Quantification Errors Research Articles

Echo planar imaging-induced errors in intracardiac 4D flow MRI quantification.

PurposeTo assess errors associated with EPI‐accelerated intracardiac 4D flow MRI (4DEPI) with EPI factor 5, compared with non‐EPI gradient echo (4DGRE).MethodsThree 3T MRI experiments were performed comparing 4DEPI to 4DGRE: steady flow through straight tubes, pulsatile flow in a left‐ventricle phantom, and intracardiac flow in 10 healthy volunteers. For each experiment, 4DEPI was repeated with readout and blip phase‐encoding gradient in different orientations, parallel or perpendicular to the flow direction. In vitro flow rates were compared with timed volumetric collection. In the left‐ventricle phantom and in vivo, voxel‐based speed and spatio‐temporal median speed were compared between sequences, as well as mitral and aortic transvalvular net forward volume.ResultsIn steady‐flow phantoms, the flow rate error was largest (12%) for high velocity (>2 m/s) with 4DEPI readout gradient parallel to the flow. Voxel‐based speed and median speed in the left‐ventricle phantom were ≤5.5% different between sequences. In vivo, mean net forward volume inconsistency was largest (6.4 ± 8.5%) for 4DEPI with nonblip phase‐encoding gradient parallel to the main flow. The difference in median speed for 4DEPI versus 4DGRE was largest (9%) when the 4DEPI readout gradient was parallel to the flow.ConclusionsVelocity and flow rate are inaccurate for 4DEPI with EPI factor 5 when flow is parallel to the readout or blip phase‐encoding gradient. However, mean differences in flow rate, voxel‐based speed, and spatio‐temporal median speed were acceptable (≤10%) when comparing 4DEPI to 4DGRE for intracardiac flow in healthy volunteers.

Abstract 14294: Quantification Correction for Rapid Myocardial T1ρ Mapping in Mice Using the T1ρ* Relaxation Pathway

Introduction: Cardiac MRI offers detailed tissue characterization and high sensitivity to macromolecules using T 1ρ based imaging. However, fast and accurate T 1ρ mapping in the myocardium is still a major challenge in small animal models. The sequence design based on prospective triggering and breath gating causes a corruption of T 1ρ relaxation leading to quantification errors in the in vivo experiment. In this study, we present a T 1ρ quantification correction using the T 1ρ * relaxation pathway. Methods: A new signal equation was derived by analytical means describing spin-lock prepared fast gradient echo readouts: S(t SL ) = S 0 / [ exp(t SL /T 1ρ ) - cos(α) NR exp(-(N R T R +T rec )/T 1 ) ] The equation was validated in phantom experiments and used in vivo for corrected T 1ρ mapping in mice. Here, the correlation of T 1ρ and the breath dependent spin recovery time T rec was examined. The results were compared to the conventional monoexponential T 1ρ fitting. Results: In the phantom study, the novel signal equation provided improved quantification accuracy compared to monoexponential fitting. The mean quantification error could be reduced from -7.4% to -0.97%. In vivo, a correlation of T 1ρ with the respiratory cycle was observed for monoexponential fitting (PCC=0.804). Using the new correction method, this correlation was reduced (PCC=0.495) and the standard deviation was decreased from ±4.2% to ±1.3%. Conclusions: Our corrected formalism enables fast and accurate T 1ρ quantification for small animals and can improve the comparability of study results. The new technique thus offers a reasonable tool for assessing myocardial diseases. Besides, the derived signal equation could be used for sequence optimization or for subsequent correction of study results.

Encoding capability prediction of acquisition schedules in CEST MR fingerprinting for pH quantification.

