Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “SLICING”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,549 records · Page 86

WE-AB-BRA-08: Correction of Patient Motion in C-Arm Cone-Beam CT Using 3D-2D Registration

Purpose: Intraoperative C-arm cone-beam CT (CBCT) is subject to artifacts arising from patient motion during the fairly long (∼5–20 s) scan times. We present a fiducial free method to mitigate motion artifacts using 3D-2D image registration that simultaneously corrects residual errors in geometric calibration. Methods: A 3D-2D registration process was used to register each projection to DRRs computed from the 3D image by maximizing gradient orientation (GO) using the CMA-ES optimizer. The resulting rigid 6 DOF transforms were applied to the system projection matrices, and a 3D image was reconstructed via model-based image reconstruction (MBIR, which accommodates the resulting noncircular orbit). Experiments were conducted using a Zeego robotic C-arm (20 s, 200°, 496 projections) to image a head phantom undergoing various types of motion: 1) 5° lateral motion; 2) 15° lateral motion; and 3) 5° lateral motion with 10 mm periodic inferior-superior motion. Images were reconstructed using a penalized likelihood (PL) objective function, and structural similarity (SSIM) was measured for axial slices of the reconstructed images. A motion-free image was acquired using the same protocol for comparison. Results: There was significant improvement (p < 0.001) in the SSIM of the motion-corrected (MC) images compared to uncorrected images. The SSIM in MC-PL images was >0.99, indicating near identity to the motion-free reference. The point spread function (PSF) measured from a wire in the phantom was restored to that of the reference in each case. Conclusion: The 3D-2D registration method provides a robust framework for mitigation of motion artifacts and is expected to hold for applications in the head, pelvis, and extremities with reasonably constrained operative setup. Further improvement can be achieved by incorporating multiple rigid components and non-rigid deformation within the framework. The method is highly parallelizable and could in principle be run with every acquisition. Research supported by National Institutes of Health Grant No. R01-EB-017226 and academic-industry partnership with Siemens Healthcare (AX Division, Forcheim, Germany).

60 APPLIED LIFE SCIENCES↗

WE-FG-202-11: Longitudinal Diffusion MRI for Treatment Assessment of Sarcoma Patients with Pre-Operative Radiation Therapy

Purpose: Diffusion weighted MRI (DWI) is a promising imaging technique for early prediction of tumor response to radiation therapy. A recently proposed longitudinal DWI strategy using a Co-60 MRI guided RT system (MRIgRT) may bring functional MRI guided adaptive radiation therapy closer to clinical utility. We report our preliminary results of using this longitudinal DWI approach performed on the MRIgRT system for predicting the response of sarcoma patient to preop RT. Methods: Three sarcoma patients who underwent fractionated IMRT were recruited in this study. For all three patients DWI images were acquired immediately following his/her treatment. For each imaging session, ten slices were acquired interleaved with the b values covering the gross tumor volume (GTV). The diffusion images were processed to obtain the ADC maps using standard exponential fitting for each voxel. Regions of interest were drawn in the tumor on the diffusion images based on each patient’s clinical GTV contours. Each patient subsequently underwent surgery and the tumor necrosis score was available from standard pathology. The ADC values for each patient were compared to the necrosis scores to assess the predictive value of our longitudinal DWI for tumor response. Results: Each patient underwent 3 to 5 diffusion MRI scans depending on their treatment length. Patient 1 had a relatively unchanged ADC during the course of RT and a necrosis score of 30% at surgery. For patient 2, the mean ADC values decreased from 1.56 × 10-3 to 1.12 × 10-3 mm2/s and the patient’s necrosis score was less than 10%. Patient 3 had a slight increase in the ADC values from 0.59 × 10-3 to 0.71 × 10-3 mm2/s and patient’s necrosis score was 50%. Conclusion: Based on limited data from 3 patients, our longitudinal changes in tumor ADC assessed using the MRIgRT system correlated well with pathology results.

