Federated learning for predicting clinical outcomes in patients with COVID-19

Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC) >0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site's data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare.

Sparse-view spectral CT reconstruction using spectral patch-based low-rank penalty

Spectral computed tomography (CT) is a promising technique with the potential for improving lesion detection, tissue characterization, and material decomposition. In this paper, we are interested in kVp switching-based spectral CT that alternates distinct kVp X-ray transmissions during gantry rotation. This system can acquire multiple X-ray energy transmissions without additional radiation dose. However, only sparse views are generated for each spectral measurement; and the spectra themselves are limited in number. To address these limitations, we propose a penalized maximum likelihood method using spectral patch-based low-rank penalty, which exploits the self-similarity of patches that are collected at the same position in spectral images. The main advantage is that the relatively small number of materials within each patch allows us to employ the low-rank penalty that is less sensitive to intensity changes while preserving edge directions. In our optimization formulation, the cost function consists of the Poisson log-likelihood for X-ray transmission and the nonconvex patch-based low-rank penalty. Since the original cost function is difficult to minimize directly, we propose an optimization method using separable quadratic surrogate and concave convex procedure algorithms for the log-likelihood and penalty terms, which results in an alternating minimization that provides a computational advantage because each subproblem can be solved independently. We performed computer simulations and a real experiment using a kVp switching-based spectral CT with sparse-view measurements, and compared the proposed method with conventional algorithms. We confirmed that the proposed method improves spectral images both qualitatively and quantitatively. Furthermore, our GPU implementation significantly reduces the computational cost.

Direct reconstruction of cardiac PET kinetic parametric images using a preconditioned conjugate gradient approach

PURPOSE

Our research goal is to develop an algorithm to reconstruct cardiac positron emission tomography (PET) kinetic parametric images directly from sinograms and compare its performance with the conventional indirect approach.

METHODS

Time activity curves of a NCAT phantom were computed according to a one-tissue compartmental kinetic model with realistic kinetic parameters. The sinograms at each time frame were simulated using the activity distribution for the time frame. The authors reconstructed the parametric images directly from the sinograms by optimizing a cost function, which included the Poisson log-likelihood and a spatial regularization terms, using the preconditioned conjugate gradient (PCG) algorithm with the proposed preconditioner. The proposed preconditioner is a diagonal matrix whose diagonal entries are the ratio of the parameter and the sensitivity of the radioactivity associated with parameter. The authors compared the reconstructed parametric images using the direct approach with those reconstructed using the conventional indirect approach.

RESULTS

At the same bias, the direct approach yielded significant relative reduction in standard deviation by 12%-29% and 32%-70% for 50 × 10(6) and 10 × 10(6) detected coincidences counts, respectively. Also, the PCG method effectively reached a constant value after only 10 iterations (with numerical convergence achieved after 40-50 iterations), while more than 500 iterations were needed for CG.

CONCLUSIONS

The authors have developed a novel approach based on the PCG algorithm to directly reconstruct cardiac PET parametric images from sinograms, and yield better estimation of kinetic parameters than the conventional indirect approach, i.e., curve fitting of reconstructed images. The PCG method increases the convergence rate of reconstruction significantly as compared to the conventional CG method.

Microarray image denoising using complex Gaussian scale mixtures of complex wavelets

Microarray images when contaminated with noise may severely affect the detection and quantification of gene expression. In this paper, we propose to use the complex Gaussian scale mixture (CGSM) model in complex wavelet domain for noise reduction in complementary DNA microarray images. Based on the joint information in the red and green channel microarray images, we model the complex wavelet coefficients of the channel images jointly using the CGSM, and subsequently perform image denoising using Bayes least square estimator in complex wavelet domain. The experimental results show that using the CGSM of complex wavelet coefficients provides better noise reduction of microarray images compared to other complex wavelet-based models.

Effect of time-of-flight and point spread function modeling on detectability of myocardial defects in PET

PURPOSE

A study was designed to investigate the impact of time-of-flight (TOF) and point spread function (PSF) modeling on the detectability of myocardial defects.

METHODS

Clinical FDG-PET data were used to generate populations of defect-present and defect-absent images. Defects were incorporated at three contrast levels, and images were reconstructed by ordered subset expectation maximization (OSEM) iterative methods including ordinary Poisson, alone and with PSF, TOF, and PSF+TOF. Channelized Hotelling observer signal-to-noise ratio (SNR) was the surrogate for human observer performance.

RESULTS

For three iterations, 12 subsets, and no postreconstruction smoothing, TOF improved overall defect detection SNR by 8.6% as compared to its non-TOF counterpart for all the defect contrasts. Due to the slow convergence of PSF reconstruction, PSF yielded 4.4% less SNR than non-PSF. For reconstruction parameters (iteration number and postreconstruction smoothing kernel size) optimizing observer SNR, PSF showed larger improvement for faint defects. The combination of TOF and PSF improved mean detection SNR as compared to non-TOF and non-PSF counterparts by 3.0% and 3.2%, respectively.