To identify a reliable metric for predicting the encoding capability of CEST MR fingerprinting acquisition schedules for pH quantification, which may facilitate CEST MR fingerprinting protocol optimization. Numerical simulations and Cr phantom MRI experiments were conducted at 3 Tesla under representative CEST MR fingerprinting sampling scenarios, including the pseudorandomization of imaging parameters (e.g., saturation power B1 , saturation frequency offset, saturation time, and relaxation time), and variation of the maximum saturation power B1max , B1 number, and sampling pattern. The CEST effect at 2 ppm was measured using asymmetry analysis and matched to a predefined dictionary to determine the pH. The pH quantification error was assessed using RMSE. Three metrics, namely the Cramer-Rao bound, dot product, and Euclidean distance, were calculated for each sampling scenario, and their relationships with the pH RMSE were investigated to examine their effectiveness for predicting the encoding capability of sampling schedules for pH quantification. Both simulation and phantom studies revealed that the Cramer-Rao bound metric consistently exhibited superior performance for predicting the pH quantification error. Although dot product exhibited good encoding capability prediction in most sampling scenarios, it failed in the scenario with varied B1 numbers. In contrast, Euclidean distance exhibited the worst performance among the 3 metrics in all scenarios. Superior over dot product and Euclidean distance, the Cramer-Rao bound metric may reliably predicting the encoding capability of CEST MR fingerprinting sampling strategies and may be useful for guiding CEST MRI protocol optimization.

Hydrogeochemical Characterization and Its Seasonal Changes of Groundwater Based on Self-Organizing Maps

Water resources are scarce in arid or semiarid areas; groundwater is an important water source to maintain residents’ lives and the social economy; and identifying the hydrogeochemical characteristics of groundwater and its seasonal changes is a prerequisite for sustainable use and protection of groundwater. This study takes the Hongjiannao Basin as an example, and the Piper diagram, the Gibbs diagram, the Gaillardet diagram, the Chlor-alkali index, the saturation index, and the ion ratio were used to analyze the hydrogeochemical characteristics of groundwater. Meanwhile, based on self-organizing maps (SOM), quantification error (QE), topological error (TE), and the K-means algorithm, groundwater chemical data analysis was carried out to explore its seasonal variability. The results show that (1) the formation of groundwater chemistry in the study area was controlled by water–rock interactions and cation exchange, and the hydrochemical facies were HCO3-Ca type, HCO3-Na type, and Cl-Na type. (2) Groundwater chemical composition was mainly controlled by silicate weathering and carbonate dissolution, and the dissolution of halite, gypsum, and fluorite dominated the contribution of ions, while most dolomite and calcite were in a precipitated state or were reactive minerals. (3) All groundwater samples in wet and dry seasons were divided into five clusters, and the hydrochemical facies of clusters 1, 2, and 3 were HCO3-Ca type; cluster 4 was HCO3-Na type; and cluster 5 was Cl-Na type. (4) Thirty samples changed in the same clusters, and the groundwater chemistry characteristics of nine samples showed obvious seasonal variability, while the seasonal changes of groundwater hydrogeochemical characteristics were not significant.

Comparison of different deep learning architectures for synthetic CT generation from MR images.

Among the different available methods for synthetic CT generation from MR images for the task of MR-guided radiation planning, the deep learning algorithms have and do outperform their conventional counterparts. In this study, we investigated the performance of some most popular deep learning architectures including eCNN, U-Net, GAN, V-Net, and Res-Net for the task of sCT generation. As a baseline, an atlas-based method is implemented to which the results of the deep learning-based model are compared. A dataset consisting of 20 co-registered MR-CT pairs of the male pelvis is applied to assess the different sCT production methods' performance. The mean error (ME), mean absolute error (MAE), Pearson correlation coefficient (PCC), structural similarity index (SSIM), and peak signal-to-noise ratio (PSNR) metrics were computed between the estimated sCT and the ground truth (reference) CT images. The visual inspection revealed that the sCTs produced by eCNN, V-Net, and ResNet, unlike the other methods, were less noisy and greatly resemble the ground truth CT image. In the whole pelvis region, the eCNN yielded the lowest MAE (26.03±8.85 HU) and ME (0.82±7.06 HU), and the highest PCC metrics were yielded by the eCNN (0.93±0.05) and ResNet (0.91±0.02) methods. The ResNet model had the highest PSNR of 29.38±1.75 among all models. In terms of the Dice similarity coefficient, the eCNN method revealed superior performance in major tissue identification (air, bone, and soft tissue). All in all, the eCNN and ResNet deep learning methods revealed acceptable performance with clinically tolerable quantification errors.