60 APPLIED LIFE SCIENCES↗

WE-FG-206-05: New Arterial Spin Labeling Method for Simultaneous Estimation of Arterial Cerebral Blood Volume, Cerebral Blood Flow and Arterial Transit Time

Purpose: To demonstrate the feasibility of a novel Arterial Spin Labeling (ASL) method for simultaneously measuring cerebral blood flow (CBF), arterial transit time (ATT), and arterial cerebral blood volume (aCBV) without the use of a contrast agent. Methods: A series of multi-TI ASL images were acquired from one healthy subject on a 3T Siemens Skyra, with the following parameters: PCASL labeling with variable TI [300, 400, 500, 600, 700, 800, 900, 1000, 1500, 2000, 2500, 3000, 3500, 4000] ms, labeling bolus 1400 ms when TI allows, otherwise 100 ms less than TI, TR was minimized for each TI, two sinc shaped pre-saturation pulses were applied in the imaging plane immediately before 2D EPI acquisition. 64×64×24 voxels, 5 mm slice thickness, 1 mm gap, full brain coverage, 6 averages per TI, no crusher gradients, 11 ms TE, scan time of 4:56. The perfusion weighted time-series was created for each voxel and fit to a novel model. The model has two components: 1) the traditional model developed by Buxton et al., accounting for CBF and ATT, and 2) a box car function characterizing the width of the labeling bolus, with variable timing and height in proportion to the aCBV. All three parameters were fit using a nonlinear fitting routine that constrained all parameters to be positive. The main purpose of the high-temporal resolution TI sampling for the first second of data acquisition was to precisely estimate the blood volume component for better detection of arrival time and magnitude of signal. Results: Whole brain maps of CBF, ATT, and aCBV were produced, and all three parameters maps are consistent with similar maps described in the literature. Conclusion: Simultaneous mapping of CBF, ATT, and aCBV is feasible with a clinically tractable scan time (under 5 minutes).

62 RADIOLOGY AND NUCLEAR MEDICINE↗

WE-FG-207B-06: Plaque Composition Measurement with Dual Energy Computed Tomography

Purpose: To investigate the feasibility of characterizing arterial plaque composition in terms of water, lipid and protein or calcium using dual energy computed tomography. Characterization of plaque composition can potentially help distinguish vulnerable from stable plaques. Methods: Simulations studies were performed by the CT simulator based on ASTRA tomography toolbox. The beam energy for dual energy images was selected to be 80 kVp and 135 kVp. The radiation dose and energy spectrum for the CT simulator were carefully calibrated with respect to a 320-slice CT scanner. A digital chest phantom was constructed using Matlab for calibration and plaque measurement. Pure water, lipid, protein or calcium was used for calibration and a mixture of different volume percentages of these materials were used for validation purposes. Non-calcified plaque was simulated using water, lipid and protein with volumetric percentage range of 35%∼65%, 5%∼60% and 5%∼40%, respectively. Calcified plaque was simulated using water, lipid and calcium with volumetric percentage range of 50%∼80%, 8%∼45% and 3%∼13%, respectively. We employed iterative sinogram processing (ISP) to reduce the beam hardening effect in the simulation to improve the decomposition results. Results: The simulated known composition and dual energy decomposition results were in good agreement. Water, lipid and protein (calcium) mixtures were decomposed into water, lipid and protein (calcium) contents. The RMS errors of volumetric percentage for the water, lipid and protein (non-calcified plaque) decomposition, as compared to known values, were estimated to be approximately 5.74%, 2.54%, and 0.95% respectively. The RMS errors of volumetric percentage for the water, lipid and Calcium (calcified plaque) decomposition, as compared to known values, were estimated to be approximately 7.4%, 8.64%, and 0.08% respectively. Conclusion: The results of this study suggest that the dual energy decomposition can potentially be used to quantify the water, lipid, and protein or calcium composition of a plaque with relatively good accuracy. Grant funding from Toshiba Medical Systems and Philips Medical Systems.

60 APPLIED LIFE SCIENCES↗

TH-AB-BRA-11: Using 3D Dosimeters for the Investigation of the Electron Return Effect (ERE) in MR-Guided Radiation Therapy: A Feasibility Study

