Search

Our study aims to improve the signal-to-noise ratio of positron emission tomography (PET) imaging using conditional unsupervised learning. The proposed method does not require low- and high-quality pairs for network training which can be easily applied to existing PET/computed tomography (CT) and PET/magnetic resonance (MR) datasets. This method consists of two steps: populational training and individual fine-tuning. As for populational training, a network was first pre-trained by a group of patients' noisy PET images and the corresponding anatomical prior images from CT or MR. As for individual fine-tuning, a new network with initial parameters inherited from the pre-trained network was fine-tuned by the test patient's noisy PET image and the corresponding anatomical prior image. Only the last few layers were fine-tuned to take advantage of the populational information and the pre-training efforts. Both networks shared the same structure and took the CT or MR images as the network input so that the network output was conditioned on the patient's anatomic prior information. The noisy PET images were used as the training and fine-tuning labels. The proposed method was evaluated on a 68Ga-PPRGD2 PET/CT dataset and a 18F-FDG PET/MR dataset. For the PET/CT dataset, with the original noisy PET image as the baseline, the proposed method has a significantly higher contrast-to noise ratio (CNR) improvement (71.85% ± 27.05%) than Gaussian (12.66% ± 6.19%, P = 0.002), nonlocal mean method (22.60% ± 13.11%, P = 0.002) and conditional deep image prior method (52.94% ± 21.79%, P = 0.0039). For the PET/MR dataset, compared to Gaussian (18.73% ± 9.98%, P < 0.0001), NLM (26.01% ± 19.40%, P < 0.0001) and CDIP (47.48% ± 25.36%, P < 0.0001), the CNR improvement ratio of the proposed method (58.07% ± 28.45%) is the highest. In addition, the denoised images using both datasets also showed that the proposed method can accurately restore tumor structures while also smoothing out the noise. [ABSTRACT FROM AUTHOR]

Positron emission tomography-magnetic resonance (PET-MR) scanners could improve radiotherapy planning through combining PET and MR functional imaging. This depends on acquiring high quality and quantitatively accurate images in the radiotherapy position. This study evaluated PET-MR image quality using a flat couch and coil bridge for pelvic radiotherapy. MR and PET image quality phantoms were imaged in three setups: phantom on the PET-MR couch with anterior coil on top (diagnostic), phantom on a flat couch with coil on top (couch), and phantom on the flat couch with coil on a coil bridge (radiotherapy). PET images were also acquired in each setup without the anterior coil. PET attenuation correction of the flat couch and coil bridge were generated using kilovoltage computed tomography (CT) images and of the anterior coil using megavoltage CT images. MR image quality was substantially affected, with MR signal to noise ratio (SNR) relative to the diagnostic setup of 89% ± 2% (mean ± standard error of the mean, couch) and 54% ± 1% (radiotherapy), likely due to the increased distance between the patient and receive coils. The reduction impacted the low-contrast detectability score: 23 ± 1 (diagnostic), 19.7 ± 0.3 (couch) and 15 ± 1 (radiotherapy). All other MR metrics agreed within one standard error. PET quantitative accuracy was also affected, with measured activity with anterior coil being different to diagnostic without anterior coil by −16.7% ± 0.2% (couch) and −17.7 ± 0.1% (radiotherapy), without attenuation correction modification. Including the couch and coil bridge attenuation correction reduced this difference to −7.5% ± 0.1%, and including the anterior coil reduced this to −2.7% ± 0.1%. This was better than the diagnostic setup with anterior coil (difference −8.3% ± 0.2%). This translated into greater PET SNR performance for the fully corrected radiotherapy setup compared to diagnostic with coil. However contrast recovery was unchanged by the modified attenuation correction, with the diagnostic setup remaining ∼2% better. Quantitative PET in the radiotherapy setup is possible if appropriate attenuation correction is used. Pelvic radiotherapy PET-MR imaging protocols will need to consider the impact on PET-MR image quality. [ABSTRACT FROM AUTHOR]

