Automatic chest computed tomography image noise quantification using deep learning.

PURPOSE: This study aimed to develop a deep learning (DL) method for noise quantification for clinical chest computed tomography (CT) images without the need for repeated scanning or homogeneous tissue regions. METHODS: A comprehensive phantom CT dataset (three dose levels, six reconstruction methods, amounting to 9240 slices) was acquired and used to train a convolutional neural network (CNN) to output an estimate of local image noise standard deviations (SD) from a single CT scan input. The CNN model consisting of seven convolutional layers was trained on the phantom image dataset representing a range of scan parameters and was tested with phantom images acquired in a variety of different scan conditions, as well as publicly available chest CT images to produce clinical noise SD maps. RESULTS: Noise SD maps predicted by the CNN agreed well with the ground truth both visually and numerically in the phantom dataset (errors of < 5 HU for most scan parameter combinations). In addition, the noise SD estimates obtained from clinical chest CT images were similar to running-average based reference estimates in areas without prominent tissue interfaces. CONCLUSIONS: Predicting local noise magnitudes without the need for repeated scans is feasible using DL. Our implementation trained with phantom data was successfully applied to open-source clinical data with heterogeneous tissue borders and textures. We suggest that automatic DL noise mapping from clinical patient images could be used as a tool for objective CT image quality estimation and protocol optimization.

Adenosine-induced splenic switch-off on [15O]H2O PET perfusion for the assessment of vascular vasodilatation.

BACKGROUND: Splenic switch-off (SSO) is a marker of adequate adenosine-induced vasodilatation on cardiac magnetic resonance perfusion imaging. We evaluate the feasibility of quantitative assessment of SSO in myocardial positron emission tomography (PET) perfusion imaging using [15O]H2O. METHODS: Thirty patients underwent [15O]H2O PET perfusion with adenosine stress. Time-activity curves, as averaged standardized uptake values (SUVavg), were extracted from dynamic PET for spleen and liver. Maximum SUVavg, stress and rest spleen-to-liver ratio (SLR), and the splenic activity concentration ratio (SAR) were computed. Optimal cut-off values for SSO assessment were estimated from receiver operating characteristics (ROC) curve for maximum SUVavg and SLR. Also, differences between coronary artery disease, myocardial ischemia, beta-blockers, and diabetes were assessed. Data are presented as median [interquartile range]. RESULTS: In concordance with the SSO phenomenon, both the spleen maximum SUVavg and SLR were lower in adenosine stress when compared to rest perfusion (8.1 [6.5, 9.2] versus 16.4 [13.4, 19.0], p < 0.001) and (0.81 [0.63, 1.08] versus 1.86 [1.73, 2.06], p < 0.001), respectively. During adenosine stress, the SSO effect was most prominent 40-160 s after radiotracer injection. Cut-off values of 12.6 and 1.57 for maximum SUVavg and SLR, respectively, were found based on ROC analysis. No differences in SAR, SLRRest, or SLRStress were observed in patients with coronary artery disease, myocardial ischemia, or diabetes. CONCLUSIONS: SSO can be quantified from [15O]H2O PET perfusion and used as a marker for adequate adenosine-induced vasodilatation response. In contrary to other PET perfusion tracers, adenosine-induced SSO is time dependent with [15O]H2O.

Deep-Learning-Based Contrast Synthesis From MRF Parameter Maps in the Knee Joint.