Mathematical Models for Blood Flow Quantification in Dialysis Access Using Angiography: A Comparative Study.

Blood flow rate in dialysis (vascular) access is the key parameter to examine patency and to evaluate the outcomes of various endovascular interve7ntions. While angiography is extensively used for dialysis access–salvage procedures, to date, there is no image-based blood flow measurement application commercially available in the angiography suite. We aim to calculate the blood flow rate in the dialysis access based on cine-angiographic and fluoroscopic image sequences. In this study, we discuss image-based methods to quantify access blood flow in a flow phantom model. Digital subtraction angiography (DSA) and fluoroscopy were used to acquire images at various sampling rates (DSA—3 and 6 frames/s, fluoroscopy—4 and 10 pulses/s). Flow rates were computed based on two bolus tracking algorithms, peak-to-peak and cross-correlation, and modeled with three curve-fitting functions, gamma variate, lagged normal, and polynomial, to correct errors with transit time measurement. Dye propagation distance and the cross-sectional area were calculated by analyzing the contrast enhancement in the vessel. The calculated flow rates were correlated versus an in-line flow sensor measurement. The cross-correlation algorithm with gamma-variate curve fitting had the best accuracy and least variability in both imaging modes. The absolute percent error (mean ± SEM) of flow quantification in the DSA mode at 6 frames/s was 21.4 ± 1.9%, and in the fluoroscopic mode at 10 pulses/s was 37.4 ± 3.6%. The radiation dose varied linearly with the sampling rate in both imaging modes and was substantially low to invoke any tissue reactions or stochastic effects. The cross-correlation algorithm and gamma-variate curve fitting for DSA acquisition at 6 frames/s had the best correlation with the flow sensor measurements. These findings will be helpful to develop a software-based vascular access flow measurement tool for the angiography suite and to optimize the imaging protocol amenable for computational flow applications.

Impacts of acquisition and reconstruction parameters on the absolute technetium quantification of the cadmium\u2013zinc\u2013telluride-based SPECT/CT system: a phantom study

BackgroundThe digital cadmium–zinc–telluride (CZT)-based SPECT system has many advantages, including better spatial and energy resolution. However, the impacts of different acquisition and reconstruction parameters on CZT SPECT quantification might still need to be validated. This study aimed to evaluate the impacts of acquisition parameters (the main energy window and acquisition time per frame) and reconstruction parameters (the number of iterations, subsets in iterative reconstruction, post-filter, and image correction methods) on the technetium quantification of CZT SPECT/CT.MethodsA phantom (PET NEMA/IEC image quality, USA) was filled with four target-to-background (T/B) ratios (32:1, 16:1, 8:1, and 4:1) of technetium. Mean uptake values (the calculated mean concentrations for spheres) were measured to evaluate the recovery coefficient (RC) changes under different acquisition and reconstruction parameters. The corresponding standard deviations of mean uptake values were also measured to evaluate the quantification error. Image quality was evaluated using the National Electrical Manufacturers Association (NEMA) NU 2–2012 standard.ResultsFor all T/B ratios, significant correlations were found between iterations and RCs (r = 0.62–0.96 for 1–35 iterations, r = 0.94–0.99 for 35–90 iterations) as well as between the full width at half maximum (FWHM) of the Gaussian filter and RCs (r = − 0.86 to − 1.00, all P values < 0.05). The regression coefficients of 1–35 iterations were higher than those of 35–90 iterations (0.51–1.60 vs. 0.02–0.19). RCs calculated with AC (attenuation correction) + SC (scatter correction) + RR (resolution recovery correction) combination were more accurate (53.82–106.70%) than those calculated with other combinations (all P values < 0.05). No significant statistical differences (all P values > 0.05) were found between the 15% and 20% energy windows except for the 32:1 T/B ratio (P value = 0.023) or between the 10 s/frame and 120 s/frame acquisition times except for the 4:1 T/B ratio (P value = 0.015) in terms of RCs.ConclusionsCZT-SPECT/CT of technetium resulted in good quantification accuracy. The favourable acquisition parameters might be a 15% energy window and 40 s/frame of acquisition time. The favourable reconstruction parameters might be 35 iterations, 20 subsets, the AC + SC + RR correction combination, and no filter.