Purpose: To demonstrate the capability of 3D radiochromic PRESAGE and Fricke-type dosimeters to measure the influence of magnetic fields on dose distribution, including the electron return effect (ERE), for MR-guided radiation therapy applications. Methods: Short cylindrical 3D dosimeters with PRESAGE and Fricke-type formulations were created in-house prior to irradiations in a 1.5T/7MV MR-linac. Each dosimeter was prepared with a concentric cylindrical air cavity with diameters of 1.5 cm and 2.5 cm, and the diameters of the dosimeters were 7.2 cm and 8.8 cm for PRESAGE and Fricke-type respectively. The dosimeters were irradiated within the bore of the MR-linac with the flat face of the dosimeters perpendicular to the magnetic field. Dosimeters were irradiated to approximately 9 Gy and 29 Gy to the center of dosimeters with a 15×15 cm{sup 2} field. The PRESAGE dosimeter was scanned using an optical-CT 2 hours post-irradiation; the Fricke-type dosimeter was immediately imaged with the MR component of the MR-linac post-irradiation. Results: Axial slices of the dose distributions show a clear demonstration of the dose enhancement due to the ERE above the cavity and the region of reduced dose below the cavity. The regions of increased and reduced dose are rotated with respect to the radiation beam axis due to the average directional change of the electrons. Measurements from line profiles show the dose enhanced up to ∼0.5 cm around the cavity by up to a factor of 1.3 and 1.4 for PRESAGE and Fricke-type dosimeters respectively. Conclusion: PRESAGE and Fricke-type dosimeters are able to qualitatively measure the ERE with good agreement with previously published simulation and 2D dosimetry demonstrations of the ERE. Further investigation of these 3D dosimeters as promising candidates for quality assurance of MR-guided radiation therapy systems is encouraged to assess changes in response and measurement accuracy due to the magnetic field.

60 APPLIED LIFE SCIENCES↗

TH-AB-207A-11: Anatomical and Functional Assessment of Coronary Artery Disease Using Low-Dose Whole-Organ Dynamic Computed Tomography

Purpose: To evaluate the accuracy of a low-dose, first-pass-analysis (FPA) dynamic computed tomography angiography and perfusion (CTAP) technique, for whole-organ anatomical and functional assessment of coronary artery disease (CAD). Methods: An angioplasty balloon was advanced into the left anterior descending (LAD) coronary artery of five swine (35–45 kg) to induce several levels of stenosis at maximal hyperemia (intracoronary adenosine, 240 µg/min). Reference fluorescence microspheres and intravenous contrast (370 mg/mL iodine, 25 mL, 7 mL/s) were injected centrally and dynamic imaging was performed using a 320-slice CT scanner at 100 kVp and 200 mA. Twenty volume scans were acquired per stenosis level to capture complete aortic and myocardial enhancement curves, but only two volume scans were used for whole-organ dynamic FPA CTAP measurement. All CTAP measurements in the LAD were compared to the reference microsphere perfusion measurements using linear regression, concordance correlation, and Bland-Altman analysis. Results: The result of dynamic FPA CTAP measurement in the LAD was in good agreement with the reference microsphere perfusion measurement (P-CTAP = 1.01 P-MICRO + 0.16, R{sup 2} = 0.95). The root mean square error (RMSE) and difference (RMSD) of measurement were 0.51 mL/min/g and 0.47 mL/min/g, respectively. Bland-Altman analysis demonstrated negligible systematic measurement bias. Additionally, the concordance correlation coefficient (CCC) was found to be ρ = 0.97, indicating excellent agreement between dynamic FPA CTAP measurement and the reference microsphere perfusion measurement. Lastly, the effective dose of the proposed technique using the “simulated” two-volume scan CTAP acquisition protocol was 2.6 mSv; much lower than the ∼10 mSv effective dose of current dynamic CTP techniques alone. Conclusion: The results indicate the potential for significant improvements in CAD assessment through low-dose, quantitative dynamic FPA CTAP. Such improvements are afforded through whole-organ CT scanning technology, and have the potential to improve patient outcomes and lead to healthier patient lives. Conflict of Interest (only if applicable): Grant funding from Toshiba America Medical Systems.

60 APPLIED LIFE SCIENCES↗

TH-CD-206-07: Determination of Patient-Specific Myocardial Mass at Risk Using Computed Tomography Angiography