CONCLUSIONS

For typical reconstruction protocol used in clinical practice, i.e., less than five iterations, TOF improved defect detectability. In contrast, PSF generally yielded less detectability. For large number of iterations, TOF+PSF yields the best observer performance.

Magnetic Resonance-based Motion Correction for Quantitative PET in Simultaneous PET-MR Imaging

Motion degrades image quality and quantitation of PET images, and is an obstacle to quantitative PET imaging. Simultaneous PET-MR offers a tool that can be used for correcting the motion in PET images by using anatomic information from MR imaging acquired concurrently. Motion correction can be performed by transforming a set of reconstructed PET images into the same frame or by incorporating the transformation into the system model and reconstructing the motion-corrected image. Several phantom and patient studies have validated that MR-based motion correction strategies have great promise for quantitative PET imaging in simultaneous PET-MR.

Tumor Prognostic Prediction of Nasopharyngeal Carcinoma Using CT-Based Radiomics in Non-Chinese Patients

Purpose

We aimed to construct predictive models for the overall survival (OS), progression-free survival (PFS), and distant metastasis-free survival (DMFS) for nasopharyngeal carcinoma (NPC) patients by using CT-based radiomics.

Materials and Methods

We collected data from 197 NPC patients. For each patient, radiomic features were extracted from the CT image acquired at pretreatment via PyRadiomics. Feature selection was performed in two steps. First, features with high inter-observer variability based on multiple tumor delineations were excluded. Then, stratified bootstrappings were performed to identify feature combinations that most frequently achieved the highest (i) area under the receiver operating characteristic curve (AUC) for predicting 3-year OS, PFS, and DMFS or (ii) Harrell’s C-index for predicting time to event. Finally, regularized logistic regression and Cox proportional hazard models with the most frequently selected feature combinations as input were tuned using cross-validation. Additionally, we examined the robustness of the constructed model to variation in tumor delineation by simulating 100 realizations of radiomic feature values to mimic features extracted from different tumor boundaries.

Results

The combined model that used both radiomics and clinical features yielded significantly higher AUC and Harrell’s C-index than models using either feature set alone for all outcomes (p < 0.05). The AUCs and Harrell’s C-indices of the clinical-only and radiomics-only models ranged from 0.758 ± 0.091 to 0.789 ± 0.082 and from 0.747 ± 0.062 to 0.767 ± 0.074, respectively. In comparison, the combined models achieved AUC of 0.801 ± 0.075 to 0.813 ± 0.078 and Harrell’s C-indices of 0.779 ± 0.066 to 0.796 ± 0.069. The results showed that our models were robust to variation in tumor delineation with the coefficient of variation ranging from 4.8% to 6.4% and from 6.7% to 9.3% for AUC and Harrell’s C-index, respectively.

Conclusion

Our results demonstrated that using CT-based radiomic features together with clinical features provided superior NPC prognostic prediction than using either clinical or radiomic features alone.

Direct parametric reconstruction in dynamic PET myocardial perfusion imaging: in vivo studies

Dynamic PET myocardial perfusion imaging (MPI) used in conjunction with tracer kinetic modeling enables the quantification of absolute myocardial blood flow (MBF). However, MBF maps computed using the traditional indirect method (i.e. post-reconstruction voxel-wise fitting of kinetic model to PET time-activity-curves-TACs) suffer from poor signal-to-noise ratio (SNR). Direct reconstruction of kinetic parameters from raw PET projection data has been shown to offer parametric images with higher SNR compared to the indirect method. The aim of this study was to extend and evaluate the performance of a direct parametric reconstruction method using in vivo dynamic PET MPI data for the purpose of quantifying MBF. Dynamic PET MPI studies were performed on two healthy pigs using a Siemens Biograph mMR scanner. List-mode PET data for each animal were acquired following a bolus injection of ~7-8 mCi of 18F-flurpiridaz, a myocardial perfusion agent. Fully-3D dynamic PET sinograms were obtained by sorting the coincidence events into 16 temporal frames covering ~5 min after radiotracer administration. Additionally, eight independent noise realizations of both scans-each containing 1/8th of the total number of events-were generated from the original list-mode data. Dynamic sinograms were then used to compute parametric maps using the conventional indirect method and the proposed direct method. For both methods, a one-tissue compartment model accounting for spillover from the left and right ventricle blood-pools was used to describe the kinetics of 18F-flurpiridaz. An image-derived arterial input function obtained from a TAC taken in the left ventricle cavity was used for tracer kinetic analysis. For the indirect method, frame-by-frame images were estimated using two fully-3D reconstruction techniques: the standard ordered subset expectation maximization (OSEM) reconstruction algorithm on one side, and the one-step late maximum a posteriori (OSL-MAP) algorithm on the other side, which incorporates a quadratic penalty function. The parametric images were then calculated using voxel-wise weighted least-square fitting of the reconstructed myocardial PET TACs. For the direct method, parametric images were estimated directly from the dynamic PET sinograms using a maximum a posteriori (MAP) parametric reconstruction algorithm which optimizes an objective function comprised of the Poisson log-likelihood term, the kinetic model and a quadratic penalty function. Maximization of the objective function with respect to each set of parameters was achieved using a preconditioned conjugate gradient algorithm with a specifically developed pre-conditioner. The performance of the direct method was evaluated by comparing voxel- and segment-wise estimates of [Formula: see text], the tracer transport rate (ml · min-1 · ml-1), to those obtained using the indirect method applied to both OSEM and OSL-MAP dynamic reconstructions. The proposed direct reconstruction method produced [Formula: see text] maps with visibly lower noise than the indirect method based on OSEM and OSL-MAP reconstructions. At normal count levels, the direct method was shown to outperform the indirect method based on OSL-MAP in the sense that at matched level of bias, reduced regional noise levels were obtained. At lower count levels, the direct method produced [Formula: see text] estimates with significantly lower standard deviation across noise realizations than the indirect method based on OSL-MAP at matched bias level. In all cases, the direct method yielded lower noise and standard deviation than the indirect method based on OSEM. Overall, the proposed direct reconstruction offered a better bias-variance tradeoff than the indirect method applied to either OSEM and OSL-MAP. Direct parametric reconstruction as applied to in vivo dynamic PET MPI data is therefore a promising method for producing MBF maps with lower variance.

