Reconstruction Kernel Research Articles

Improving Generalization Capability of Multiorgan Segmentation Models Using Dual-Energy CT

Multiorgan segmentation in computed tomography (CT) images is essential for a variety of clinical applications. Due to variations in acquisition protocols, clinical data differ in terms of soft-tissue contrast, noise, and artifacts. Devising automatic multiorgan segmentation approaches, which are generalized to data acquired using different CT protocols, are challenging and essential when conducting any multicenter/scanner analyses. In this study, we investigate the use of dual-energy CT (DECT) images to train a fully convolutional segmentation network which is generalized to CT images acquired using different protocols (i.e., at different energy levels and using a variety of reconstruction kernels) from different CT scanners. Furthermore, a novel image fusion approach in the frequency domain is proposed and compared to state-of-the-art fusion approaches, in terms of the segmentation quality achieved by the network. Overall, the experiments indicate that the generalization capability of the segmentation network is improved using DECT image fusion. The proposed fusion method outperforms all single-energy CT approaches. It provided a significant improvement in segmentation accuracy, ranging from 16.0% to 23.35% with <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$p \leq 0.03$ </tex-math></inline-formula> . Furthermore, two image fusion methods statistically significantly improve segmentation quality in the abdominal organs compared to simply using all available DECT data.

Convolutional Neural Network Addresses the Confounding Impact of CT Reconstruction Kernels on Radiomics Studies.

Achieving high feature reproducibility while preserving biological information is one of the main challenges for the generalizability of current radiomics studies. Non-clinical imaging variables, such as reconstruction kernels, have shown to significantly impact radiomics features. In this study, we retrain an open-source convolutional neural network (CNN) to harmonize computerized tomography (CT) images with various reconstruction kernels to improve feature reproducibility and radiomic model performance using epidermal growth factor receptor (EGFR) mutation prediction in lung cancer as a paradigm. In the training phase, the CNN was retrained and tested on 32 lung cancer patients’ CT images between two different groups of reconstruction kernels (smooth and sharp). In the validation phase, the retrained CNN was validated on an external cohort of 223 lung cancer patients’ CT images acquired using different CT scanners and kernels. The results showed that the retrained CNN could be successfully applied to external datasets with different CT scanner parameters, and harmonization of reconstruction kernels from sharp to smooth could significantly improve the performance of radiomics model in predicting EGFR mutation status in lung cancer. In conclusion, the CNN based method showed great potential in improving feature reproducibility and generalizability by harmonizing medical images with heterogeneous reconstruction kernels.

OL93 - CT imaging texture analysis: evaluation of the effect of reconstruction algorithms and kernels by different vendors

OL93 - CT imaging texture analysis: evaluation of the effect of reconstruction algorithms and kernels by different vendors

EFFECTS OF SMOOTH, MEDIUM SMOOTH AND MEDIUM RECONSTRUCTION KERNELS ON IMAGE QUALITY IN THREE-PHASE CT OF LIVER

Reconstruction kernel is one of the parameters that affects the computed tomography (CT) image quality. This study aimed to evaluate the effects of applying three different reconstruction kernels on image quality in 3-phased CT of the liver. A total of 63 CT liver images including normal liver (n = 43) and liver lesion (n = 20) were retrospectively reviewed. Smooth (B20f), medium smooth (B30f) and medium (B40f) reconstruction kernels were employed in the image reconstruction process. Mean attenuation, image noise, and signal-to-noise ratio (SNR) values from each kernel reconstruction were quantified and compared among those kernels using One Way Analysis of Variance (ANOVA) statistical analysis. Significant changes in image noise and SNR were observed in the normal liver (p &lt; 0.001, respectively) following the application of those reconstruction kernels. However, no significant changes in mean attenuation, image noise, and SNR were demonstrated in the liver lesion (p &gt; 0.05). Application of smooth (B20f), medium smooth (B30f), and medium (B40f) kernel reconstructions would significantly affect the image noise and SNR in the normal liver of CT images instead of liver lesions. Hence, proper selection of reconstruction kernel is important in CT images reconstruction to improve precision in diagnostic CT interpretation.

Deep Learning-Based Image Conversion Improves the Reproducibility of Computed Tomography Radiomics Features: A Phantom Study.