BACKGROUND: Magnetic resonance fingerprinting (MRF) is a method to speed up acquisition of quantitative MRI data. However, MRF does not usually produce contrast-weighted images that are required by radiologists, limiting reachable total scan time improvement. Contrast synthesis from MRF could significantly decrease the imaging time. PURPOSE: To improve clinical utility of MRF by synthesizing contrast-weighted MR images from the quantitative data provided by MRF, using U-nets that were trained for the synthesis task utilizing L1- and perceptual loss functions, and their combinations. STUDY TYPE: Retrospective. POPULATION: Knee joint MRI data from 184 subjects from Northern Finland 1986 Birth Cohort (ages 33-35, gender distribution not available). FIELD STRENGTH AND SEQUENCE: A 3 T, multislice-MRF, proton density (PD)-weighted 3D-SPACE (sampling perfection with application optimized contrasts using different flip angle evolution), fat-saturated T2-weighted 3D-space, water-excited double echo steady state (DESS). ASSESSMENT: Data were divided into training, validation, test, and radiologist's assessment sets in the following way: 136 subjects to training, 3 for validation, 3 for testing, and 42 for radiologist's assessment. The synthetic and target images were evaluated using 5-point Likert scale by two musculoskeletal radiologists blinded and with quantitative error metrics. STATISTICAL TESTS: Friedman's test accompanied with post hoc Wilcoxon signed-rank test and intraclass correlation coefficient. The statistical cutoff P <0.05 adjusted by Bonferroni correction as necessary was utilized. RESULTS: The networks trained in the study could synthesize conventional images with high image quality (Likert scores 3-4 on a 5-point scale). Qualitatively, the best synthetic images were produced with combination of L1- and perceptual loss functions and perceptual loss alone, while L1-loss alone led to significantly poorer image quality (Likert scores below 3). The interreader and intrareader agreement were high (0.80 and 0.92, respectively) and significant. However, quantitative image quality metrics indicated best performance for the pure L1-loss. DATA CONCLUSION: Synthesizing high-quality contrast-weighted images from MRF data using deep learning is feasible. However, more studies are needed to validate the diagnostic accuracy of these synthetic images. EVIDENCE LEVEL: 4. TECHNICAL EFFICACY: Stage 1.

Automatic head computed tomography image noise quantification with deep learning.

PURPOSE: Computed tomography (CT) image noise is usually determined by standard deviation (SD) of pixel values from uniform image regions. This study investigates how deep learning (DL) could be applied in head CT image noise estimation. METHODS: Two approaches were investigated for noise image estimation of a single acquisition image: direct noise image estimation using supervised DnCNN convolutional neural network (CNN) architecture, and subtraction of a denoised image estimated with denoising UNet-CNN experimented with supervised and unsupervised noise2noise training approaches. Noise was assessed with local SD maps using 3D- and 2D-CNN architectures. Anthropomorphic phantom CT image dataset (N = 9 scans, 3 repetitions) was used for DL-model comparisons. Mean square error (MSE) and mean absolute percentage errors (MAPE) of SD values were determined using the SD values of subtraction images as ground truth. Open-source clinical head CT low-dose dataset (Ntrain = 37, Ntest = 10 subjects) were used to demonstrate DL applicability in noise estimation from manually labeled uniform regions and in automated noise and contrast assessment. RESULTS: The direct SD estimation using 3D-CNN was the most accurate assessment method when comparing in phantom dataset (MAPE = 15.5%, MSE = 6.3HU). Unsupervised noise2noise approach provided only slightly inferior results (MAPE = 20.2%, MSE = 13.7HU). 2DCNN and unsupervised UNet models provided the smallest MSE on clinical labeled uniform regions. CONCLUSIONS: DL-based clinical image assessment is feasible and provides acceptable accuracy as compared to true image noise. Noise2noise approach may be feasible in clinical use where no ground truth data is available. Noise estimation combined with tissue segmentation may enable more comprehensive image quality characterization.

Virtual monochromatic imaging reduces beam hardening artefacts in cardiac interior photon counting computed tomography: a phantom study with cadaveric specimens.