Many deep learning (DL)-based image restoration methods for low-dose CT (LDCT) problems directly employ the end-to-end networks on low-dose training data without considering dose differences. However, the radiation dose difference has a great impact on the ultimate results, and lower doses increase the difficulty of restoration. Moreover, there is increasing demand to design and estimate acceptable scanning doses for patients in clinical practice, necessitating dose-aware networks embedded with adaptive dose estimation. In this paper, we consider these dose differences of input LDCT images and propose an adaptive dose-aware network. First, considering a large dose distribution range for simulation convenience, we coarsely define five dose levels in advance as lowest, lower, mild, higher and highest radiation dose levels. Instead of directly building the end-to-end mapping function between LDCT images and high-dose CT counterparts, the dose level is primarily estimated in the first stage. In the second stage, the adaptively learned low-dose level is used to guide the image restoration process as the pattern of prior information through the channel feature transform. We conduct experiments on a simulated dataset based on original high dose parts of American Association of Physicists in Medicine challenge datasets from the Mayo Clinic. Ablation studies validate the effectiveness of the dose-level estimation, and the experimental results show that our method is superior to several other DL-based methods. Specifically, our method provides obviously better performance in terms of the peak signal-to-noise ratio and visual quality reflected in subjective scores. Due to the dual-stage process, our method may suffer limitations under more parameters and coarse dose-level definitions, and thus, further improvements in clinical practical applications with different CT equipment vendors are planned in future work. [ABSTRACT FROM AUTHOR]

Reducing radiation dose is important for PET imaging. However, reducing injection doses causes increased image noise and low signal-to-noise ratio (SNR), subsequently affecting diagnostic and quantitative accuracies. Deep learning methods have shown a great potential to reduce the noise and improve the SNR in low dose PET data. In this work, we comprehensively investigated the quantitative accuracy of small lung nodules, in addition to visual image quality, using deep learning based denoising methods for oncological PET imaging. We applied and optimized an advanced deep learning method based on the U-net architecture to predict the standard dose PET image from 10% low-dose PET data. We also investigated the effect of different network architectures, image dimensions, labels and inputs for deep learning methods with respect to both noise reduction performance and quantitative accuracy. Normalized mean square error (NMSE), SNR, and standard uptake value (SUV) bias of different nodule regions of interest (ROIs) were used for evaluation. Our results showed that U-net and GAN are superior to CAE with smaller SUVmean and SUVmax bias at the expense of inferior SNR. A fully 3D U-net has optimal quantitative performance compared to 2D and 2.5D U-net with less than 15% SUVmean bias for all the ten patients. U-net outperforms Residual U-net (r-U-net) in general with smaller NMSE, higher SNR and lower SUVmax bias. Fully 3D U-net is superior to several existing denoising methods, including Gaussian filter, anatomical-guided non-local mean (NLM) filter, and MAP reconstruction with Quadratic prior and relative difference prior, in terms of superior image quality and trade-off between noise and bias. Furthermore, incorporating aligned CT images has the potential to further improve the quantitative accuracy in multi-channel U-net. We found the optimal architectures and parameters of deep learning based methods are different for absolute quantitative accuracy and visual image quality. Our quantitative results demonstrated that fully 3D U-net can both effectively reduce image noise and control bias even for sub-centimeter small lung nodules when generating standard dose PET using 10% low count down-sampled data. [ABSTRACT FROM AUTHOR]