This study aimed to evaluate the usefulness of deep learning-based image conversion to improve the reproducibility of computed tomography (CT) radiomics features. This study was conducted using an abdominal phantom with liver nodules. We developed an image conversion algorithm using a residual feature aggregation network to reproduce radiomics features with CT images under various CT protocols and reconstruction kernels. External validation was performed using images from different scanners, consisting of 8 different protocols. To evaluate the variability of radiomics features, regions of interest (ROIs) were drawn by targeting the liver parenchyma, vessels, paraspinal area, and liver nodules. We extracted 18 first-order, 68 second-order, and 688 wavelet radiomics features. Measurement variability was assessed using the concordance correlation coefficient (CCC), compared with the ground-truth image. In the ROI-based analysis, there was an 83.3% improvement of CCC (80/96; 4 ROIs with 3 categories of radiomics features and 8 protocols) in synthetic images compared with the original images. Among them, the 56 CCC pairs showed a significant increase after image synthesis. In the radiomics feature-based analysis, 62.0% (3838 of 6192; 774 radiomics features with 8 protocols) features showed increased CCC after image synthesis, and a significant increase was noted in 26.9% (1663 of 6192) features. In particular, the first-order feature (79.9%, 115/144) showed better improvement in terms of the reproducibility of radiomics feature than the second-order (59.9%, 326/544) or wavelet feature (61.7%, 3397/5504). Our study demonstrated that a deep learning model for image conversion can improve the reproducibility of radiomics features across various CT protocols, reconstruction kernels, and CT scanners.

Development, validation, and relevance of in vivo low-contrast task transfer function to estimate detectability in clinical CT images.

The current state-of-the-art calculation of detectability index (d') is largely phantom-based, with the latest being based on a hybrid phantom noise power spectrum (NPS) combined with patient-specific noise magnitude and high-contrast air-skin interface. The purpose of this study was to develop and assess the use of fully patient-specific measurements of noise and low-contrast resolution, derived entirely from patient images on d'. This study developed a d' calculation that is patient- and task-specific, employing newly developed algorithms for estimating patient-specific NPS and low-contrast task transfer function (TTF). The TTF estimation methodology used a trained regression support vector machine (SVM) to estimate a fitted form of the TTF given a variance-normalized estimate of the NPS (referred to as the TTFNPS ). The regression SVM was trained and tested using five-fold cross-validation on 192 scans (4 dose levels x 6 reconstruction kernels x 4 repeats) of a phantom with low-contrast polyethylene insert and reconstructed with filtered backprojection and iterative reconstructions across 12 clinically relevant kernels (FBP: B20f, B31f, B45f; SAFIRE: I26f, I31f, J45f with strengths: 2, 3, 5). To test the low-contrast TTF estimation method, the estimated TTFNPS measurements were compared to (1) TTF measurements from the air-phantom interface (referred to as the TTFair , representing the most patient-specific clinical alternative) and (2) TTF measurements from the edge of the low-contrast polyethylene insert (referred to as the TTFpoly ), which represented the gold standard of low-contrast TTF measurement. Patient-specific NPS, patient-specific noise magnitude, and patient-specific low-contrast TTF were further combined with a reference task function to calculate a d' (according to a non-prewhitening matched filter model) across 1120 lesions previously evaluated in 2AFC human observer detection of liver lesions. The resulting values were compared to the observer results using a generalized linear mixed-effects statistical model. The correlations between the model and observer results were also compared with previously reported values (using a hybrid method with phantom-derived NPS and TTFair ). The TTFNPS more accurately represented resolution across the considered reconstruction settings, compared with the TTFair . The out-of-fold predictions of the TTFNPS had statistically better root-mean-square error concordance (p<0.05, one-tailed Wilcoxon rank-sum test) to gold standard than the TTFair (the alternative, measured from the air-phantom interface). Detectability indices informed by purely patient-specific NPS and TTF were strongly correlated with 2AFC outcomes (p<0.05). R2 between human detection accuracy and model-predicted detection accuracy were shown to be greater for those measured with patient-specific d' than for the hybrid d' but failed to rise to the level of statistical significance (p≥0.05, bootstrap resampled corrected paired Student's t-test). The results suggest that fully patient-specific characterization of image quality based on in vivo NPS and low-contrast TTF offer advantages over hybrid methods. The results in terms of d' favorably relate to observer detection of liver lesions. The method can potentially be integrated into an automated image quality tracking system to assess image quality across a computed tomography clinical operation without needing phantom scans.

Repeatability and Reproducibility of Computed Tomography Radiomics for Pulmonary Nodules: A Multicenter Phantom Study.