Purpose: To evaluate the accuracy of a patient-specific coronary perfusion territory assignment algorithm that uses CT angiography (CTA) and a minimum-cost-path approach to assign coronary perfusion territories on a voxel-by-voxel basis for determination of myocardial mass at risk. Methods: Intravenous (IV) contrast (370 mg/mL iodine, 25 mL, 7 mL/s) was injected centrally into five swine (35–45 kg) and CTA was performed using a 320-slice CT scanner at 100 kVp and 200 mA. Additionally, a 4F catheter was advanced into the left anterior descending (LAD), left circumflex (LCX), and right coronary artery (RCA) and contrast (30 mg/mL iodine, 10 mL, 1.5 mL/s) was directly injected into each coronary artery for isolation of reference coronary perfusion territories. Semiautomatic myocardial segmentation of the CTA data was then performed and the centerlines of the LAD, LCX, and RCA were digitally extracted through image processing. Individual coronary perfusion territories were then assigned using a minimum-cost-path approach, and were quantitatively compared to the reference coronary perfusion territories. Results: The results of the coronary perfusion territory assignment algorithm were in good agreement with the reference coronary perfusion territories. The average volumetric assignment error from mitral orifice to apex was 5.5 ± 1.1%, corresponding to 2.1 ± 0.7 grams of myocardial mass misassigned for each coronary perfusion territory. Conclusion: The results indicate that accurate coronary perfusion territory assignment is possible on a voxel-by-voxel basis using CTA data and an assignment algorithm based on a minimum-cost-path approach. Thus, the technique can potentially be used to accurately determine patient-specific myocardial mass at risk distal to a coronary stenosis, improving coronary lesion assessment and treatment. Conflict of Interest (only if applicable): Grant funding from Toshiba America Medical Systems.

60 APPLIED LIFE SCIENCES↗

TH-CD-207B-11: Multi-Vendor Phantom Study of CT Lung Density Metrics: Is a Reproducibility of Less Than 1 HU Achievable?

Purpose: To standardize the calibration procedures of CT lung density measurements using low-density reference foams in a phantom, and to demonstrate a reproducibility of less than 1 HU for lung equivalent foam densities measured across CT vendor platforms and protocols. Methods: A phantom study was conducted on CT scanner models from 4 vendors at 100, 120, and 135/140 kVp and 1.5, 3, and 6 mGy dose settings, using a lung density phantom containing air, water, and 3 reference foams (indirectly calibrated) with discrete densities simulating a 5-cm slice of the human chest. Customized segmentation software was used to analyze the images and generate a mean HU and variance for each of the density for the 22 vendor/protocols. A 3-step calibration process was devised to remove a scanner-dependent parameter using linear regression of the HU value vs the relative electron density. The results were mapped to a single energy (80 keV) for final comparison. Results: The heterogeneity across vendor platforms for each density assessed by a random effects model was reduced by 50% after re-calibration, while the standard deviation of the mean HU values also improved by about the same amount. The 95% CI of the final HU value was within +/−1 HU for all 3 reference foam densities. For the backing lung foam in the phantom (served as an “unknown”), this CI is +/− 1.6 HU. The kVp and dose settings did not appear to have significant contributions to the variability. Conclusion: With the proposed calibration procedures, the inter-scanner reproducibility of better than 1 HU is demonstrated in the current phantom study for the reference foam densities, but not yet achieved for a test density. The sources of error are being investigated in the next round of scanning with a certified Standard Reference Material for direct calibration. Fain: research funding from GE Healthcare to develop pulmonary MRI techniques. Hoppel: employee of Toshiba Medical Research Institute USA/financial interest with GE Healthcare. M. Fuld: employee of Siemens Healthcare for medical device equipment and software. This project is supported partially by RSNA QIBA Concept Award (Fain), NIH/NIBIB, HHSN268201300071C (Y).

60 APPLIED LIFE SCIENCES↗

TH-EF-207A-03: Photon Counting Implementation Challenges Using An Electron Multiplying Charged-Coupled Device Based Micro-CT System

Purpose: To successfully implement and operate a photon counting scheme on an electron multiplying charged-coupled device (EMCCD) based micro-CT system. Methods: We built an EMCCD based micro-CT system and implemented a photon counting scheme. EMCCD detectors use avalanche transfer registries to multiply the input signal far above the readout noise floor. Due to intrinsic differences in the pixel array, using a global threshold for photon counting is not optimal. To address this shortcoming, we generated a threshold array based on sixty dark fields (no x-ray exposure). We calculated an average matrix and a variance matrix of the dark field sequence. The average matrix was used for the offset correction while the variance matrix was used to set individual pixel thresholds for the photon counting scheme. Three hundred photon counting frames were added for each projection and 360 projections were acquired for each object. The system was used to scan various objects followed by reconstruction using an FDK algorithm. Results: Examination of the projection images and reconstructed slices of the objects indicated clear interior detail free of beam hardening artifacts. This suggests successful implementation of the photon counting scheme on our EMCCD based micro-CT system. Conclusion: This work indicates that it is possible to implement and operate a photon counting scheme on an EMCCD based micro-CT system, suggesting that these devices might be able to operate at very low x-ray exposures in a photon counting mode. Such devices could have future implications in clinical CT protocols. NIH Grant R01EB002873; Toshiba Medical Systems Corp.