Simulated dose painting of hypoxic sub-volumes in pancreatic cancer stereotactic body radiotherapy

Dose painting of hypoxic tumour sub-volumes using positron-emission tomography (PET) has been shown to improve tumour control in silico in several sites, predominantly head and neck and lung cancers. Pancreatic cancer presents a more stringent challenge, given its proximity to critical gastro-intestinal organs-at-risk (OARs), anatomic motion, and impediments to reliable PET hypoxia quantification. A radiobiological model was developed to estimate clonogen survival fraction (SF), using 18F-fluoroazomycin arabinoside PET (FAZA PET) images from ten patients with unresectable pancreatic ductal adenocarcinoma to quantify oxygen enhancement effects. For each patient, four simulated five-fraction stereotactic body radiotherapy (SBRT) plans were generated: (1) a standard SBRT plan aiming to cover the planning target volume with 40 Gy, (2) dose painting plans delivering escalated doses to a maximum of three FAZA-avid hypoxic sub-volumes, (3) dose painting plans with simulated spacer separating the duodenum and pancreatic head, and (4), plans with integrated boosts to geometric contractions of the gross tumour volume (GTV). All plans saturated at least one OAR dose limit. SF was calculated for each plan and sensitivity of SF to simulated hypoxia quantification errors was evaluated. Dose painting resulted in a 55% reduction in SF as compared to standard SBRT; 78% with spacer. Integrated boosts to hypoxia-blind geometric contractions resulted in a 41% reduction in SF. The reduction in SF for dose-painting plans persisted for all hypoxia quantification parameters studied, including registration and rigid motion errors that resulted in shifts and rotations of the GTV and hypoxic sub-volumes by as much as 1 cm and 10 degrees. Although proximity to OARs ultimately limited dose escalation, with estimated SFs (∼10−5) well above levels required to completely ablate a ∼10 cm3 tumour, dose painting robustly reduced clonogen survival when accounting for expected treatment and imaging uncertainties and thus, may improve local response and associated morbidity.

Combined Mixed Potential Electrochemical Sensors and Artificial Neural Networks for the Quantificationand Identification of Methane in Natural Gas Emissions Monitoring

Sensors capable of quantifying methane concentration and discriminating between possible sources are needed for natural gas leak detection where multiple spatially overlapping sources including wetlands and agriculture may be present. We report on the fabrication by an additive manufacturing process of a four electrode La0.87Sr0.13CrO3, Indium Tin Oxide (In2O3 90 wt%, SnO2 10 wt%), Au, Pt mixed potential electrochemical sensor using yttria-stabilized zirconia (YSZ) as a solid electrolyte to natural gas detection. Artificial neural networks (ANNs) are used to automatically decode the possible source and concentration of methane. The ANNs trained on sensor data are capable of correctly discriminating between three sources of methane emissions from simulated mixtures of emissions from cattle, wetlands, or natural gas with >98% accuracy. Quantification error for methane in mixtures of CH4 in air, CH4 + NH3 in air, and simulated natural gas is less than 1.5% ppm when a two-temperature dataset is employed.

Cytochrome P450 3A4 (CYP3A4) protein quantification using capillary western blot technology and total protein normalization