BackgroundRadiomics can yield minable information from medical images, which can facilitate computer-aided diagnosis. However, the lack of repeatability and reproducibility of radiomic features (RFs) may hinder their generalizability in clinical applications.ObjectivesThe aims of this study were to explore 3 main sources of variability in RFs, investigate their influencing magnitudes and patterns, and identify a subset of robust RFs for further studies.Materials and MethodsA chest phantom with nodules was scanned with different computed tomography (CT) scanners repeatedly with varying acquisition and reconstruction parameters (April-May 2019) to evaluate 3 sources of variability: test-retest, inter-CT, and intra-CT protocol variability. The robustness of the RFs was measured using the concordance correlation coefficient, dynamic range, and intraclass correlation coefficient (ICC). The influencing magnitudes and patterns were analyzed using the Friedman test and Spearman rank correlation coefficient. Stable and informative RFs were selected, and their redundancy was eliminated using hierarchical clustering. Clinical validation was also performed to verify the clinical effectiveness and potential enhancement of the generalizability of radiomics research.ResultsA total of 1295 RFs that showed all 3 sources of variability were included. The reconstruction kernel and the iteration level showed the greatest (ICC, 0.35 ± 0.31) and the least (ICC, 0.63 ± 0.27) influence on magnitudes. The different sources of variability showed relatively consistent patterns of influence (false discovery rate <0.001). Finally, we obtained a subset of 19 stable, informative, and nonredundant RFs under all 3 sources of variability. These RFs exhibited clinical effectiveness and showed better prediction performance than unstable RFs in the validation dataset (P = 0.017, Delong test).ConclusionsThe stability of RFs was affected to different degrees by test-retest and differences in CT manufacturers and models and CT acquisition and reconstruction parameters, but the influences of these factors showed relatively consistent patterns. We also obtained a subset of 19 stable, informative, and nonredundant RFs that should be preferably used to enhance the generalizability of further radiomics research.

Influence of reconstruction kernels on the accuracy of CT-derived fractional flow reserve

ObjectivesWe evaluated the influence of image reconstruction kernels on the diagnostic accuracy of CT-derived fractional flow reserve (FFRCT) compared to invasive FFR in patients with coronary artery disease.MethodsSixty-nine patients, in whom coronary CT angiography was performed and who were further referred for invasive coronary angiography with FFR measurement via pressure wire, were retrospectively included. CT data sets were acquired using a third-generation dual-source CT system and rendered with medium smooth (Bv40) and sharp (Bv49) reconstruction kernels. FFRCT was calculated on-site using prototype software. Coronary stenoses with invasive FFR ≤ 0.80 were classified as significant. Agreement between FFRCT and invasive FFR was determined for both reconstruction kernels.ResultsOne hundred analyzed vessels in 69 patients were included. Twenty-five vessels were significantly stenosed according to invasive FFR. Using a sharp reconstruction kernel for FFRCT resulted in a significantly higher correlation with invasive FFR (r = 0.74, p < 0.01 vs. r = 0.58, p < 0.01; p = 0.04) and a higher AUC in ROC curve analysis to correctly identify/exclude significant stenosis (AUC = 0.92 vs. AUC = 0.82 for sharp vs. medium smooth kernel, respectively, p = 0.02). A FFRCT value of ≤ 0.8 using a sharp reconstruction kernel showed a sensitivity of 88% and a specificity of 92% for detecting ischemia-causing lesions, resulting in a diagnostic accuracy of 91%. The medium smooth reconstruction kernel performed worse (sensitivity 60%, specificity 89%, accuracy 82%).ConclusionCompared to invasively measured FFR, FFRCT using a sharp image reconstruction kernel shows higher diagnostic accuracy for detecting lesions causing ischemia, potentially altering decision-making in a clinical setting.Key Points• Image reconstruction parameters influence the diagnostic accuracy of simulated fractional flow reserve derived from coronary computed tomography angiography.• Using a sharp kernel image reconstruction algorithm delivers higher diagnostic accuracy compared to medium smooth kernel image reconstruction (gold standard invasive fractional flow reserve).

A generative adversarial network-based CT image standardization model for predicting progression-free survival of lung cancer.