A Bayesian spatial temporal mixtures approach to kinetic parametric images in dynamic positron emission tomography

PURPOSE

Estimation of parametric maps is challenging for kinetic models in dynamic positron emission tomography. Since voxel kinetics tend to be spatially contiguous, the authors consider groups of homogeneous voxels together. The authors propose a novel algorithm to identify the groups and estimate kinetic parameters simultaneously. Uncertainty estimates for kinetic parameters are also obtained.

METHODS

Mixture models were used to fit the time activity curves. In order to borrow information from spatially nearby voxels, the Potts model was adopted. A spatial temporal model was built incorporating both spatial and temporal information in the data. Markov chain Monte Carlo was used to carry out parameter estimation. Evaluation and comparisons with existing methods were carried out on cardiac studies using both simulated data sets and a pig study data. One-compartment kinetic modeling was used, in which K1 is the parameter of interest, providing a measure of local perfusion.

RESULTS

Based on simulation experiments, the median standard deviation across all image voxels, of K1 estimates were 0, 0.13, and 0.16 for the proposed spatial mixture models (SMMs), standard curve fitting, and spatial K-means methods, respectively. The corresponding median mean squared biases for K1 were 0.04, 0.06, and 0.06 for abnormal region of interest (ROI); 0.03, 0.03, and 0.04 for normal ROI; and 0.007, 0.02, and 0.05 for the noise region.

CONCLUSIONS

SMM is a fully Bayesian algorithm which determines the optimal number of homogeneous voxel groups, voxel group membership, parameter estimation, and parameter uncertainty estimation simultaneously. The voxel membership can also be used for classification purposes. By borrowing information from spatially nearby voxels, SMM substantially reduces the variability of parameter estimates. In some ROIs, SMM also reduces mean squared bias.

Federated Learning used for predicting outcomes in SARS-COV-2 patients

'Federated Learning' (FL) is a method to train Artificial Intelligence (AI) models with data from multiple sources while maintaining anonymity of the data thus removing many barriers to data sharing. During the SARS-COV-2 pandemic, 20 institutes collaborated on a healthcare FL study to predict future oxygen requirements of infected patients using inputs of vital signs, laboratory data, and chest x-rays, constituting the "EXAM" (EMR CXR AI Model) model. EXAM achieved an average Area Under the Curve (AUC) of over 0.92, an average improvement of 16%, and a 38% increase in generalisability over local models. The FL paradigm was successfully applied to facilitate a rapid data science collaboration without data exchange, resulting in a model that generalised across heterogeneous, unharmonized datasets. This provided the broader healthcare community with a validated model to respond to COVID-19 challenges, as well as set the stage for broader use of FL in healthcare.

Simultaneous 99mTc-MDP/123I-MIBG tumor imaging using SPECT-CT: phantom and constructed patient studies

PURPOSE

Authors' goal is to evaluate the performance of simultaneous (99m)Tc-MDP/(123)I-MIBG tumor imaging with fast Monte-Carlo (MC) based joint iterative reconstruction as compared to sequential (99m)Tc-MDP and (123)I-MIBG tumor imaging.