The western blot (WB) is the predominate method for protein quantification, frequently used in pharmacological and toxicological studies. To control for technical variation, WB signals are normalized through immunodetection of an internal standard “house-keeping” gene or total protein quantification via staining of the same blot or a duplicate, sister blot. Increasing evidence suggests that house-keeping genes are subject to change after drug treatment or under disease states, causing protein quantification errors in WB. Recent advances in automated capillary-based WB technologies enable measurement of the protein of interest, internal standards, and total protein in a single capillary. Using this approach, we quantified cytochrome P450 3A4 (CYP3A4) across 179 liver samples and compared normalization by both β-actin and total protein to determine which better functions as an internal standard. CYP3A4 is responsible for metabolism of a wide array of xenobiotics and is known to exhibit large inter-person variation, making it a good candidate to evaluate protein quantification. We observed significant differences in β-actin protein levels between liver samples (~20-fold) and found better correlation between CYP3A4 protein and mRNA using total protein normalization than β-actin, indicating total protein normalization to be less error prone for estimation of CYP3A4. Furthermore, by using total protein normalization, we confirmed significant association between CYP3A4 protein expression and the functional CYP3A4 variant CYP3A4*22, which contains two linked SNPs rs35599367 and rs62471956. Our results indicate that the automatic capillary WB instrument combined with total protein normalization provides a high throughput and robust approach for protein quantification.

Quantifying location error to define uncertainty in volcanic mass flow hazard simulations

Abstract. The use of mass flow simulations in volcanic hazard zonation and mapping is often limited by model complexity (i.e. uncertainty in correct values of model parameters), a lack of model uncertainty quantification, and limited approaches to incorporate this uncertainty into hazard maps. When quantified, mass flow simulation errors are typically evaluated on a pixel-pair basis, using the difference between simulated and observed (“actual”) map-cell values to evaluate the performance of a model. However, these comparisons conflate location and quantification errors, neglecting possible spatial autocorrelation of evaluated errors. As a result, model performance assessments typically yield moderate accuracy values. In this paper, similarly moderate accuracy values were found in a performance assessment of three depth-averaged numerical models using the 2012 debris avalanche from the Upper Te Maari crater, Tongariro Volcano, as a benchmark. To provide a fairer assessment of performance and evaluate spatial covariance of errors, we use a fuzzy set approach to indicate the proximity of similarly valued map cells. This “fuzzification” of simulated results yields improvements in targeted performance metrics relative to a length scale parameter at the expense of decreases in opposing metrics (e.g. fewer false negatives result in more false positives) and a reduction in resolution. The use of this approach to generate hazard zones incorporating the identified uncertainty and associated trade-offs is demonstrated and indicates a potential use for informed stakeholders by reducing the complexity of uncertainty estimation and supporting decision-making from simulated data.

Evaluation of Iodine-123 and Iodine-131 SPECT activity quantification: a Monte Carlo study

PurposeThe quantitative accuracy of Nuclear Medicine images, acquired for both planar and SPECT studies, is influenced by the isotope-collimator combination as well as image corrections incorporated in the iterative reconstruction process. These factors can be investigated and optimised using Monte Carlo simulations. This study aimed to evaluate SPECT quantification accuracy for 123I with both the low-energy high resolution (LEHR) and medium-energy (ME) collimators and 131I with the high-energy (HE) collimator.MethodsSimulated SPECT projection images were reconstructed using the OS-EM iterative algorithm, which was optimised for the number of updates, with appropriate corrections for scatter, attenuation and collimator detector response (CDR), including septal scatter and penetration compensation. An appropriate calibration factor (CF) was determined from four different source geometries (activity-filled: water-filled cylindrical phantom, sphere in water-filled (cold) cylindrical phantom, sphere in air and point-like source), investigated with different volume of interest (VOI) diameters. Recovery curves were constructed from recovery coefficients to correct for partial volume effects (PVEs). The quantitative method was evaluated for spheres in voxel-based digital cylindrical and patient phantoms.ResultsThe optimal number of OS-EM updates was 60 for all isotope-collimator combinations. The CFpoint with a VOI diameter equal to the physical size plus a 3.0-cm margin was selected, for all isotope-collimator geometries. The spheres’ quantification errors in the voxel-based digital cylindrical and patient phantoms were less than 3.2% and 5.4%, respectively, for all isotope-collimator combinations.ConclusionThe study showed that quantification errors of less than 6.0% could be attained, for all isotope-collimator combinations, if corrections for; scatter, attenuation, CDR (including septal scatter and penetration) and PVEs are performed. 123I LEHR and 123I ME quantification accuracies compared well when appropriate corrections for septal scatter and penetration were applied. This can be useful in departments that perform 123I studies and may not have access to ME collimators.