Progression-free survival (PFS) prediction using computed tomography (CT) images is important for treatment planning in lung cancer. However, the generalization ability of current analysis methods is usually affected by the scanning parameters of CT images, such as slice thickness and reconstruction kernel. In this paper, we proposed a generative adversarial network (GAN)-based model to convert heterogenous CT images into standardized CT images with uniform slice thickness and reconstruction kernel to increase the generalization of the predictive model. This model was trained in 173 patients with multiple CT sequences including both thin/thick voxel-spacing and sharp/soft reconstruction kernel. Afterward, we built a 3D-CNN model to predict the individualized 1year PFS of lung cancer using the standardized CT images in 281 patients. Finally, we evaluated the predictive model by 5-fold cross-validation and the mean area under the receiver operating characteristic curve (AUC). After transforming to the heterogenous CT images into the uniform thin-spacing and sharp kernel CT images, the AUC value of the 3D-CNN model improved from 0.614 to 0.686. Furthermore, this model can stratify the patients into high-risk and low-risk groups, where patients in these two groups showed significant difference in PFS (P < 0.001).

Spectral convergence of graph Laplacian and heat kernel reconstruction in L∞ from random samples

In the manifold setting, we provide a series of spectral convergence results quantifying how the eigenvectors and eigenvalues of the graph Laplacian converge to the eigenfunctions and eigenvalues of the Laplace-Beltrami operator in the L ∞ sense. Based on these results, convergence of the proposed heat kernel approximation algorithm, as well as the convergence rate, to the exact heat kernel is guaranteed. To our knowledge, this is the first work exploring the spectral convergence in the L ∞ sense and providing a numerical heat kernel reconstruction from the point cloud with theoretical guarantees.

Development and validation of an automated methodology to assess perceptual in vivo noise texture in liver CT.

Purpose: Developing, validating, and evaluating a method for measuring noise texture directly from patient liver CT images (i.e., in vivo). Approach: The method identifies target regions within patient scans that are least likely to have major contribution of patient anatomy, detrends them locally, and measures noise power spectrum (NPS) there using a previously phantom-validated technique targeting perceptual noise-non-anatomical fluctuations in the image that may interfere with the detection of focal lesions. Method development and validation used scanner-specific CT simulations of computational, anthropomorphic phantom (XCAT phantom, three phases of contrast-enhancement) with known ground truth of the NPS. Simulations were based on a clinical scanner (Definition Flash, Siemens) and clinically relevant settings (tube voltage of 120kV at three dose levels). Images were reconstructed with filtered backprojection (kernel: B31, B41, and B50) and Sinogram Affirmed Iterative Reconstruction (kernel: I31, I41, and I50) using a manufacturer-specific reconstruction software (ReconCT, Siemens). All NPS measurements were made in the liver. Ground-truth NPS were taken as the sum of (1)a measurement in parenchymal regions of anatomy-subtracted (i.e., noise only) scans, and (2)a measurement in the same region of noise-free (pre-noise-insertion) images. To assess in vivo NPS performance, correlation of NPS average frequency ( ), was reported. Sensitivity of accuracy [root-mean-square-error (RMSE)] to number of pixels included in measurement was conducted via bootstrapped pixel-dropout. Sensitivity of NPS to dose and reconstruction kernel was assessed to confirm that ground truth NPS similarities were maintained in patient-specific measurements. Results: Pearson and Spearman correlation coefficients 0.97 and 0.96 for indicated good correlation. Results suggested accurate NPS measurements (within 5% total RMSE) could be acquired with . Conclusions: Relationships of similar NPS due to reconstruction kernel and dose were preserved between gold standard and observed in vivo estimations. The NPS estimation method was further deployed on clinical cases to demonstrate the feasibility of clinical analysis.

Accuracy and Reproducibility of a Software Prototype for Semi-Automated Computer-Aided Volumetry of the solid and subsolid Components of part-solid Pulmonary Nodules