METHODS

Noise-free (99m)Tc and (123)I SPECT projections were acquired separately using an anthropomorphic torso phantom modified to include a fillable tube around the lungs to mimic ribs. Additionally, (99m)Tc and (123)I projections were acquired separately using a 1-cm spherical "tumor" placed at various distances from one detector. Tumor-present data were generated by adding tumor projections to the torso phantom data, which were scaled to the total counts in typical clinical studies. Twenty-five noise realizations were generated by adding Poisson noise to the projection data for each radionuclide. Dual-radionuclide data were synthesized by summing the (99m)Tc and (123)I projections. Image reconstruction was performed using: (1) SR-OSEM, ordered subset expectation maximization (OSEM) without scatter correction (SC) using single-radionuclide (SR) data; (2) SR-MC-OSEM, OSEM with a fast MC-based SC using SR data; (3) DR-OSEM, OSEM without SC using dual-radionuclide (DR) data; and (4) DR-MC-JOSEM, joint OSEM with a fast MC-based SC using DR data. Ten (99m)Tc-MDP and ten (123)I-MIBG data sets, which had tumors mathematically inserted, were also used to evaluate the performance of authors' approach. For the phantom study, relative bias and relative standard deviation of tumor uptake were computed for each tumor using the tumor uptake in the noise-free single-radionuclide images, which were reconstructed by SR-MC-OSEM, as the gold standard. For both the phantom and constructed patient studies, mean contrast and standard deviation of contrast were computed for each tumor for both the single- and dual-radionuclide images. Additionally, contrast recovery was computed as the ratio between mean contrast and the mean contrast for SR-MC-OSEM.

RESULTS

For the phantom study, DR-MC-JOSEM yielded 2.7% on average relative bias of tumor uptake using the images, which were reconstructed from the noise-free SR data with SR-MC-OSEM, as the gold-standard. For both the phantom and constructed patient studies, DR-MC-JOSEM yielded 94.7% and 95.2% tumor contrast recovery on average using SR-MC-OSEM as the reference, in the phantom and constructed patient studies, respectively. DR-MC-JOSEM yielded comparable relative standard deviation of bias and standard deviation of contrast to SR-MC-OSEM.

CONCLUSIONS

Simultaneous (99m)Tc-MDP/(123)I-MIBG tumor imaging using authors' dual-radionuclide reconstruction approach yielded comparable image quality to sequential (99m)Tc-MDP and (123)I-MIBG imaging. For patients who need to undergo both scans, authors' approach offers perfectly registered dual-tracer images under identical conditions without compromising image quality, and reduces the imaging cost while increasing patient throughput.

Spectral CT using multiple balanced K-edge filters

Our goal is to validate a spectral computed tomography (CT) system design that uses a conventional X-ray source with multiple balanced K-edge filters. By performing a simultaneously synthetic reconstruction in multiple energy bins, we obtained a good agreement between measurements and model expectations for a reasonably complex phantom. We performed simulation and data acquisition on a phantom containing multiple rods of different materials using a NeuroLogica CT scanner. Five balanced K-edge filters including Molybdenum, Cerium, Dysprosium, Erbium, and Tungsten were used separately proximal to the X-ray tube. For each sinogram bin, measured filtered vector can be defined as a product of a transmission matrix, which is determined by the filters and is independent of the imaging object, and energy-binned intensity vector. The energy-binned sinograms were then obtained by inverting the transmission matrix followed by a multiplication of the filter measurement vector. For each energy bin defined by two consecutive K-edges, a synthesized energy-binned attenuation image was obtained using filtered back-projection reconstruction. The reconstructed attenuation coefficients for each rod obtained from the experiment was in good agreement with the corresponding simulated results. Furthermore, the reconstructed attenuation coefficients for a given energy bin, agreed with National Institute of Standards and Technology reference values when beam hardening within the energy bin is small. The proposed cost-effective system design using multiple balanced K-edge filters can be used to perform spectral CT imaging at clinically relevant flux rates using conventional detectors and integrating electronics.

Determination of radiation dose and low-dose protocol for digital chest tomosynthesis using radiophotoluminescent (RPL) glass dosimeters

PURPOSE

This study aimed to determine a low-dose protocol for digital chest tomosynthesis (DTS).

METHODS

Five simulated nodules with a CT number of approximately 100 HU with size diameter of 3, 5, 8, 10, and 12 mm were inserted into an anthropomorphic chest phantom (N1 Lungman model), and then scanned by DTS system (Definium 8000) with varying tube voltage, copper filter thickness, and dose ratio. Three radiophotoluminescent (RPL) glass dosimeters, type GD-352 M with a dimension of 1.5 × 12 mm, were used to measure the entrance surface air kerma (ESAK) in each protocol. The effective dose (ED) was calculated using the recorded total dose-area-product (DAP). The signal-to-noise ratio (SNR) was determined for qualitative image quality evaluation. The image criteria and nodule detection capability were scored by two experienced radiologists. The selected low-dose protocol was further applied in a clinical study with 30 pulmonary nodule follow-up patients.