60 APPLIED LIFE SCIENCES↗

TH-EF-207A-04: A Dynamic Contrast Enhanced Cone Beam CT Technique for Evaluation of Renal Functions

Purpose: To develop a simple but robust method for the early detection and evaluation of renal functions using dynamic contrast enhanced cone beam CT technique. Methods: Experiments were performed on an integrated imaging and radiation research platform developed by our lab. Animals (n=3) were anesthetized with 20uL Ketamine/Xylazine cocktail, and then received 200uL injection of iodinated contrast agent Iopamidol via tail vein. Cone beam CT was acquired following contrast injection once per minute and up to 25 minutes. The cone beam CT was reconstructed with a dimension of 300×300×800 voxels of 130×130×130um voxel resolution. The middle kidney slices in the transvers and coronal planes were selected for image analysis. A double exponential function was used to fit the contrast enhanced signal intensity versus the time after contrast injection. Both pixel-based and region of interest (ROI)-based curve fitting were performed. Four parameters obtained from the curve fitting, namely the amplitude and flow constant for both contrast wash in and wash out phases, were investigated for further analysis. Results: Robust curve fitting was demonstrated for both pixel based (with R{sup 2}>0.8 for >85% pixels within the kidney contour) and ROI based (R{sup 2}>0.9 for all regions) analysis. Three different functional regions: renal pelvis, medulla and cortex, were clearly differentiated in the functional parameter map in the pixel based analysis. ROI based analysis showed the half-life T1/2 for contrast wash in and wash out phases were 0.98±0.15 and 17.04±7.16, 0.63±0.07 and 17.88±4.51, and 1.48±0.40 and 10.79±3.88 minutes for the renal pelvis, medulla and cortex, respectively. Conclusion: A robust method based on dynamic contrast enhanced cone beam CT and double exponential curve fitting has been developed to analyze the renal functions for different functional regions. Future study will be performed to investigate the sensitivity of this technique in the detection of radiation induced kidney dysfunction.

60 APPLIED LIFE SCIENCES↗

RASER MRI: Magnetic resonance images formed spontaneously exploiting cooperative nonlinear interaction

The spatial resolution of magnetic resonance imaging (MRI) is limited by the width of Lorentzian point spread functions associated with the transverse relaxation rate 1/T 2 * . Here, we show a different contrast mechanism in MRI by establishing RASER (radio-frequency amplification by stimulated emission of radiation) in imaged media. RASER imaging bursts emerge out of noise and without applying radio-frequency pulses when placing spins with sufficient population inversion in a weak magnetic field gradient. Small local differences in initial population inversion density can create stronger image contrast than conventional MRI. This different contrast mechanism is based on the cooperative nonlinear interaction between all slices. On the other hand, the cooperative nonlinear interaction gives rise to imaging artifacts, such as amplitude distortions and side lobes outside of the imaging domain. Contrast mechanism and artifacts are explored experimentally and predicted by simulations on the basis of a proposed RASER MRI theory.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

The Crystal Structure of Human Argonaute2

Argonaute proteins form the functional core of the RNA-induced silencing complexes that mediate RNA silencing in eukaryotes. The 2.3 angstrom resolution crystal structure of human Argonaute2 (Ago2) reveals a bilobed molecule with a central cleft for binding guide and target RNAs. Nucleotides 2 to 6 of a heterogeneous mixture of guide RNAs are positioned in an A-form conformation for base pairing with target messenger RNAs. Between nucleotides 6 and 7, there is a kink that may function in microRNA target recognition or release of sliced RNA products. Tandem tryptophan-binding pockets in the PIWI domain define a likely interaction surface for recruitment of glycine-tryptophan-182 (GW182) or other tryptophan-rich cofactors. These results will enable structure-based approaches for harnessing the untapped therapeutic potential of RNA silencing in humans.