To test the accuracy and reproducibility of a software prototype for semi-automated computer-aided volumetry (CAV) of part-solid pulmonary nodules (PSN) with separate segmentation of the solid part. 66 PSNs were retrospectively identified in 34 thin-slice unenhanced chest CTs of 19 patients. CAV was performed by two medical students. Manual volumetry (MV) was carried out by two radiology residents. The reference standard was determined by an experienced radiologist in consensus with one of the residents. Visual assessment of CAV accuracy was performed. Measurement variability between CAV/MV and the reference standard as a measure of accuracy, CAV inter- and intra-rater variability as well as CAV intrascan variability between two recontruction kernels was determined via the Bland-Altman method and intraclass correlation coefficients (ICC). Subjectively assessed accuracy of CAV/MV was 77 %/79 %-80 % for the solid part and 67 %/73 %-76 % for the entire nodule. Measurement variability between CAV and the reference standard ranged from -151-117 % for the solid part and -106-54 % for the entire nodule. Interrater variability was -16-16 % for the solid part (ICC 0.998) and -102-65 % for the entire nodule (ICC 0.880). Intra-rater variability was -70-49 % for the solid part (ICC 0.992) and -111-31 % for the entire nodule (ICC 0.929). Intrascan variability between the smooth and the sharp reconstruction kernel was -45-39 % for the solid part and -21-46 % for the entire nodule. Although the software prototype delivered satisfactory results when segmentation is evaluated subjectively, quantitative statistical analysis revealed room for improvement especially regarding the segmentation accuracy of the solid part and the reproducibility of measurements of the nodule's subsolid margins. · Assessed visually CAV delivers similar accuracy compared to manual volumetry. · Accuracy of CAV was higher for the entire nodule. · Reproducibility was better for the solid part. · Variability between the kernels was higher for the solid part.

Deep learning for lung disease segmentation on CT: Which reconstruction kernel should be used?

PurposeThe purpose of this study was to determine whether a single reconstruction kernel or both high and low frequency kernels should be used for training deep learning models for the segmentation of diffuse lung disease on chest computed tomography (CT). Materials and methodsTwo annotated datasets of COVID-19 pneumonia (323,960 slices) and interstitial lung disease (ILD) (4,284 slices) were used. Annotated CT images were used to train a U-Net architecture to segment disease. All CT slices were reconstructed using both a lung kernel (LK) and a mediastinal kernel (MK). Three different trainings, resulting in three different models were compared for each disease: training on LK only, MK only or LK+MK images. Dice similarity scores (DSC) were compared using the Wilcoxon signed-rank test. ResultsModels only trained on LK images performed better on LK images than on MK images (median DSC = 0.62 [interquartile range (IQR): 0.54, 0.69] vs. 0.60 [IQR: 0.50, 0.70], P < 0.001 for COVID-19 and median DSC = 0.62 [IQR: 0.56, 0.69] vs. 0.50 [IQR 0.43, 0.57], P < 0.001 for ILD). Similarly, models only trained on MK images performed better on MK images (median DSC = 0.62 [IQR: 0.53, 0.68] vs. 0.54 [IQR: 0.47, 0.63], P < 0.001 for COVID-19 and 0.69 [IQR: 0.61, 0.73] vs. 0.63 [IQR: 0.53, 0.70], P < 0.001 for ILD). Models trained on both kernels performed better or similarly than those trained on only one kernel. For COVID-19, median DSC was 0.67 (IQR: =0.59, 0.73) when applied on LK images and 0.67 (IQR: 0.60, 0.74) when applied on MK images (P < 0.001 for both). For ILD, median DSC was 0.69 (IQR: 0.63, 0.73) when applied on LK images (P = 0.006) and 0.68 (IQR: 0.62, 0.72) when applied on MK images (P > 0.99). ConclusionReconstruction kernels impact the performance of deep learning-based models for lung disease segmentation. Training on both LK and MK images improves the performance.

Dose reduction potential in cone-beam CT imaging of upper extremity joints with a twin robotic x-ray system

Cone-beam computed tomography is a powerful tool for 3D imaging of the appendicular skeleton, facilitating detailed visualization of bone microarchitecture. This study evaluated various combinations of acquisition and reconstruction parameters for the cone-beam CT mode of a twin robotic x-ray system in cadaveric wrist and elbow scans, aiming to define the best possible trade-off between image quality and radiation dose. Images were acquired with different combinations of tube voltage and tube current–time product, resulting in five scan protocols with varying volume CT dose indices: full-dose (FD; 17.4 mGy), low-dose (LD; 4.5 mGy), ultra-low-dose (ULD; 1.15 mGy), modulated low-dose (mLD; 0.6 mGy) and modulated ultra-low-dose (mULD; 0.29 mGy). Each set of projection data was reconstructed with three convolution kernels (very sharp [Ur77], sharp [Br69], intermediate [Br62]). Five radiologists subjectively assessed the image quality of cortical bone, cancellous bone and soft tissue using seven-point scales. Irrespective of the reconstruction kernel, overall image quality of every FD, LD and ULD scan was deemed suitable for diagnostic use in contrast to mLD (very sharp/sharp/intermediate: 60/55/70%) and mULD (0/3/5%). Superior depiction of cortical and cancellous bone was achieved in FDUr77 and LDUr77 examinations (p < 0.001) with LDUr77 scans also providing favorable bone visualization compared to FDBr69 and FDBr62 (p < 0.001). Fleiss’ kappa was 0.618 (0.594–0.641; p < 0.001), indicating substantial interrater reliability. In this study, we demonstrate that considerable dose reduction can be realized while maintaining diagnostic image quality in upper extremity joint scans with the cone-beam CT mode of a twin robotic x-ray system. Application of sharper convolution kernels for image reconstruction facilitates superior display of bone microarchitecture.