Quantitative frother analysis on coal mine process water with a benchtop NMR spectrometer

This paper investigates the use of a benchtop NMR for quantification of a commonly used frothing agent, methyl isobutyl carbinol (MIBC) in the process water of a coal preparation facility. Solid phase extraction is used to increase the concentration of MIBC in the sample so that it is quantifiable by a benchtop NMR. A polymeric, reversed phase column with methanol as solvent gives a MIBC recovery rate of 67 ± 4% as determined using 400 MHz high-field NMR. The recovery rate consistently falls in the above narrow range even in the presence of diesel and inorganic electrolytes which are likely present as background chemicals in the process water. Using the average MIBC recovery rate, we use a quantum mechanical model to analyse the intensity of MIBC in the benchtop spectra. The quantum mechanical modelling algorithm effectively excludes the effect of the diesel on the measured NMR signal. The quantification error when the inlet concentration of MIBC is between 1 and 12 mg/L (1.2–15 ppm v/v), is within 0.5 mg/L (0.6 ppm v/v).

Monte Carlo strategy for SEM-EDS micro-nanoanalysis of geopolymer composites

The development, optimization and application of new geopolymer composite materials must necessarily go through a precise and accurate physico-chemical and mineralogical characterization down to the micro and nanoscale. In this regard, SEM-EDS X-ray microanalysis is widely and successfully employed by the scientific community and industry. However, the nano-to-micrometre sized architecture of many geopolymer composites introduces many difficulties and issues in SEM-EDS quantification, with potential large sources of error that should carefully be taken into account and investigated. In this work, a SEM-EDS Monte Carlo approach is proposed to study the complex physical phenomena at the basis of the quantification issues and errors, through the investigation of: (i) a not completely reacted sodium-poly(sialate-siloxo) geopolymer, and (ii) a geopolymer composite with a potassium-poly(sialate-siloxo) matrix and basalt-derived glass fibres reinforcement. The Monte Carlo simulation evinced a strong influence of the nano-microsized specimen architecture (e.g., basalt fibre size and shape, different elemental composition between fibre and matrix) on the measured X-ray intensity, with contributions also depending on the SEM electron beam energy. The proposed approach provided fundamental indications for selecting optimal operative conditions depending on the type of geopolymer sample, shape, size with the specific SEM-EDS setup and silicon drift EDS detector here used.

Fibre-Reinforced Geopolymer Composites Micro-Nanochemistry by SEM-EDS Simulations

The focus of the present study is on fibre-reinforced geopolymer composites, whose optimization and application necessarily need a detailed chemical characterization at the micro-nanoscale. In this regard, many geopolymer composites presenting micro and nanometric architectures pose a challenge for scanning electron microscopy with energy dispersive X-ray microanalysis (SEM-EDS) quantification, because of several potential sources of errors. For this reason, the present work reports a SEM-EDS Monte Carlo approach to carefully investigate the complex physical phenomena related to the cited quantification errors. The model used for this theoretical analysis is a simplified fibre-reinforced geopolymer with basalt-derived glass fibres immersed in a potassium-poly(sialate-siloxo) matrix. The simulated SEM-EDS spectra showed a strong influence on the measured X-ray intensity of (i) the sample nano-to-micro architecture, (ii) the electron beam probing energy and (iii) the electron probe-sample-EDS detector relative position. The results showed that, compared to a bulk material, the X-ray intensity for a nano-micrometric sized specimen may give rise to potential underestimation and/or overestimation of the elemental composition of the sample. The proposed Monte Carlo approach indicated the optimal instrumental setup depending on the sample and on the specific SEM-EDS equipment here considered.