RESULTS

The average ESAK obtained from the standard default protocol was 1.68 ± 0.15 mGy, while an ESAK of 0.47 ± 0.02 mGy was found for a low-dose protocol. The EDs for the default and low-dose protocols were 313.98 ± 0.72 µSv and 100.55 ± 0.28 µSv, respectively. There were small non-significant differences in the image criteria and nodule detection scoring between the low-dose and default protocols interpreted by two radiologists. The effective dose of 98.87 ± 0.08 µSv was obtained in clinical study after applying the low-dose protocol.

CONCLUSIONS

The low-dose protocol obtained in this study can substantially reduce radiation dose while preserving an acceptable image quality compared to the standard protocol.

Numerical observer for atherosclerotic plaque classification in spectral computed tomography

Spectral computed tomography (SCT) generates better image quality than conventional computed tomography (CT). It has overcome several limitations for imaging atherosclerotic plaque. However, the literature evaluating the performance of SCT based on objective image assessment is very limited for the task of discriminating plaques. We developed a numerical-observer method and used it to assess performance on discrimination vulnerable-plaque features and compared the performance among multienergy CT (MECT), dual-energy CT (DECT), and conventional CT methods. Our numerical observer was designed to incorporate all spectral information and comprised two-processing stages. First, each energy-window domain was preprocessed by a set of localized channelized Hotelling observers (CHO). In this step, the spectral image in each energy bin was decorrelated using localized prewhitening and matched filtering with a set of Laguerre-Gaussian channel functions. Second, the series of the intermediate scores computed from all the CHOs were integrated by a Hotelling observer with an additional prewhitening and matched filter. The overall signal-to-noise ratio (SNR) and the area under the receiver operating characteristic curve (AUC) were obtained, yielding an overall discrimination performance metric. The performance of our new observer was evaluated for the particular binary classification task of differentiating between alternative plaque characterizations in carotid arteries. A clinically realistic model of signal variability was also included in our simulation of the discrimination tasks. The inclusion of signal variation is a key to applying the proposed observer method to spectral CT data. Hence, the task-based approaches based on the signal-known-exactly/background-known-exactly (SKE/BKE) framework and the clinical-relevant signal-known-statistically/background-known-exactly (SKS/BKE) framework were applied for analytical computation of figures of merit (FOM). Simulated data of a carotid-atherosclerosis patient were used to validate our methods. We used an extended cardiac-torso anthropomorphic digital phantom and three simulated plaque types (i.e., calcified plaque, fatty-mixed plaque, and iodine-mixed blood). The images were reconstructed using a standard filtered backprojection (FBP) algorithm for all the acquisition methods and were applied to perform two different discrimination tasks of: (1) calcified plaque versus fatty-mixed plaque and (2) calcified plaque versus iodine-mixed blood. MECT outperformed DECT and conventional CT systems for all cases of the SKE/BKE and SKS/BKE tasks (all [Formula: see text]). On average of signal variability, MECT yielded the SNR improvements over other acquisition methods in the range of 46.8% to 65.3% (all [Formula: see text]) for FBP-Ramp images and 53.2% to 67.7% (all [Formula: see text]) for FBP-Hanning images for both identification tasks. This proposed numerical observer combined with our signal variability framework is promising for assessing material characterization obtained through the additional energy-dependent attenuation information of SCT. These methods can be further extended to other clinical tasks such as kidney or urinary stone identification applications.

Joint reconstruction of Ictal/inter-ictal SPECT data for improved epileptic foci localization

PURPOSE

To improve the performance for localizing epileptic foci, we have developed a joint ictal/inter-ictal SPECT reconstruction method in which ictal and inter-ictal SPECT projections are simultaneously reconstructed to obtain the differential image.

METHODS

We have developed a SPECT reconstruction method that jointly reconstructs ictal and inter-ictal SPECT projection data. We performed both phantom and patient studies to evaluate the performance of our joint method for epileptic foci localization as compared with the conventional subtraction method in which the differential image is obtained by subtracting the inter-ictal image from the co-registered ictal image. Two low-noise SPECT projection datasets were acquired using ^99m Tc and a Hoffman head phantom at two different positions and orientations. At one of the two phantom locations, a low-noise dataset was also acquired using a ^99m Tc-filled 3.3-cm sphere with a cold attenuation background identical to the Hoffman phantom. These three datasets were combined and scaled to mimic low-noise clinical ictal (three different lesion-to-background contrast levels: 1.25, 1.55, and 1.70) and inter-ictal scans. For each low-noise dataset, 25 noise realizations were generated by adding Poisson noise to the projections. The mean and standard deviation (SD) of lesion contrast in the differential images were computed using both the conventional subtraction and our joint methods. We also applied both methods to the 35 epileptic patient datasets. Each differential image was presented to two nuclear medicine physicians to localize a lesion and specify a confidence level. The readers' data were analyzed to obtain the localized-response receiver operating characteristic (LROC) curves for both the subtraction and joint methods.