In interior cardiac computed tomography (CT) imaging, the x-ray beam is collimated to a limited field-of-view covering the heart volume, which decreases the radiation exposure to surrounding tissues. Spectral CT enables the creation of virtual monochromatic images (VMIs) through a computational material decomposition process. This study investigates the utility of VMIs for beam hardening (BH) reduction in interior cardiac CT, and further, the suitability of VMIs for coronary artery calcium (CAC) scoring and volume assessment is studied using spectral photon counting detector CT (PCD-CT).Ex vivocoronary artery samples (N = 18) were inserted in an epoxy rod phantom. The rod was scanned in the conventional CT geometry, and subsequently, the rod was positioned in a torso phantom and re-measured in the interior PCD-CT geometry. The total energy (TE) 10-100 keV reconstructions from PCD-CT were used as a reference. The low energy 10-60 keV and high energy 60-100 keV data were used to perform projection domain material decomposition to polymethyl methacrylate and calcium hydroxylapatite basis. The truncated basis-material sinograms were extended using the adaptive detruncation method. VMIs from 30-180 keV range were computed from the detruncated virtual monochromatic sinograms using filtered back projection. Detrending was applied as a post-processing method prior to CAC scoring. The results showed that BH artefacts from the exterior structures can be suppressed with high (≥100 keV) VMIs. With appropriate selection of the monoenergy (46 keV), the underestimation trend of CAC scores and volumes shown in Bland-Altman (BA) plots for TE interior PCD-CT was mitigated, as the BA slope values were -0.02 for the 46 keV VMI compared to -0.21 the conventional TE image. To conclude, spectral PCD-CT imaging using VMIs could be applied to reduce BH artefacts interior CT geometry, and further, optimal selection of VMI may improve the accuracy of CAC scoring assessment in interior PCD-CT.

Generative adversarial networks improve interior computed tomography angiography reconstruction.

In interior computed tomography (CT), the x-ray beam is collimated to a limited field-of-view (FOV) (e.g. the volume of the heart) to decrease exposure to adjacent organs, but the resulting image has a severe truncation artifact when reconstructed with traditional filtered back-projection (FBP) type algorithms. In some examinations, such as cardiac or dentomaxillofacial imaging, interior CT could be used to achieve further dose reductions. In this work, we describe a deep learning (DL) method to obtain artifact-free images from interior CT angiography. Our method employs the Pix2Pix generative adversarial network (GAN) in a two-stage process: (1) An extended sinogram is computed from a truncated sinogram with one GAN model, and (2) the FBP reconstruction obtained from that extended sinogram is used as an input to another GAN model that improves the quality of the interior reconstruction. Our double GAN (DGAN) model was trained with 10 000 truncated sinograms simulated from real computed tomography angiography slice images. Truncated sinograms (input) were used with original slice images (target) in training to yield an improved reconstruction (output). DGAN performance was compared with the adaptive de-truncation method, total variation regularization, and two reference DL methods: FBPConvNet, and U-Net-based sinogram extension (ES-UNet). Our DGAN method and ES-UNet yielded the best root-mean-squared error (RMSE) (0.03 ± 0.01), and structural similarity index (SSIM) (0.92 ± 0.02) values, and reference DL methods also yielded good results. Furthermore, we performed an extended FOV analysis by increasing the reconstruction area by 10% and 20%. In both cases, the DGAN approach yielded best results at RMSE (0.03 ± 0.01 and 0.04 ± 0.01 for the 10% and 20% cases, respectively), peak signal-to-noise ratio (PSNR) (30.5 ± 2.6 dB and 28.6 ± 2.6 dB), and SSIM (0.90 ± 0.02 and 0.87 ± 0.02). In conclusion, our method was able to not only reconstruct the interior region with improved image quality, but also extend the reconstructed FOV by 20%.

Quantitative dual contrast photon-counting computed tomography for assessment of articular cartilage health.

Photon-counting detector computed tomography (PCD-CT) is a modern spectral imaging technique utilizing photon-counting detectors (PCDs). PCDs detect individual photons and classify them into fixed energy bins, thus enabling energy selective imaging, contrary to energy integrating detectors that detects and sums the total energy from all photons during acquisition. The structure and composition of the articular cartilage cannot be detected with native CT imaging but can be assessed using contrast-enhancement. Spectral imaging allows simultaneous decomposition of multiple contrast agents, which can be used to target and highlight discrete cartilage properties. Here we report, for the first time, the use of PCD-CT to quantify a cationic iodinated CA4+ (targeting proteoglycans) and a non-ionic gadolinium-based gadoteridol (reflecting water content) contrast agents inside human osteochondral tissue (n = 53). We performed PCD-CT scanning at diffusion equilibrium and compared the results against reference data of biomechanical and optical density measurements, and Mankin scoring. PCD-CT enables simultaneous quantification of the two contrast agent concentrations inside cartilage and the results correlate with the structural and functional reference parameters. With improved soft tissue contrast and assessment of proteoglycan and water contents, PCD-CT with the dual contrast agent method is of potential use for the detection and monitoring of osteoarthritis.