Structural health monitoring of tendons in a multibody floating offshore wind turbine under varying environmental and operating conditions

The structural health monitoring of a Floating Offshore Wind Turbine (FOWT) tendons, taking into account the comprehensive damage diagnosis problem of damage detection, damaged tendon identification and damage precise quantification under varying environmental and operating conditions (EOCs), is investigated for the first time. The study examines a new concept of a 10 MW multibody FOWT whose tower is supported by a platform consisting of two rigid-body tanks connected by 12 tendons. Normal and the most severe EOCs from a site located in the northern coast of Scotland, are selected for the simulation of the FOWT structure under constant current but varying wind and wave conditions. Dynamic responses of the platform under different damage states are obtained based on the simulated FOWT. The damage scenarios are modelled via stiffness reduction (%) at the tendon's connection point to the platform's upper tank. Damage diagnosis is achieved via an advanced method, the Functional Model Based Method, that is formulated to operate using a single response signal and stochastic Functional Models representing the structural dynamics under the effects of varying EOCs and any magnitude of the considered damages. Due to the robustness and high number of the existing tendons, the effects of considered damages on the FOWT dynamics are minor and overlapped by the effects of the varying EOCs, indicating a highly challenging damage diagnosis problem. Very good damage detection results are obtained with the damage detection almost faultless and with no false alarms. Accurate damaged tendon identification is achieved for the 95% of the considered test cases, while the mean error in damage quantification is approximately equal to 4% using measurements from just a single accelerometer within a very limited frequency bandwidth of [0–5] Hz.

SuperDTI: Ultrafast DTI and fiber tractography with deep learning.

To develop a deep learning-based reconstruction framework for ultrafast and robust diffusion tensor imaging and fiber tractography. SuperDTI was developed to learn the nonlinear relationship between DWIs and the corresponding diffusion tensor parameter maps. It bypasses the tensor fitting procedure, which is highly susceptible to noises and motions in DWIs. The network was trained and tested using data sets from the Human Connectome Project and patients with ischemic stroke. Results from SuperDTI were compared against widely used methods for tensor parameter estimation and fiber tracking. Using training and testing data acquired using the same protocol and scanner, SuperDTI was shown to generate fractional anisotropy and mean diffusivity maps, as well as fiber tractography, from as few as six raw DWIs, with a quantification error of less than 5% in all white-matter and gray-matter regions of interest. It was robust to noises and motions in the testing data. Furthermore, the network trained using healthy volunteer data showed no apparent reduction in lesion detectability when directly applied to stroke patient data. Our results demonstrate the feasibility of superfast DTI and fiber tractography using deep learning with as few as six DWIs directly, bypassing tensor fitting. Such a significant reduction in scan time may allow the inclusion of DTI into the clinical routine for many potential applications.

Accelerated multi-target chemical exchange saturation transfer magnetic resonance imaging of the mouse heart

Cardiac chemical exchange saturation transfer-magnetic resonance imaging (CEST-MRI) has been used to probe levels of various metabolites that provide insight into myocardial structure and function. However, imaging of the heart using CEST-MRI is prolonged by the need to repeatedly acquire multiple images for a full Z-spectrum and to perform saturation and acquisition around cardiac and respiratory cycles. Compressed sensing (CS) reconstruction of sparse data enables accelerated acquisition, but reconstruction artifacts may bias subsequently derived measures of CEST contrast. In this study, we examine the impact of CS reconstruction of increasingly under-sampled cardiac CEST-MRI data on subsequent CEST contrasts of amine-containing metabolites and amide-containing proteins. Cardiac CEST-MRI data sets were acquired in six mice using low and high RF saturation for single and dual contrast generation, respectively. CEST-weighted images were reconstructed using CS methods at 2–5× levels of under-sampling. CEST contrasts were derived from corresponding Z-spectra and the impact of accelerated imaging on accuracy was assessed via analysis of variance. CS reconstruction preserved myocardial signal to noise ratio as compared to conventional reconstruction. However, greater absolute error and distribution of derived contrasts was observed with increasing acceleration factors. The results from this study indicate that acquisition of radial cardiac CEST-MRI data can be modestly, but meaningfully, accelerated via CS reconstructions with little error in CEST contrast quantification.

Fully Automated Gross Tumor Volume Delineation From PET in Head and Neck Cancer Using Deep Learning Algorithms.