RESULTS

For the phantom study, the difference between the mean lesion contrast in the differential images obtained using the conventional subtraction versus our joint method decreases as the iteration number increases. Compared with the conventional subtraction approach, the SD reduction of lesion contrast at the 10^th iteration using our joint method ranges from 54.7% to 68.2% (P < 0.0005), and 33.8% to 47.9% (P < 0.05) for 2 and 4 million total inter-ictal counts, respectively. In the patient study, our joint method increases the area under LROC from 0.24 to 0.34 and from 0.15 to 0.20 for the first and second reader, respectively. We have demonstrated improved performance of our method as compared to the standard subtraction method currently used in clinical practice.

CONCLUSION

The proposed joint ictal/inter-ictal reconstruction method yields better performance for epileptic foci localization than the conventional subtraction method.

Depressive symptoms due to stroke are strongly predicted by the volume and location of the cerebral infarction, white matter hyperintensities, hypertension, and age: A precision nomothetic psychiatry analysis

OBJECTIVES

To delineate the effects of white matter hyperintensities (WMHs) as measured by Fluid-attenuated inversion recovery (FLAIR) and infarction volume as measured by Diffusion-weighted imaging (DWI) on post-stroke depression symptoms.

METHODS

Baseline National Institutes of Health Stroke Score (NIHSS) and Modified Rankin Scale (mRS) scores, and FLAIR and DWI MRIs to assess WMHs and acute infarct volumes, respectively, were assessed in 47 patients (≥55 years) with acute ischemic stroke and 17 normal controls. The Montgomery-Åsberg Depression Rating Scale (MDRS) was assessed three months after the stroke.

RESULTS

The MADRS score was significantly increased in stroke patients as compared with normal controls. The MADRS scale is not unidimensional and cannot be used as an accurate indicator of depression severity in stroke patients. Three months after stroke, key depressive (sadness and inability to feel) and concentration-tension symptoms, and lassitude are significantly predicted by the infarct volume. Right side infarction strongly predicts key depressive symptoms and left side infarction strongly predicts concentration-tension and lassitude scores. Total WMHs significantly predict key depressive and concentration-tension symptoms, and lassitude, with these effects being mediated by right and left DWI stroke volumes and associated disabilities.

CONCLUSIONS

Interactions between age, hypertension, a chronic atherosclerotic process, and acute stroke account for the onset of key depressive symptoms three months after the acute infarct. Chronic and acute neuro-immune and neuro-oxidative stress pathways associated with the formation of WMHs and acute stroke may explain the incidence of post-stroke key depressive and concentration-tension symptoms, and lassitude.

Impact of injection time on migration of SPECT seizure onset in temporal lobe epilepsy

In this work, we investigated an impact of injection time on migration of seizure-onset in ictal/interictal single photon emission computed tomography (SPECT) study for patients with temporal lobe epilepsy. We selected 33 patients with refractory temporal lobe epilepsy who underwent ictal/interictal SPECT studies and had preoperative intracranial EEG result or surgical resection which was used as reference for seizure location. We divided all patients into two groups, which are the fast and the delayed groups; the delayed group comprised patients with injection time more than a cutoff time and vice versa. Using the subtraction ictal-interictal SPECT co-registered with SPECT (SISCOS) with varied Z-threshold (1.0, 1.5, 2.0 and 2.5), a method similar to subtraction ictal SPECT co-registered to magnetic resonance imaging (MRI) (SISCOM), a seizure-onset region in the SISCOS image was localized at the region with maximum sum of Z-scores. For each pair of cutoff time and Z-threshold, we determined the migratory proportion which was defined as the proportion of patients whose seizure-onset location based on SISCOS image was discordant with the reference. At cutoff time of 32-35 seconds and the Z-threshold of 2.0, the migratory proportion values were 7/26 (26.9%) and 5/7 (71.4%) in the fast and the delayed groups, respectively. At the same range of cutoff time with the Z-threshold of 2.5, the migratory proportion was 8/26 (30.8%) in the fast group while the proportion was 5/7 (71.4%) in the delayed group. Using Fisher's exact test, the migratory proportion values at the Z-threshold of 2.0 and 2.5 were significantly different between the fast and the delayed groups (p = 0.0709 and 0.0838, respectively), suggesting that patients with temporal lobe epilepsy who undergo an ictal/interictal SPECT study with injection time longer than 35 seconds tend to have seizure-onset zone migration in the SISCOS analysis with the traditionally-used Z-threshold of 2.0.