Optimizing iterative reconstruction for quantification of calcium hydroxyapatite with photon counting flat-detector computed tomography: a cardiac phantom study.

Purpose: Coronary artery calcium (CAC) scoring with computed tomography (CT) has been proposed as a screening tool for coronary artery disease, but concerns remain regarding the radiation dose of CT CAC scoring. Photon counting detectors and iterative reconstruction (IR) are promising approaches for patient dose reduction, yet the preservation of CAC scores with IR has been questioned. The purpose of this study was to investigate the applicability of IR for quantification of CAC using a photon counting flat-detector. Approach: We imaged a cardiac rod phantom with calcium hydroxyapatite (CaHA) inserts with different noise levels using an experimental photon counting flat-detector CT setup to simulate the clinical CAC scoring protocol. We applied filtered back projection (FBP) and two IR algorithms with different regularization strengths. We compared the air kerma values, image quality parameters [noise magnitude, noise power spectrum, modulation transfer function (MTF), and contrast-to-noise ratio], and CaHA quantification accuracy between FBP and IR. Results: IR regularization strength influenced CAC scores significantly ( p < 0.05 ). The CAC volumes and scores between FBP and IRs were the most similar when the IR regularization strength was chosen to match the MTF of the FBP reconstruction. Conclusion: When the regularization strength is selected to produce comparable spatial resolution with FBP, IR can yield comparable CAC scores and volumes with FBP. Nonetheless, at the lowest radiation dose setting, FBP produced more accurate CAC volumes and scores compared to IR, and no improved CAC scoring accuracy at low dose was demonstrated with the utilized IR methods.

T2 -weighted magnetic resonance imaging texture as predictor of low back pain: A texture analysis-based classification pipeline to symptomatic and asymptomatic cases.

Low back pain is a very common symptom and the leading cause of disability throughout the world. Several degenerative imaging findings seen on magnetic resonance imaging are associated with low back pain but none of them is specific for the presence of low back pain as abnormal findings are prevalent among asymptomatic subjects as well. The purpose of this population-based study was to investigate if more specific magnetic resonance imaging predictors of low back pain could be found via texture analysis and machine learning. We used this methodology to classify T2 -weighted magnetic resonance images from the Northern Finland Birth Cohort 1966 data to symptomatic and asymptomatic groups. Lumbar spine magnetic resonance imaging was performed using a fast spin-echo sequence at 1.5 T. Texture analysis pipeline consisting of textural feature extraction, principal component analysis, and logistic regression classifier was applied to the data to classify them into symptomatic (clinically relevant pain with frequency ≥30 days and intensity ≥6/10) and asymptomatic (frequency ≤7 days, intensity ≤3/10, and no previous pain episodes in the follow-up period) groups. Best classification results were observed applying texture analysis to the two lowest intervertebral discs (L4-L5 and L5-S1), with accuracy of 83%, specificity of 83%, sensitivity of 82%, negative predictive value of 94%, precision of 56%, and receiver operating characteristic area-under-curve of 0.91. To conclude, textural features from T2 -weighted magnetic resonance images can be applied in low back pain classification.

Interior photon counting computed tomography for quantification of coronary artery calcium: pre-clinical phantom study.