Recent studies emphasized the role of the bone lacuno-canalicular network (LCN) in the understanding of bone diseases such as osteoporosis. However, suitable methods to investigate this structure are lacking. The aim of this paper is to introduce a methodology to segment the LCN from three-dimensional (3D) synchrotron radiation nano-CT images. Segmentation of such structures is challenging due to several factors such as limited contrast and signal-to-noise ratio, partial volume effects and huge number of data that needs to be processed, which restrains user interaction. We use an approach based on minimum-cost paths and geodesic voting, for which we propose a fully automatic initialization scheme based on a tessellation of the image domain. The centroids of pre-segmented lacunæ are used as Voronoi-tessellation seeds and as start-points of a fast-marching front propagation, whereas the end-points are distributed in the vicinity of each Voronoi-region boundary. This initialization scheme was devised to cope with complex biological structures involving cells interconnected by multiple thread-like, branching processes, while the seminal geodesic-voting method only copes with tree-like structures. Our method has been assessed quantitatively on phantom data and qualitatively on real datasets, demonstrating its feasibility. To the best of our knowledge, presented 3D renderings of lacunæ interconnected by their canaliculi were achieved for the first time. [ABSTRACT FROM AUTHOR]

Three phase-retrieval algorithms, based on the transport-of-intensity equation and on the contrast transfer function for propagation-based imaging, and on the linearized geometrical optics approximation for analyser-based imaging, are investigated. The algorithms are compared in terms of their effect on propagation of noise from projection images to the corresponding phase-retrieved images and further to the computed tomography (CT) images/slices of a monomorphous object reconstructed using filtered backprojection algorithm. The comparison is carried out in terms of an integral noise characteristic, the variance, as well as in terms of a simple figure-of-merit, i.e. signal-to-noise ratio per unit dose. A gain factor is introduced that quantitatively characterizes the effect of phase retrieval on the variance of noise in the reconstructed projection images and in the axial slices of the object. Simple analytical expressions are derived for the gain factor and the signal-to-noise ratio, which indicate that the application of phase-retrieval algorithms can increase these parameters by up to two orders of magnitude compared to raw projection images and conventional CT, thus allowing significant improvement in the image quality and/or reduction of the x-ray dose delivered to the patient. [ABSTRACT FROM AUTHOR]

This study presents an analytical model for the edge spread function (ESF) of a clinical CT system that allows reliable fits of noisy ESF data. The model was used for the calculation of the material-specific transfer function TF and an estimation of the signal transfer and the signal-to-noise ratio (SNR) in 2D. Images of the Catphan phantom were acquired with a clinical Siemens Somatom Sensation Cardiac 64 CT scanner combining four different x-ray tube outputs (40, 150, 250 and 350 mAs) with four different reconstruction filters, which covered the range from very smooth (B10s) to very sharp (B70s). The images of the high- and mid-contrast cylinders of the phantom’s ‘Geometry and Sensitometry’ module (air, Teflon, Delrin and PMP) were used to sample material-specific ESF curves. The ESF curves were fitted with the analytical model we developed based on a linear combination of Boltzmann and Gaussian functions. The analytical model of the ESF was used to obtain the Fourier-based material-specific transfer function TF, as well as the spatial-domain point spread function (PSF). TF was subsequently used to estimate the signal transfer, which was compared to the actual reconstructed image of a 3.0 mm diameter Teflon pin. The noise power spectrum (NPS) was calculated from images of a uniform water phantom under the same technique parameters. The task-specific SNR was calculated for all technique parameters from the model-based TF, the measured NPS and simulated 3 mm diameter disc signals modeling the aforementioned materials. Bootstrapping was performed to estimate the standard deviation of the TF and the SNR. The analytical model we developed accurately captured the features of the CT ESF data. The coefficient of determination R2, a metric that describes the goodness of the fit, had a median value of 0.9995, and decreased for low tube output, low contrast and the sharp reconstruction filter. Our analysis showed that ESF, PSF and TF depended not only on the reconstruction filter, but also on the tube output and the material of the cylinders. For B40s and B70s, the TF of Delrin was significantly higher than the TF of other materials in the frequency range of 0.4–0.9 mm−1. The estimated signal transfer agreed well with the actual reconstructed image of the Teflon pin. For the technique parameters we used the SNR values ranged between [64, 320], [64, 281], [37, 137] and [33, 117] for air, Teflon, Delrin and PMP respectively. While for high-contrast materials the smoothest reconstruction filter resulted in the highest SNR, for mid-contrast materials the standard filter gave the best results. The presented approach provides an accurate, analytical description of the material-specific ESF, PSF and TF as well as an estimate of the signal transfer. The transfer function TF together with the NPS and simulated signals allow the calculation of a task-specific SNR. [ABSTRACT FROM AUTHOR]