Improved prediction of radiation-induced hypothyroidism in nasopharyngeal carcinoma using pre-treatment CT radiomics

When planning radiation therapy, late effects due to the treatment should be considered. One of the most common complications of head and neck radiation therapy is hypothyroidism. Although clinical and dosimetric data are routinely used to assess the risk of hypothyroidism after radiation, the outcome is still unsatisfactory. Medical imaging can provide additional information that improves the prediction of hypothyroidism. In this study, pre-treatment computed tomography (CT) radiomics features of the thyroid gland were combined with clinical and dosimetric data from 220 participants to predict the occurrence of hypothyroidism within 2 years after radiation therapy. The findings demonstrated that the addition of CT radiomics consistently and significantly improves upon conventional model, achieving the highest area under the receiver operating characteristic curve (AUCs) of 0.81 ± 0.06 with a random forest model. Hence, pre-treatment thyroid CT imaging provides useful information that have the potential to improve the ability to predict hypothyroidism after nasopharyngeal radiation therapy.

Quantification and normalization of noise variance with sparsity regularization to enhance diffuse optical tomography

Conventional reconstruction of diffuse optical tomography (DOT) is based on the Tikhonov regularization and the white Gaussian noise assumption. Consequently, the reconstructed DOT images usually have a low spatial resolution. In this work, we have derived a novel quantification method for noise variance based on the linear Rytov approximation of the photon diffusion equation. Specifically, we have implemented this quantification of noise variance to normalize the measurement signals from all source-detector channels along with sparsity regularization to provide high-quality DOT images. Multiple experiments from computer simulations and laboratory phantoms were performed to validate and support the newly developed algorithm. The reconstructed images demonstrate that quantification and normalization of noise variance with sparsity regularization (QNNVSR) is an effective reconstruction approach to greatly enhance the spatial resolution and the shape fidelity for DOT images. Since noise variance can be estimated by our derived expression with relatively limited resources available, this approach is practically useful for many DOT applications.

Analysis of dynamic antral scintigraphy using empirical mode decomposition

In this work, we proposed to use the empirical mode decomposition (EMD) to analyze signals from dynamic antral scintigraphy (DAS) for antral contraction rate estimation. The proposed EMD-Fourier method analyzes a DAS time-activity curve (TAC) which is called a DAS signal by estimating the frequency from the Fourier transform of each intrinsic mode function (IMF), thus yielding multiple frequency values for each TAC as opposed to one frequency value obtained from the conventional Fourier method. Twenty-three TACs extracted from DAS data acquired on twenty-three healthy volunteers were analyzed using both the EMD-Fourier method and the Fourier method. The mean and standard deviation of frequency across all volunteers were computed. The result showed that the EMD-Fourier method provided a reduction of SD-to-mean ratio from the Fourier method ranging from 64.3% to 85.8%. The mean frequencies from the third IMF in the EMD-Fourier method and from the Fourier method were 2.76 and 3.44 cycles per minute were close to a widely-used normal antral contraction rate (3.0 cycles per minute), while the EMD-Fourier method yielded significant SD reduction (from 4.77 to 0.57: p <; 0.0001). Moreover, it was found that the first IMF yielded a frequency estimate of 11.26 ± 2.2 cycles per minute, while the second IMF yielded a frequency estimate of 4.3 ± 1.3 cycles per minute, which are similar to the contraction rates of the duodenum and the large intestine, respectively.

A multivariable normal tissue complication probability model for predicting radiation-induced hypothyroidism in nasopharyngeal carcinoma patients in the modern radiotherapy era

Radiation-induced hypothyroidism (RHT) is a common long-term complication for nasopharyngeal carcinoma (NPC) survivors. A model using clinical and dosimetric factors for predicting risk of RHT could suggest a proper dose-volume parameters for the treatment planning in an individual level. We aim to develop a multivariable normal tissue complication probability (NTCP) model for RHT in NPC patients after intensity-modulated radiotherapy or volumetric modulated arc therapy. The model was developed using retrospective clinical data and dose-volume data of the thyroid and pituitary gland based on a standard backward stepwise multivariable logistic regression analysis and was then internally validated using 10-fold cross-validation. The final NTCP model consisted of age, pretreatment thyroid-stimulating hormone and mean thyroid dose. The model performance was good with an area under the receiver operating characteristic curve of 0.749 on an internal (200 patients) and 0.812 on an external (25 patients) validation. The mean thyroid dose at ≤45 Gy was suggested for treatment plan, owing to an RHT incidence of 2% versus 61% in the >45 Gy group.

Subtraction of single-photon emission computed tomography (SPECT) in radioembolization: a comparison of four methods

BACKGROUND

Subtraction of single-photon emission computed tomography (SPECT) images has a number of clinical applications in e.g. foci localization in ictal/inter-ictal SPECT and defect detection in rest/stress cardiac SPECT. In this work, we investigated the technical performance of SPECT subtraction for the purpose of quantifying the effect of a vasoconstricting drug (angiotensin-II, or AT2) on the Tc-99m-MAA liver distribution in hepatic radioembolization using an innovative interventional hybrid C-arm scanner. Given that subtraction of SPECT images is challenging due to high noise levels and poor resolution, we compared four methods to obtain a difference image in terms of image quality and quantitative accuracy. These methods included (i) image subtraction: subtraction of independently reconstructed SPECT images, (ii) projection subtraction: reconstruction of a SPECT image from subtracted projections, (iii) projection addition: reconstruction by addition of projections as a background term during the iterative reconstruction, and (iv) image addition: simultaneous reconstruction of the difference image and the subtracted image.