Quantification of airway dimensions using a high-resolution CT scanner: A phantom study.

Small airways with inner diameters less than 2mm are sites of major airflow limitations in patients with chronic obstructive pulmonary disease (COPD) and asthma. The purpose of this study is to investigate the limitations for accurate assessment of small airway dimensions using both high-resolution CT (HRCT) and conventional normal-resolution CT at low dose levels. To model the normal human airways from the 3rd to 20th generations, a cylindrical polyurethane phantom with 14 airway tubes of inner diameters (ID) ranging from 0.3 to 3.4mm and wall thicknesses (WT) ranging from 0.15 to 1.6mm was placed within an Anthropomorphic QRM-Thorax phantom. The Aquilion Precision (Canon Medical Systems Corporation) HRCT scanner was used to acquire images at 80, 100, and 120kV using high resolution mode (HR, 0.25mm×160 detector configuration) and normal-resolution (NR) mode (0.5mm×80 detector configuration). The HR data were reconstructed using a 1024×1024matrix (0.22×0.22×0.25mm voxel size) and the NR data were reconstructed using a 512×512matrix (0.43×0.43×0.50mm). Two reconstruction algorithms (filtered back projection; FBP and an adaptive iterative dose reduction 3D algorithm; AIDR 3D) and three reconstruction kernels (FC30, FC52, and FC56) were investigated. The dose values ranged from 0.2 to 6.2mGy. A refined automated full-width half-maximum (FWHM) method was used for the measurement of airway dimensions, where the density profiles were computed by radial oversampling using a polar coordinate system. Both ID and WT were compared to the known dimensions using a regression model, and the root-mean-square error (RMSE) and average error were computed across all 14 airway tubes. The results indicate that the ID can be measured within a 15% error down to approximately 0.8 and 2.0mm using the HR and NR modes, respectively. The overall RMSE (and average error) of ID measurements for HR and NR were 0.10mm (-0.70%) and 0.31mm (-2.63%), respectively. The RMSE (and average error) of WT measurements using HR and NR were 0.10mm (23.27%) and 0.27mm (53.56%), respectively. The WT measurement using HR yielded a factor of two improvement in accuracy as compared to NR. High-resolution CT can provide more accurate measurements of airway dimensions as compared with NR CT, potentially improving quantitative assessment of pathologies such as COPD and asthma. The HR mode acquired and reconstructed with AIDR3D and the FC52kernel provides most accurate measurement of airway dimensions. Low-dose HR measurements at dose level above 0.9mGy can provide improved accuracy on both inner diameters and wall thicknesses compared to full dose NR airway phantom measurements.

Fully automated whole-liver volume quantification on CT-image data: Comparison with manual volumetry using enhanced and unenhanced images as well as two different radiation dose levels and two reconstruction kernels.

ObjectivesTo evaluate the accuracy of fully automated liver volume quantification vs. manual quantification using unenhanced as well as enhanced CT-image data as well as two different radiation dose levels and also two image reconstruction kernels.Material and methodsThe local ethics board gave its approval for retrospective data analysis. Automated liver volume quantification in 300 consecutive livers in 164 male and 103 female oncologic patients (64±12y) performed at our institution (between January 2020 and May 2020) using two different dual-energy helicals: portal-venous phase enhanced, ref. tube current 300mAs (CARE Dose4D) for tube A (100 kV) and ref. 232mAs tube current for tube B (Sn140kV), slice collimation 0.6mm, reconstruction kernel I30f/1, recon. thickness of 0.6mm and 5mm, 80–100 mL iodine contrast agent 350 mg/mL, (flow 2mL/s) and unenhanced ref. tube current 100mAs (CARE Dose4D) for tube A (100 kV) and ref. 77mAs tube current for tube B (Sn140kV), slice collimation 0.6mm (kernel Q40f) were analyzed. The post-processing tool (syngo.CT Liver Analysis) is already FDA-approved. Two resident radiologists with no and 1-year CT-experience performed both the automated measurements independently from each other. Results were compared with those of manual liver volume quantification using the same software which was supervised by a senior radiologist with 30-year CT-experience (ground truth).ResultsIn total, a correlation of 98% was obtained for liver volumetry based on enhanced and unenhanced data sets compared to the manual liver quantification. Radiologist #1 and #2 achieved an inter-reader agreement of 99.8% for manual liver segmentation (p<0.0001). Automated liver volumetry resulted in an overestimation (>5% deviation) of 3.7% for unenhanced CT-image data and 4.0% for contrast-enhanced CT-images. Underestimation (<5%) of liver volume was 2.0% for unenhanced CT-image data and 1.3% for enhanced images after automated liver volumetry. Number and distribution of erroneous volume measurements using either thin or thick slice reconstructions was exactly the same, both for the enhanced as well for the unenhanced image data sets (p> 0.05).ConclusionResults of fully automated liver volume quantification are accurate and comparable with those of manual liver volume quantification and the technique seems to be confident even if unenhanced lower-dose CT image data is used.