59 BASIC BIOLOGICAL SCIENCES↗

Structural basis for microRNA targeting

MicroRNAs (miRNAs) control expression of thousands of genes in plants and animals. miRNAs function by guiding Argonaute proteins to complementary sites in messenger RNAs (mRNAs) targeted for repression. In this paper, we determined crystal structures of human Argonaute-2 (Ago2) bound to a defined guide RNA with and without target RNAs representing miRNA recognition sites. These structures suggest a stepwise mechanism, in which Ago2 primarily exposes guide nucleotides (nt) 2 to 5 for initial target pairing. Pairing to nt 2 to 5 promotes conformational changes that expose nt 2 to 8 and 13 to 16 for further target recognition. Interactions with the guide-target minor groove allow Ago2 to interrogate target RNAs in a sequence-independent manner, whereas an adenosine binding-pocket opposite guide nt 1 further facilitates target recognition. Spurious slicing of miRNA targets is avoided through an inhibitory coordination of one catalytic magnesium ion. Finally, these results explain the conserved nucleotide-pairing patterns in animal miRNA target sites first observed over two decades ago.

59 BASIC BIOLOGICAL SCIENCES↗

Magnetic resonance spectroscopy of the human brain

I give a brief description of the magnetic resonance spectroscopy (MRS) in the human brain examinations. MRS allows a noninvasive chemical analysis of the brain using a standard high field MR system. Nowadays, the dominant form of MR brain spectroscopy is proton spectroscopy. Two main techniques of MRS, which utilize the chemical shift of metabolites in the external magnetic field, are SVS (single voxel) and CSI (single slice). The major peaks in the spectrum of a normal brain include NAA, Cr, Cho and m-Ins, which are neuronal, energetic, membrane turnover and glial markers, respectively. In disease, two pathological metabolites can be found in the brain spectra: Lac, which is end product of anaerobic glycolysis and Lip, which is a marker of membrane breakdown, occurring in necrosis. The common way to analyze clinical spectra is to determine metabolite ratios, e.g. NAA/Cr, Cho/Cr, Cho/NAA. This analysis permits a safe and noninvasive examination of the brain tissue as each disease state has its own characteristic spectroscopic image. MRS is a valuable diagnostic tool in such clinical applications as detecting brain tumors and differentiating tumors from inflammatory and infectious processes. Proton MRS is also very helpful in diagnostic of ischemic lesions, Alzheimer’s disease and hepatic encephalopathy. The MRS brain spectra should always be correlated with the Magnetic Resonance Imaging (MRI) results and alone cannot make neurological diagnosis.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Pure Yang–Mills Solutions on dS{sub 4}

We consider pure SU(2) Yang–Mills theory on four-dimensional de Sitter space dS{sub 4} and construct smooth and spatially homogeneous classical Yang–Mills fields. Slicing dS{sub 4} as R×S{sup 3}, via an SU(2)-equivariant ansatz we reduce the Yang–Mills equations to ordinary matrix differential equations and further to Newtonian dynamics in a particular three-dimensional potential. Its classical trajectories yield spatially homogeneous Yang–Mills solutions in a very simple explicit form, depending only on de Sitter time with an exponential decay in the past and future. These configurations have not only finite energy, but their action is also finite and bounded from below. We present explicit coordinate representations of the simplest examples (for the fundamental SU(2) representation). Instantons (Yang–Mills solutions on the Wick-rotated S{sup 4} ) and solutions on AdS{sub 4} are also briefly discussed.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

A study in three-dimensional chaotic dynamics: Granular flow and transport in a bi-axial spherical tumbler

We study three-dimensional (3D) chaotic dynamics through an analysis of transport in a granular flow in a half-full spherical tumbler rotated sequentially about two orthogonal axes (a bi-axial “blinking” tumbler). The flow is essentially quasi-two-dimensional in any vertical slice of the sphere during rotation about a single axis, and we provide an explicit exact solution to the model in this case. Hence, the cross-sectional flow can be represented by a twist map, allowing us to express the 3D flow as a linked twist map (LTM). We prove that if the rates of rotation about each axis are equal, then (in the absence of stochasticity) particle trajectories are restricted to two-dimensional (2D) surfaces consisting of a portion of a hemispherical shell closed by a “cap''; if the rotation rates are unequal, then particles can leave the surface they start on and traverse a volume of the tumbler. The period-one structures of the governing LTM are examined in detail: analytical expressions are provided for the location of period-one curves, their extent into the bulk of the granular material, and their dependence on the protocol parameters (rates and durations of rotations). Exploiting the restriction of trajectories to 2D surfaces in the case of equal rotation rates about the axes, a method is proposed for identifying and constructing 3D Kolmogorov--Arnold--Moser (KAM) tubes around the normally elliptic period-one curves. The invariant manifold structure arising from the normally hyperbolic period-one curves is also examined. When the motion is restricted to 2D surfaces, the structure of manifolds of the hyperbolic points in the bulk differs from that corresponding to hyperbolic points in the flowing layer. Each is reminiscent of a template provided by a non-integrable perturbation to a Hamiltonian system, though the governing LTM is not. This highlights the novel 3D chaotic behaviors observed in this model dynamical system.