Computed tomography (CT) is the reference method for cardiac imaging, but concerns have been raised regarding the radiation dose of CT examinations. Recently, photon counting detectors (PCDs) and interior tomography, in which the radiation beam is limited to the organ-of-interest, have been suggested for patient dose reduction. In this study, we investigated interior PCD-CT (iPCD-CT) for non-enhanced quantification of coronary artery calcium (CAC) using an anthropomorphic torso phantom and ex vivo coronary artery samples. We reconstructed the iPCD-CT measurements with filtered back projection (FBP), iterative total variation (TV) regularization, padded FBP, and adaptively detruncated FBP and adaptively detruncated TV. We compared the organ doses between conventional CT and iPCD-CT geometries, assessed the truncation and cupping artifacts with iPCD-CT, and evaluated the CAC quantification performance of iPCD-CT. With approximately the same effective dose between conventional CT geometry (0.30 mSv) and interior PCD-CT with 10.2 cm field-of-view (0.27 mSv), the organ dose of the heart was increased by 52.3% with interior PCD-CT when compared to CT. Conversely, the organ doses to peripheral and radiosensitive organs, such as the stomach (55.0% reduction), were often reduced with interior PCD-CT. FBP and TV did not sufficiently reduce the truncation artifact, whereas padded FBP and adaptively detruncated FBP and TV yielded satisfactory truncation artifact reduction. Notably, the adaptive detruncation algorithm reduced truncation artifacts effectively when it was combined with reconstruction detrending. With this approach, the CAC quantification accuracy was good, and the coronary artery disease grade reclassification rate was particularly low (5.6%). Thus, our results confirm that CAC quantification can be performed with the interior CT geometry, that the artifacts are effectively reduced with suitable interior reconstruction methods, and that interior tomography provides efficient patient dose reduction.

Framework for Photon Counting Quantitative Material Decomposition.

In this paper, the accuracy of material decomposition (MD) using an energy discriminating photon counting detector was studied. An MD framework was established and validated using calcium hydroxyapatite (CaHA) inserts of known densities (50 mg/cm3, 100 mg/cm3, 250 mg/cm3, 400 mg/cm3), and diameters (1.2, 3.0, and 5.0 mm). These inserts were placed in a cardiac rod phantom that mimics a tissue equivalent heart and measured using an experimental photon counting detector cone beam computed tomography (PCD-CBCT) setup. The quantitative coronary calcium scores (density, mass, and volume) obtained from the MD framework were compared with the nominal values. In addition, three different calibration techniques, signal-to-equivalent thickness calibration (STC), polynomial correction (PC), and projected equivalent thickness calibration (PETC) were compared to investigate the effect of the calibration method on the quantitative values. The obtained MD estimates agreed well with the nominal values for density (mass) with mean absolute percent errors (MAPEs) 8 ± 11% (9 ± 15%) and 4 ± 6% (9 ± 14%) for STC and PETC calibration methods, respectively. PC displayed large MAPEs for density (27 ± 9%), and mass (25 ± 12%). Volume estimation resulted in large deviations between true and measured values with notable MAPEs for STC (40 ± 90%), PC (40 ± 80%), and PETC (40 ± 90%). The framework demonstrated the feasibility of quantitative CaHA mass and density scoring using PCD-CBCT.

Regular chondrocyte spacing is a potential cause for coherent ultrasound backscatter in human articular cartilage.

The potential of quantitative ultrasound (QUS) to assess the regular cellular spacing in the superficial cartilage zones was investigated experimentally and numerically. Nine osteochondral samples, extracted from two human cadaver knee joints, were measured using a 50-MHz ultrasound scanning device and evaluated using Mankin score. Simulated backscattered power spectra from models with an idealized cell alignment exhibited a pronounced frequency peak. From the peak, cell spacing in the range between 15 and 40 µm between cell layers was detected with an average error of 0.2 µm. The mean QUS-based cell spacing was 28.3 ± 5.3 µm. Strong correlation (R2 = 0.59, p ≤ 0.001) between spacing estimates from light microscopy (LM) and QUS was found for samples with Mankin score ≤3. For higher scores, QUS-based spacing was significantly higher (p ≤ 0.05) compared to LM-based spacing. QUS-based spacing estimates together with other QUS parameters may serve as future biomarkers for detecting early signs of osteoarthrosis.

Differences in acoustic impedance of fresh and embedded human trabecular bone samples-Scanning acoustic microscopy and numerical evaluation.