CT-based thermometry with virtual monoenergetic images by dual-energy of fat, muscle and bone using FBP, iterative and deep learning\u2013based reconstruction

ObjectivesThe aim of this study was to evaluate the sensitivity of CT-based thermometry for clinical applications regarding a three-component tissue phantom of fat, muscle and bone. Virtual monoenergetic images (VMI) by dual-energy measurements and conventional polychromatic 120-kVp images with modern reconstruction algorithms adaptive statistical iterative reconstruction-Volume (ASIR-V) and deep learning image reconstruction (DLIR) were compared.MethodsA temperature-regulating water circuit system was developed for the systematic evaluation of the correlation between temperature and Hounsfield units (HU). The measurements were performed on a Revolution CT with gemstone spectral imaging technology (GSI). Complementary measurements were performed without GSI (voltage 120 kVp, current 130–545 mA). The measured object was a tissue equivalent phantom in a temperature range of 18 to 50°C. The evaluation was carried out for VMI at 40 to 140 keV and polychromatic 120-kVp images.ResultsThe regression analysis showed a significant inverse linear dependency between temperature and average HU regardless of ASIR-V and DLIR. VMI show a higher temperature sensitivity compared to polychromatic images. The temperature sensitivities were 1.25 HU/°C (120 kVp) and 1.35 HU/°C (VMI at 140 keV) for fat, 0.38 HU/°C (120 kVp) and 0.47 HU/°C (VMI at 40 keV) for muscle and 1.15 HU/°C (120 kVp) and 3.58 HU/°C (VMI at 50 keV) for bone.ConclusionsDual-energy with VMI enables a higher temperature sensitivity for fat, muscle and bone. The reconstruction with ASIR-V and DLIR has no significant influence on CT-based thermometry, which opens up the potential of drastic dose reductions.Key Points• Virtual monoenergetic images (VMI) enable a higher temperature sensitivity for fat (8%), muscle (24%) and bone (211%) compared to conventional polychromatic 120-kVp images.• With VMI, there are parameters, e.g. monoenergy and reconstruction kernel, to modulate the temperature sensitivity. In contrast, there are no parameters to influence the temperature sensitivity for conventional polychromatic 120-kVp images.• The application of adaptive statistical iterative reconstruction-Volume (ASIR-V) and deep learning–based image reconstruction (DLIR) has no effect on CT-based thermometry, opening up the potential of drastic dose reductions in clinical applications.

Proton stopping power prediction based on dual-energy CT-generated virtual monoenergetic images.

The purpose of this work was to assess a proof of concept for a novel method for predicting proton stopping power ratios (SPRs) based on a pair of dual-energy CT generated virtual monoenergetic (VM) images. A rapid kV-switching dual-energy CT scanner was used to acquire Gemstone Spectral Imaging (GSI) and 120kV conventional single-energy CT (SECT) image data of the CIRS 062M phantom. The proposed method was applied to every possible pairing of VM images between 40 and 140keV to find the optimal energy pairs for SPR prediction in lung tissue, soft tissue, and bone. The predicted SPRs were compared against SPRs predicted from the SECT data using the conventional SECT-based method. The impact of different scan and reconstruction parameters was also investigated. The SPR residual root mean square errors (RMSE) yielded by the optimal pairs were 7.2% for lung tissue, 0.4% for soft tissue, and 0.8% for bone. While no direct comparison could be made to other DECT-based methods for SPR prediction, as these methods could not be directly implemented on a fast kV-switching system, the SPR RMSEs for soft tissue and bone in Table 4 are comparable to RMSEs reported in the literature. For the conventional SECT-based method, the SPR RMSEs were 5.9% for lung tissue, 0.9% for soft tissue, and 5.1% for bone. The proposed method is a valid alternative to, and has the potential to improve upon, the conventional SECT-based method for predicting SPRs. The formalism used in the method is applied directly, with no approximations made on our part, and requires neither prior knowledge of the spectra nor calibration with a phantom. This work presents a way of optimizing the proposed method for a specific scanner by determining the optimal energy pairs to use as input and demonstrates the method's robustness to different levels of ASiR-V, reconstruction kernels, and dose levels.