Three-dimensional reconstruction of cardiac vasculature from angiographic C-arm CT (rotational angiography) data is a major challenge. Motion artefacts corrupt image quality, reducing usability for diagnosis and guidance. Many state-of-the-art approaches depend on retrospective ECG-gating of projection data for image reconstruction. A trade-off has to be made regarding the size of the ECG-gating window. A large temporal window is desirable to avoid undersampling. However, residual motionwill occur in a large window, causing motion artefacts. We present an algorithm to correct for residual motion. Our approach is based on a deformable 2D-2D registration between the forward projection of an initial, ECG-gated reconstruction, and the original projection data. The approach is fully automatic and does not require any complex segmentation of vasculature, or landmarks. The estimated motion is compensated for during the backprojection step of a subsequent reconstruction. We evaluated the method using the publicly available CAVAREV platform and on six human clinical datasets.We found a better visibility of structure, reduced motion artefacts, and increased sharpness of the vessels in the compensated reconstructions compared to the initial reconstructions. At the time of writing, our algorithm outperforms the leading result of the CAVAREV ranking list. For the clinical datasets, we found an average reduction of motion artefacts by 13 ± 6%. Vessel sharpness was improved by 25 ± 12% on average [ABSTRACT FROM AUTHOR]

The purpose of this research was to study the sensitivity of Computed Tomography (CT) radiomic features to motion blurring and signal-to-noise ratios (SNR), and investigate its downstream effect regarding the classification of non-small cell lung cancer (NSCLC) histology. Forty-three radiomic features were considered and classified into one of four categories: Morphological, Intensity, Fine Texture, and Coarse Texture. First, a series of simulations were used to study feature-sensitivity to changes in spatial-temporal resolution. A dynamic digital phantom was used to generate images with different breathing amplitudes and SNR, from which features were extracted and characterized relative to initial simulation conditions. Stage I NSCLC patients were then retrospectively identified, from which three different acquisition-specific feature-spaces were generated based on free-breathing (FB), average-intensity-projection (AIP), and end-of-exhalation (EOE) CT images. These feature-spaces were derived to cover a wide range of spatial-temporal tradeoff. Normalized percent differences and concordance correlation coefficients (CCC) were used to assess the variability between the 3D and 4D acquisition techniques. Subsequently, three corresponding acquisition-specific logistic regression models were developed to classify lung tumor histology. Classification performance was compared between the different data-dependent models. Simulation results demonstrated strong linear dependences (p  >  0.95) between respiratory motion and morphological features, as well as between SNR and texture features. The feature Short Run Emphasis was found to be particularly stable to both motion blurring and changes in SNR. When comparing FB-to-EOE, 37% of features demonstrated high CCC agreement (CCC  >  0.8), compared to only 30% for FB-to-AIP. In classifying tumor histology, EoE images achieved an average AUC, Accuracy, Sensitivity, and Specificity of , respectively. FB images achieved respective values of , and AIP images achieved respective values of . Several radiomic features have been identified as being relatively robust to spatial-temporal variations based on both simulation data and patient data. In general, features that were sensitive to motion blurring were not necessarily the same features that were sensitive to changes in SNR. Our modeling results suggest that the EoE phase of a 4DCT acquisition may provide useful radiomic information, particularly for features that are highly sensitive to respiratory motion. [ABSTRACT FROM AUTHOR]

