the need for complex interpolation methods that can distort cardiac images is removed. Image quality is not sacrificed compared with standard 64-row CT scanners, as demonstrated via low contrast, resolution, and accuracy measurements presented in this work. By capturing the entire brain in one rotation, brain perfusion, bone subtraction, and quantitative perfusion analysis are now possible with a single low dose exam. Dynamic volume CT offers to change the way medicine approaches stroke patients, myocardial perfusion studies, and imaging of other moving body parts such as the
lung and joints.
The spatial characteristics of noise in X-ray CT can influence object detectability. Now, as three dimensional image
reformations become more clinically common, it has become vital to understand the structure of noise in the x, y, and z
directions independently. The purpose of this paper is to study noise structure in the radial direction and tangential
direction, under varying conditions, including a wedge filter and acquisition techniques (i.e. half scan vs full scan).
Because the effect of the reconstruction algorithm on an image is highly dependent on spatial location within the field of
view, the effect of off-center vs centered positioning in each direction is also examined. The noise spatial frequency
distribution was investigated via calculation of the noise power spectrum (NPS) through Fourier methods on simulated
water images. As expected, noise structure at center was equivalent in both the radial and tangential directions.
Towards the periphery, overall noise power was muted. However, in the tangential direction high frequency noise power
was preserved more than twice as much as in the radial direction. Towards the periphery noise becomes more low-frequency
in the radial direction, while in the tangential direction it becomes more high-frequency. The half scan
increased both noise magnitude and low-mid spatial correlation in the NPS compared to full scan. In conclusion, noise
spatial structure is directionally dependent off-center and this may have an impact on object detectability in directional
reformations.
The aim of our investigation was to assess the influence of both CT acquisition dose and reconstruction kernel on computer-aided detection (CAD) of pulmonary nodules. Our hypothesis is that the detection of small nodules is affected by the noise characteristics of the image and the signal to noise ratio of the nodule and bronchiovascular anatomy. Knowledge gained from this experiment will assist in developing an advanced CAD system designed to detect smaller and more subtle nodules with minimal false positives. Eleven research subjects were selected from the Lung Image Database Consortium (LIDC) database based on our inclusion criteria of: 1) having at least one nodule and 2) available raw CT projection data for the series that our institution submitted to the LIDC study. Using the original raw projection data, research software simulated raw projection data acquired with a dose reduced 32-40% from the original scan. Projection data for both dose levels was reconstructed with smooth to very sharp kernels (B10f, B30f, B50f, and B70f). The resulting series were used to investigate the influence of dose and reconstruction kernel on CAD performance. A prototype CAD system was used to investigate changes in sensitivity and false positives with varying imaging parameters. In a sub-study, the prototype system was compared to a commercial CAD system. We did not have enough subjects to conclude significance, but the results indicate our research system had a higher sensitivity with the smooth or medium reconstruction kernels than with the sharper kernels. The sensitivity was similar for both dose levels. The false positive rate was higher with the smooth kernels and the lower dose levels.
Low contrast detection tasks, such as the detection of subtle ground glass nodules, are low signal to noise (SNR) situations that can be greatly influenced by choice of reconstruction filter. The goal of this work is to examine tradeoffs in noise, resolution, and dose on the SNR of low contrast test objects, resembling spherical lung nodules, to potentially improve reconstruction filter selection for a given nodule detection task. To perform these experiments, the Modulation Transfer Function (MTF) was calculated for each reconstruction filter available. Next, simulated signal images were created using 2mm section thicknesses of 1 cm diameter spheres of varying contrast levels. The Noise Power Spectra (NPS) were then calculated for each reconstruction filter to be examined. The signal to noise metric used is the ideal Bayesian observer SNR metric, which takes into account the spatial correlations in noise introduced by the filter (and described by the NPS). The IBO SNR was calculated under a variety of reconstruction conditions: (a) varying mAs so that each reconstruction filter results in the same standard deviation; (b) constant mAs and varying reconstruction filter; (c) one reconstruction filter using varying mAs. These measurements provide an opportunity to examine the important tradeoffs in SNR with noise, resolution, and dose that occur with selection of a reconstruction filter and can potentially lead to a quantitative basis for filter selection, improving lesion detectability.
The effects of collimation and reconstruction algorithm on image noise in CT were investigated using several low dose techniques typical of lung cancer screening protocols. Tube current settings were 10mA, 20mA, 50mA, and 100 mA, all at 120kVp and 0.8 sec rotation time. A homogeneous water phantom was scanned with various mA setting, collimation, and reconstruction algorithm combinations. Noise was measured under each condition and radiation doses for each tube current used were extrapolated from CTDIw values measured at each collimation. Noise values for each mA, collimation, and algorithm combination were compared as a function of radiation dose (CTDIw) and were also compared with the noise and radiation dose values of currently employed lung cancer screening techniques (e.g. 120 kVp, 50 mA, .8 sec, 2.5 mm collimation, bone reconstruction algorithm). The data shows that thinner slices (those < 2.5 mm) at the same mA setting and reconstruction algorithm yield higher noise values and higher radiation dose values than current techniques, as high as nearly 3 times the original CTDIw. In lung cancer screening imaging with CT, moving to thinner slices presents some difficult tradeoffs between dose and noise. Reconstruction algorithm can be used to reduce image noise, but at a price of reduced in-plane spatial resolution, offsetting some of the benefit of using thinner slices to detect smaller lesions.
The effects of CT reconstruction algorithm on density mask score (percentage of voxels < -910 HU) and total lung volume were investigated for emphysema patients with low density mask scores (approximately 20% or below) and patients with higher scores. Based on MTF curves, reconstruction algorithms were classed as standard (i.e. non-enhancing) or over-enhancing. Each image data set was reconstructed with both the standard reconstruction algorithm and the over-enhancing algorithm. All other factors, such as slice collimation and reconstruction interval, were constant. Twenty-nine patients were divided into a high density mask score group (n=10) and low density mask score group (n=19). For the low density mask subgroup, the over-enhancing category yielded an average increase in density mask score of 12.6% compared to standard (i.e. a shift in average score from 14.8% to 27.4%). The maximum shift in score for the low density mask group was 15.9% while the minimum shift was 9.2%. The high density mask group yielded an average shift of 8.7%, with a minimum shift of 3.8% and a maximum shift of 15.3%. The low density group displayed a 1.2% decrease in volume for the over-enhancing category and a 0.8% decrease for the standard category. These volume changes are likely clinically insignificant. Reconstruction algorithm does, however, have a significant effect on the density mask quantitative measure of emphysema. This effect may be significantly larger for density mask scores in patients with smaller amounts of emphysema.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.