MRI-aided kernel PET image reconstruction method based on texture features

We investigate the reconstruction of low-count positron emission tomography (PET) projection, which is an important, but challenging, task. Using the texture feature extraction method of radiomics, i.e. the gray-level co-occurrence matrix (GLCM), texture features can be extracted from magnetic resonance imaging images with high-spatial resolution. In this work, we propose a kernel reconstruction method combining autocorrelation texture features derived from the GLCM. The new kernel function includes the correlations of both the intensity and texture features from the prior image. By regarding the GLCM as a discrete approximation of a probability density function, the asymptotically gray-level-invariant autocorrelation texture feature is generated, which can maintain the accuracy of texture features extracted from small image regions by reducing the number of quantized image gray levels. A computer simulation shows that the proposed method can effectively reduce the noise in the reconstructed image compared to the maximum likelihood expectation maximum method and improve the image quality and tumor region accuracy compared to the original kernel method for low-count PET reconstruction. A simulation study on clinical patient images also shows that the proposed method can improve the whole image quality and that the reconstruction of a high-uptake lesion is more accurate than that achieved by the original kernel method.

Unraveling the interplay of image formation, data representation and learning in CT-based COPD phenotyping automation: The need for a meta-strategy.

In the literature on automated phenotyping of chronic obstructive pulmonary disease (COPD), there is a multitude of isolated classical machine learning and deep learning techniques, mostly investigating individual phenotypes, with small study cohorts and heterogeneous meta-parameters, e.g., different scan protocols or segmented regions. The objective is to compare the impact of different experimental setups, i.e., varying meta-parameters related to image formation and data representation, with the impact of the learning technique for subtyping automation for a variety of phenotypes. The identified associations of these parameters with automation performance and their interactions might be a first step towards a determination of optimal meta-parameters, i.e., a meta-strategy. A clinical cohort of 981 patients (53.8±15.1years, 554male) was examined. The inspiratory CT images were analyzed to automate the diagnosis of 13 COPD phenotypes given by two radiologists. A benchmark feature set that integrates many quantitative criteria was extracted from the lung and trained a variety of learning algorithms on the first 654 patients (two thirds) and the respective algorithm retrospectively assessed the remaining 327 patients (one third). The automation performance was evaluated by the area under the receiver operating characteristic curve (AUC). 1717 experiments were conducted with varying meta-parameters such as reconstruction kernel, segmented regions and input dimensionality, i.e., number of extracted features. The association of the meta-parameters with the automation performance was analyzed by multivariable general linear model decomposition of the automation performance in the contributions of meta-parameters and the learning technique. The automation performance varied strongly for varying meta-parameters. For emphysema-predominant phenotypes, an AUC of 93%-95% could be achieved for the best meta-configuration. The airways-predominant phenotypes led to a lower performance of 65%-85%, while smooth kernel configurations on average were unexpectedly superior to those with sharp kernels. The performance impact of meta-parameters, even that of often neglected ones like the missing-data imputation, was in general larger than that of the learning technique. Advanced learning techniques like 3D deep learning or automated machine learning yielded inferior automation performance for non-optimal meta-configurations in comparison to simple techniques with suitable meta-configurations. The best automation performance was achieved by a combination of modern learning techniques and a suitable meta-configuration. Our results indicate that for COPD phenotype automation, study design parameters such as reconstruction kernel and the model input dimensionality should be adapted to the learning technique and may be more important than the technique itself. To achieve optimal automation and prediction results, the interaction between input those meta-parameters and the learning technique should be considered. This might be particularly relevant for the development of specific scan protocols for novel learning algorithms, and towards an understanding of good study design for automated phenotyping.