The purpose of this study is to assess dose for radiosensitive organs and image quality in head and chest computed tomography (CT) examinations. Our focus was in the brain, eye lens and lung organs using two protocols; one protocol with fixed mAs and filtered back projection (FBP) and another with tube current modulation (TCM) and sinogram affirmed iterative reconstruction (SAFIRE). Measurements were performed on a 128-slice CT scanner by placing thermoluminescent dosimeters (TLDs) in an anthropomorphic adult phantom. Results were compared to a CT-Expo software. Objective image quality was assessed in terms of signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR). SPSS software was used for data analyses. Results showed that, using TCM, doses were reduced by 22.84%–25.06% for brain, by 21.82%–23.48% for eye lens and by 54%–53.22% for lung with TLD and CT-Expo respectively. The increased SNR and CNR values achieved for scans performed with TCM combined with iterative reconstruction techniques were 38.68%–58.81% and 38.91%–43.60% respectively. We conclude that, using TCM, a significant mean organ dose reduction is achieved for brain, eye lens and lung organs. Then, combined with iterative reconstruction, image quality was well maintained in terms of SNR and CNR. Thus it is highly recommended in clinical practice optimization in head and chest CT examinations. [ABSTRACT FROM AUTHOR]

Rapid developments in photon-counting and energy-discriminating detectors have the potential to provide an additional spectral dimension to conventional x-ray grayscale imaging. Reconstructed spectroscopic tomographic data can be used to distinguish individual materials by characteristic absorption peaks. The acquired energy-binned data, however, suffer from low signal-to-noise ratio, acquisition artifacts, and frequently angular undersampled conditions. New regularized iterative reconstruction methods have the potential to produce higher quality images and since energy channels are mutually correlated it can be advantageous to exploit this additional knowledge. In this paper, we propose a novel method which jointly reconstructs all energy channels while imposing a strong structural correlation. The core of the proposed algorithm is to employ a variational framework of parallel level sets to encourage joint smoothing directions. In particular, the method selects reference channels from which to propagate structure in an adaptive and stochastic way while preferring channels with a high data signal-to-noise ratio. The method is compared with current state-of-the-art multi-channel reconstruction techniques including channel-wise total variation and correlative total nuclear variation regularization. Realistic simulation experiments demonstrate the performance improvements achievable by using correlative regularization methods. [ABSTRACT FROM AUTHOR]

In CT imaging, a variety of applications exist which are strongly SNR limited. However, in some cases redundant data of the same body region provide additional quanta.Examples: in dual energy CT, the spatial resolution has to be compromised to provide good SNR for material decomposition. However, the respective spectral dataset of the same body region provides additional quanta which might be utilized to improve SNR of each spectral component. Perfusion CT is a high dose application, and dose reduction is highly desirable. However, a meaningful evaluation of perfusion parameters might be impaired by noisy time frames. On the other hand, the SNR of the average of all time frames is extremely high.In redundant CT acquisitions, multiple image datasets can be reconstructed and averaged to composite image data. These composite image data, however, might be compromised with respect to contrast resolution and/or spatial resolution and/or temporal resolution. These observations bring us to the idea of transferring high SNR of composite image data to low SNR ‘source’ image data, while maintaining their resolution.It has been shown that the noise characteristics of CT image data can be improved by iterative reconstruction (Popescu et al 2012 Book of Abstracts, 2nd CT Meeting (Salt Lake City, UT) p 148). In case of data dependent Gaussian noise it can be modelled with image-based iterative reconstruction at least in an approximate manner (Bruder et al 2011 Proc. SPIE7961 79610J).We present a generalized update equation in image space, consisting of a linear combination of the previous update, a correction term which is constrained by the source image data, and a regularization prior, which is initialized by the composite image data. This iterative reconstruction approach we call bimodal reconstruction (BMR).Based on simulation data it is shown that BMR can improve low contrast detectability, substantially reduces the noise power and has the potential to recover spatial resolution of the source image data.For different CT applications: dual energy imaging, liver imaging, spiral imaging, cardiac imaging, we show that SNR can efficiently be transferred from the composite image to the source image data at constant patient dose, while maintaining resolution properties of the source data. [ABSTRACT FROM AUTHOR]