The availability of automated, accurate, and robust gross tumor volume (GTV) segmentation algorithms is critical for the management of head and neck cancer (HNC) patients. In this work, we evaluated 3 state-of-the-art deep learning algorithms combined with 8 different loss functions for PET image segmentation using a comprehensive training set and evaluated its performance on an external validation set of HNC patients. 18F-FDG PET/CT images of 470 patients presenting with HNC on which manually defined GTVs serving as standard of reference were used for training (340 patients), evaluation (30 patients), and testing (100 patients from different centers) of these algorithms. PET image intensity was converted to SUVs and normalized in the range (0-1) using the SUVmax of the whole data set. PET images were cropped to 12 × 12 × 12 cm3 subvolumes using isotropic voxel spacing of 3 × 3 × 3 mm3 containing the whole tumor and neighboring background including lymph nodes. We used different approaches for data augmentation, including rotation (-15 degrees, +15 degrees), scaling (-20%, 20%), random flipping (3 axes), and elastic deformation (sigma = 1 and proportion to deform = 0.7) to increase the number of training sets. Three state-of-the-art networks, including Dense-VNet, NN-UNet, and Res-Net, with 8 different loss functions, including Dice, generalized Wasserstein Dice loss, Dice plus XEnt loss, generalized Dice loss, cross-entropy, sensitivity-specificity, and Tversky, were used. Overall, 28 different networks were built. Standard image segmentation metrics, including Dice similarity, image-derived PET metrics, first-order, and shape radiomic features, were used for performance assessment of these algorithms. The best results in terms of Dice coefficient (mean ± SD) were achieved by cross-entropy for Res-Net (0.86 ± 0.05; 95% confidence interval [CI], 0.85-0.87), Dense-VNet (0.85 ± 0.058; 95% CI, 0.84-0.86), and Dice plus XEnt for NN-UNet (0.87 ± 0.05; 95% CI, 0.86-0.88). The difference between the 3 networks was not statistically significant (P > 0.05). The percent relative error (RE%) of SUVmax quantification was less than 5% in networks with a Dice coefficient more than 0.84, whereas a lower RE% (0.41%) was achieved by Res-Net with cross-entropy loss. For maximum 3-dimensional diameter and sphericity shape features, all networks achieved a RE ≤ 5% and ≤10%, respectively, reflecting a small variability. Deep learning algorithms exhibited promising performance for automated GTV delineation on HNC PET images. Different loss functions performed competitively when using different networks and cross-entropy for Res-Net, Dense-VNet, and Dice plus XEnt for NN-UNet emerged as reliable networks for GTV delineation. Caution should be exercised for clinical deployment owing to the occurrence of outliers in deep learning-based algorithms.

Minimizing OCT quantification error via a surface-tracking imaging probe.

OCT-based quantitative tissue optical properties imaging is a promising technique for intraoperative brain cancer assessment. The attenuation coefficient analysis relies on the depth-dependent OCT intensity profile, thus sensitive to tissue surface positions relative to the imaging beam focus. However, it is almost impossible to maintain a steady tissue surface during intraoperative imaging due to the patient's arterial pulsation and breathing, the operator's motion, and the complex tissue surface geometry of the surgical cavity. In this work, we developed an intraoperative OCT imaging probe with a surface-tracking function to minimize the quantification errors in optical attenuation due to the tissue surface position variations. A compact OCT imaging probe was designed and engineered to have a long working distance of ∼ 41 mm and a large field of view of 4 × 4 mm2 while keeping the probe diameter small (9 mm) to maximize clinical versatility. A piezo-based linear motor was integrated with the imaging probe and controlled based upon real-time feedback of tissue surface position inferred from OCT images. A GPU-assisted parallel processing algorithm was implemented, enabling detection and tracking of tissue surface in real-time and successfully suppressing more than 90% of the typical physiologically induced motion range. The surface-tracking intraoperative OCT imaging probe could maintain a steady beam focus inside the target tissue regardless of the surface geometry or physiological motions and enabled to obtain tissue optical attenuation reliably for assessing brain cancer margins in challenging intraoperative settings.