RESULTS

Digital simulations (XCAT) and phantom studies (NEMA-IQ and anthropomorphic torso) showed that all four methods were able to generate difference images but their performance on specific metrics varied substantially. Image subtraction had the best quantitative performance (activity recovery coefficient) but had the worst visual quality (contrast-to-noise ratio) due to high noise levels. Projection subtraction showed a slightly better visual quality than image subtraction, but also a slightly worse quantitative accuracy. Projection addition had a substantial bias in its quantitative accuracy which increased with less counts in the projections. Image addition resulted in the best visual image quality but had a quantitative bias when the two images to subtract contained opposing features.

CONCLUSION

All four investigated methods of SPECT subtraction demonstrated the capacity to generate a feasible difference image from two SPECT images. Image subtraction is recommended when the user is only interested in quantitative values, whereas image addition is recommended when the user requires the best visual image quality. Since quantitative accuracy is most important for the dosimetric investigation of AT2 in radioembolization, we recommend using the image subtraction method for this purpose.

Pharmacokinetic Modeling of 18F-FDOPA PET in the Human Brain for Early Parkinson's Disease

Objectives:

Early detection is essential for the treatment approaches of Parkinson’s disease (PD). Clinical criteria alone may be insufficient to distinguish early PD from other conditions. This study aimed to investigate the transfer rate constants of 6-¹⁸F-fluoro-L-dopa (¹⁸F-FDOPA) in positron emission tomography (PET) brain images as a sensitive parameter to detect early PD.

Methods:

Retrospective ¹⁸F-FDOPA PET data of five patients with early PD were collected. PET data were acquired for 90 min after intravenous injection of 306-379 MBq ¹⁸F-FDOPA, and reconstructed into a series of 18 five-minute frames. Reoriented PET images were coregistered and normalized with the PET brain template on the statistical parametric mapping. The ¹⁸F-FDOPA activity concentrations were measured in the striatum, caudate, and putamen on both sides: Contralateral (as PD) and ipsilateral (as control) to the main motor symptoms. The pharmacokinetic model was generated using the SAAM II simulation software. The transfer rate constants across the blood-brain barrier (forward, K₁ and reverse, k₂) and decarboxylation rate constants (k₃) were estimated in these regions.

Results:

The activity uptakes in the contralateral striatum (0.0323%±0.0091%) and putamen (0.0169%±0.0054%) were significantly lower than the control (0.0353%±0.0086%, 0.0199%±0.0054%, respectively). The K₁ and k₃ were significantly lower in the contralateral striatum and putamen (p<0.05). There were no significant differences in any transfer rate constants in the caudate.

Conclusion:

The transfer rate constants (K₁ and k₃) of ¹⁸F-FDOPA on the contralateral striatum and putamen were significantly lower than the control. These biokinetic data could be potential indicators for quantitative detection of early PD diagnosis.

Preoperative parathyroid localization using joint planar imaging

The conventional approach for preoperative parathyroid localization with the dual ^99mTc-sestamibi (^99mTc-MIBI) and ^99mTc-pertechnetate (^99mTcO₄^- ) scintigraphy technique obtains the differential image by subtracting images from the two scans; the former depicts both the thyroid and parathyroid glands while the latter depicts the thyroid only. In this study, we developed a novel method based on Poisson noise modeling and maximum-likelihood estimation to generate the differential image in an iterative fashion using both planar images jointly. We demonstrated improved performance of our joint method as compared with the subtraction method in both phantom and patient studies. In the phantom study, we acquired two noise-free planar datasets using ^99mTc on an in-house thyroid phantom and a "lesion" bead (representing a parathyroid gland) with the same attenuation background as the thyroid phantom. These two planar datasets were combined and scaled to simulate noise-free clinical MIBI (four lesion-to-background contrast (LBC) values: 1.2, 1.3, 1.4 and 1.5), and ^99mTcO₄^- datasets. One-hundred Poisson noise realizations were generated for each datasets. The mean and standard deviation (SD) of the lesion contrast in the differential images were computed for both the subtraction and the joint methods. We also applied both the subtraction and the joint methods to one parathyroid patient dataset. The voxel-wise mean-to-SD ratios in four hyperfunctioning parathyroid lesions were calculated. The phantom results showed that the joint method at the 50^th iteration yielded a significant SD reduction compared with the subtraction method ranging from 20% to 45% (p <; 0.05). Similarly, the voxel-wise mean-to-SD ratios were substantially improved in the patient study from 0.40-1.60 (subtraction) to 2.68-3.16 (joint).