CT dose optimization is typically guided by pixel noise or contrast-to-noise ratio that does not delineate low contrast details adequately. We utilized the statistically defined low contrast detectability to study its relationship to dose and lesion size in abdominal CT. A realistically shaped medium sized abdomen phantom was customized to contain a cylindrical void of 4 cm diameter. The void was filled with a low contrast (1% and 2%) insert containing six groups of cylindrical targets ranging from 1.2 mm to 7 mm in size. Helical CT scans were performed using a Siemens 64-slice mCT and a GE Discovery 750 HD at various doses. After the subtractions between adjacent slices, the uniform sections of the filtered backprojection reconstructed images were partitioned to matrices of square elements matching the sizes of the targets. It was verified that the mean values from all the elements in each matrix follow a Gaussian distribution. The minimum detectable contrast (MDC), quantified by the mean signal to background difference equal to the distribution’s standard deviation multiplied by 3.29, corresponding to 95% confidence level, was found to be related to the phantom specific dose and the element size by a power law (R^2  >  0.990). Independent readings on the 5 mm and 7 mm targets were compared to the measured contrast to the MDC ratios. The results showed that 93% of the cases were detectable when the measured contrast exceeds the MDC. The correlation of the MDC to the pixel noise and target size was also identified and the relationship was found to be the same for the scanners in the study. To quantify the impact of iterative reconstructions to the low contrast detectability, the noise structure was studied in a similar manner at different doses and with different ASIR blending fractions. The relationship of the dose to the blending fraction and low contrast detectability is presented. [ABSTRACT FROM AUTHOR]

We investigate the improvement from the use of high-Z CdTe sensors for pre-clinical K-edge imaging with the hybrid pixel detectors XPAD3. We compare XPAD3 chips bump bonded to Si or CdTe sensors in identical experimental conditions. Image performance for narrow energy bin acquisitions and contrast-to-noise ratios of K-edge images are presented and compared. CdTe sensors achieve signal-to-noise ratios at least three times higher than Si sensors within narrow energy bins, thanks to their much higher detection efficiency. Nevertheless Si sensors provide better contrast-to-noise ratios in K-edge imaging when working at equivalent counting statistics, due to their better estimation of the attenuation coefficient of the contrast agent. Results are compared to simulated data in the case of the XPAD3/Si detector. Good agreement is observed when including charge sharing between pixels, which have a strong impact on contrast-to-noise ratios in K-edge images. [ABSTRACT FROM AUTHOR]

The state of the art to describe image quality in medical imaging is to assess the performance of an observer conducting a task of clinical interest. This can be done by using a model observer leading to a figure of merit such as the signal-to-noise ratio (SNR). Using the non-prewhitening (NPW) model observer, we objectively characterised the evolution of its figure of merit in various acquisition conditions. The NPW model observer usually requires the use of the modulation transfer function (MTF) as well as noise power spectra. However, although the computation of the MTF poses no problem when dealing with the traditional filtered back-projection (FBP) algorithm, this is not the case when using iterative reconstruction (IR) algorithms, such as adaptive statistical iterative reconstruction (ASIR) or model-based iterative reconstruction (MBIR). Given that the target transfer function (TTF) had already shown it could accurately express the system resolution even with non-linear algorithms, we decided to tune the NPW model observer, replacing the standard MTF by the TTF. It was estimated using a custom-made phantom containing cylindrical inserts surrounded by water. The contrast differences between the inserts and water were plotted for each acquisition condition. Then, mathematical transformations were performed leading to the TTF. As expected, the first results showed a dependency of the image contrast and noise levels on the TTF for both ASIR and MBIR. Moreover, FBP also proved to be dependent of the contrast and noise when using the lung kernel. Those results were then introduced in the NPW model observer. We observed an enhancement of SNR every time we switched from FBP to ASIR to MBIR. IR algorithms greatly improve image quality, especially in low-dose conditions. Based on our results, the use of MBIR could lead to further dose reduction in several clinical applications. [ABSTRACT FROM AUTHOR]