42 ENGINEERING↗

Generative Ensemble Regression: Learning Particle Dynamics from Observations of Ensembles with Physics-informed Deep Generative Models

Here, we propose a new method for inferring the governing stochastic ordinary differential equations (SODEs) by observing particle ensembles at discrete and sparse time instants, i.e., multiple “snapshots.” Particle coordinates at a single time instant, possibly noisy or truncated, are recorded in each snapshot but are unpaired across the snapshots. By training a physics-informed generative model that generates “fake” sample paths, we aim to fit the observed particle ensemble distributions with a curve in the probability measure space, which is induced from the inferred particle dynamics. We employ different metrics to quantify the differences between distributions, e.g., the sliced Wasserstein distances and the adversarial losses in generative adversarial networks. We refer to this method as generative “ensemble-regression” (GER), in analogy to the classic “point-regression,” where we infer the dynamics by performing regression in the Euclidean space. We illustrate the GER by learning the drift and diffusion terms of particle ensembles governed by SODEs with Brownian motions and Lévy processes up to 100 dimensions. We also discuss how to treat cases with noisy or truncated observations. Apart from systems consisting of independent particles, we also tackle nonlocal interacting particle systems with unknown interaction potential parameters by constructing a physics-informed loss function. Finally, we investigate scenarios of paired observations and discuss how to reduce the dimensionality in such cases by proving a convergence theorem that provides theoretical support.

97 MATHEMATICS AND COMPUTING↗

Score-Based Physics-Informed Neural Networks for High-Dimensional Fokker–Planck Equations

The Fokker-Planck (FP) equation is a foundational partial differential equation (PDE) in stochastic processes involving Brownian motions. However, the curse of dimensionality (CoD) poses a formidable challenge when dealing with high-dimensional FP equations. Although Monte Carlo simulation and (vanilla) Physics-Informed Neural Networks (PINNs) have shown the potential to tackle CoD, both methods exhibit significant numerical errors in high dimensions when dealing with the probability density function (PDF) associated with Brownian motion. The point-wise PDF values tend to decrease exponentially as dimensionality increases, surpassing the precision of numerical simulations and resulting in substantial errors. In addition, due to its massive sampling, Monte Carlo fails to offer fast sampling. Modeling the logarithm likelihood (LL) via vanilla PINNs transforms the FP equation into a notoriously difficult Hamilton-Jacobi-Bellman (HJB) equation, which is impractical for PINN learning, whose error grows rapidly with dimension. To this end, we propose a novel approach utilizing a score-based solver to fit the score function in stochastic differential equations (SDEs). The score function, defined as the gradient of the LL, plays a fundamental role in inferring LL and PDF and enables fast SDE sampling, offering an effective means to overcome the CoD. Three fitting methods, Score Matching (SM), Sliced Score Matching (SSM), and Score-PINN, are introduced, each contributing unique advantages in computational complexity, accuracy, and generality. The proposed score-based SDE solver operates in two stages: first, employing score matching or Score-PINN to acquire the score function; and second, solving the LL via an ordinary differential equation (ODE) using the obtained score function. Comparative evaluations across these methods showcase varying trade-offs. The proposed methodology is evaluated across diverse SDEs, including anisotropic Ornstein-Uhlenbeck processes, geometric Brownian motion, and Brownian motion with varying eigenspace. We also test various distributions, including Gaussian, Log-normal, Laplace, and Cauchy distributions. The numerical results demonstrate the score-based SDE solver’s stability, speed, and performance across different experimental settings, solidifying its potential as a solution to CoD for high-dimensional FP equations.

97 MATHEMATICS AND COMPUTING↗