Trabecular bone samples are traditionally embedded and polished for scanning acoustic microscopy (SAM). The effect of sample processing, including dehydration, on the acoustic impedance of bone is unknown. In this study, acoustic impedance of human trabecular bone samples (n = 8) was experimentally assessed before (fresh) and after embedding using SAM and two-dimensional (2-D) finite-difference time domain simulations. Fresh samples were polished with sandpapers of different grit (P1000, P2500, and P4000). Experimental results indicated that acoustic impedance of samples increased significantly after embedding [mean values 3.7 MRayl (fresh), 6.1 MRayl (embedded), p < 0.001]. After polishing with different papers, no significant changes in acoustic impedance were found, even though higher mean values were detected after polishing with finer (P2500 and P4000) papers. A linear correlation (r = 0.854, p < 0.05) was found between the acoustic impedance values of embedded and fresh bone samples polished using P2500 SiC paper. In numerical simulations dehydration increased the acoustic impedance of trabecular bone (38%), whereas changes in surface roughness of bone had a minor effect on the acoustic impedance (-1.56%/0.1 µm). Thereby, the numerical simulations corroborated the experimental findings. In conclusion, acoustic impedance measurement of fresh trabecular bone is possible and may provide realistic material values similar to those of living bone.

Finite difference time domain model of ultrasound propagation in agarose scaffold containing collagen or chondrocytes.

Measurement of ultrasound backscattering is a promising diagnostic technique for arthroscopic evaluation of articular cartilage. However, contribution of collagen and chondrocytes on ultrasound backscattering and speed of sound in cartilage is not fully understood and is experimentally difficult to study. Agarose hydrogels have been used in tissue engineering applications of cartilage. Therefore, the aim of this study was to simulate the propagation of high frequency ultrasound (40 MHz) in agarose scaffolds with varying concentrations of chondrocytes (1 to 32 × 10(6) cells/ml) and collagen (1.56-200 mg/ml) using transversely isotropic two-dimensional finite difference time domain method (FDTD). Backscatter and speed of sound were evaluated from the simulated pulse-echo and through transmission measurements, respectively. Ultrasound backscatter increased with increasing collagen and chondrocyte concentrations. Furthermore, speed of sound increased with increasing collagen concentration. However, this was not observed with increasing chondrocyte concentrations. The present study suggests that the FDTD method may have some applicability in simulations of ultrasound scattering and propagation in constructs containing collagen and chondrocytes. Findings of this study indicate the significant role of collagen and chondrocytes as ultrasound scatterers and can aid in development of modeling approaches for understanding how cartilage architecture affects to the propagation of high frequency ultrasound.

Species-Independent Modeling of High-Frequency Ultrasound Backscatter in Hyaline Cartilage.

Apparent integrated backscatter (AIB) is a common ultrasound parameter used to assess cartilage matrix degeneration. However, the specific contributions of chondrocytes, proteoglycan and collagen to AIB remain unknown. To reveal these relationships, this work examined biopsies and cross sections of human, ovine and bovine cartilage with 40-MHz ultrasound biomicroscopy. Site-matched estimates of collagen concentration, proteoglycan concentration, collagen orientation and cell number density were employed in quasi-least-squares linear regression analyses to model AIB. A positive correlation (R(2) = 0.51, p < 10(-4)) between AIB and a combination model of cell number density and collagen concentration was obtained for collagen orientations approximately perpendicular (>70°) to the sound beam direction. These findings indicate causal relationships between AIB and cartilage structural parameters and could aid in more sophisticated future interpretations of ultrasound backscatter.

Ultrasound Backscattering Is Anisotropic in Bovine Articular Cartilage.

Collagen, proteoglycans and chondrocytes can contribute to ultrasound scattering in articular cartilage. However, anisotropy of ultrasound scattering in cartilage is not fully characterized. We investigate this using a clinical intravascular ultrasound device with ultrasound frequencies of 9 and 40 MHz. Osteochondral samples were obtained from intact bovine patellas, and cartilage was imaged in two perpendicular directions: through articular and lateral surfaces. At both frequencies, ultrasound backscattering was higher (p < 0.05) when measured through the lateral surface of cartilage. In addition, the composition and structure of articular cartilage were investigated with multiple reference methods involving light microscopy, digital densitometry, polarized light microscopy and Fourier infrared imaging. Reference methods indicated that acoustic anisotropy of ultrasound scattering arises mainly from non-uniform distribution of chondrocytes and anisotropic orientation of collagen fibers. To conclude, ultrasound backscattering in articular cartilage was found to be anisotropic and dependent on the frequency in use.