Search

The complexity and heterogeneity of data in many real-world applications pose significant challenges for traditional machine learning and signal processing techniques. For instance, in medicine, effective analysis of diverse physiological signals is crucial for patient monitoring and clinical decision-making and yet highly challenging. We introduce MedTsLLM, a general multimodal large language model (LLM) framework that effectively integrates time series data and rich contextual information in the form of text to analyze physiological signals, performing three tasks with clinical relevance: semantic segmentation, boundary detection, and anomaly detection in time series. These critical tasks enable deeper analysis of physiological signals and can provide actionable insights for clinicians. We utilize a reprogramming layer to align embeddings of time series patches with a pretrained LLM's embedding space and make effective use of raw time series, in conjunction with textual context. Given the multivariate nature of medical datasets, we develop methods to handle multiple covariates. We additionally tailor the text prompt to include patient-specific information. Our model outperforms state-of-the-art baselines, including deep learning models, other LLMs, and clinical methods across multiple medical domains, specifically electrocardiograms and respiratory waveforms. MedTsLLM presents a promising step towards harnessing the power of LLMs for medical time series analysis that can elevate data-driven tools for clinicians and improve patient outcomes., Comment: published in Proceedings of Machine Learning Research, MLHC 2024

Radiative heat transfer is a fundamental process in high energy density physics and inertial fusion. Accurately predicting the behavior of Marshak waves across a wide range of material properties and drive conditions is crucial for design and analysis of these systems. Conventional numerical solvers and analytical approximations often face challenges in terms of accuracy and computational efficiency. In this work, we propose a novel approach to model Marshak waves using Fourier Neural Operators (FNO). We develop two FNO-based models: (1) a base model that learns the mapping between the drive condition and material properties to a solution approximation based on the widely used analytic model by Hammer & Rosen (2003), and (2) a model that corrects the inaccuracies of the analytic approximation by learning the mapping to a more accurate numerical solution. Our results demonstrate the strong generalization capabilities of the FNOs and show significant improvements in prediction accuracy compared to the base analytic model.

Verification solutions for uncertainty quantification are presented for time dependent transport problems where $c$, the scattering ratio, is uncertain. The method of polynomial chaos expansions is employed for quick and accurate calculation of the quantities of interest and uncollided solutions are used to treat part of the uncertainty calculation analytically. We find that approximately six moments in the polynomial expansion are required to represent the solutions to these problems accurately. Additionally, the results show that if the uncertainty interval spans c=1, which means it is uncertain whether the system is multiplying or not, the confidence interval will grow in time. Finally, since the QoI is a strictly increasing function, the percentile values are known and can be used to verify the accuracy of the expansion. These results can be used to test UQ methods for time-dependent transport problems.

The set of benchmark solutions used in the thermal radiative transfer community suffer some coverage gaps, in particular nonlinear, non-equilibrium problems. Also, there are no non-equilibrium, optically thick benchmarks. These shortcomings motivated the development of a numerical method free from the requirement of linearity and easily able to converge on smooth optically thick problems, a moving mesh Discontinuous Galerkin (DG) framework that utilizes an uncollided source treatment. Having already proven this method on time dependent scattering transport problems, we present here solutions to non-equilibrium thermal radiative transfer problems for familiar linearized systems together with more physical nonlinear systems in both optically thin and thick regimes, including both the full transport and the $S_2$/$P_1$ solution. Geometric convergence is observed for smooth sources at all times and some nonsmooth sources at late times when there is local equilibrium. Also, accurate solutions are achieved for step sources when the solution is not smooth.

This work is concerned with the development of a numerical modelling approach for studying the time-accurate response of aerospace fasteners subjected to high electrical current loading from a simulated lightning strike. The electromagnetic, thermal and elastoplastic response of individual fastener components is captured by this method allowing a critical analysis of fastener design and material layering. Under high electrical current loading, ionisation of gas filled cavities in the fastener assembly can lead to viable current paths across internal voids. This ionisation can lead to localised pockets of high pressure plasma through the Joule heating effect. The multi-physics approach developed in this paper extends an existing methodology that allows a two-way dynamic non-linear coupling of the plasma arc, the titanium aerospace fastener components, the surrounding aircraft panels, the internal supporting structure and internal plasma-filled cavities. Results from this model are compared with experimental measurements of a titanium fastener holding together carbon composite panels separated by thin dielectric layers. The current distribution measurements are shown to be accurately reproduced. A parameter study is used to assess the internal cavity modelling strategy and to quantify the relation between the internal cavity plasma pressure, the electrical current distribution and changes in the internal cavity geometry., Comment: 16 pages, 17 figures

For the purpose of finding benchmark quality solutions to time dependent Sn transport problems, we develop a numerical method in a Discontinuous Galerkin (DG) framework that utilizes time dependent cell edges, which we call a moving mesh, and an uncollided source treatment. The DG method for discretizing space is a powerful solution technique on smooth problems and is robust on non-smooth problems. In order to realize the potential of the DG method to spectrally resolve smooth problems, our moving mesh and uncollided source treatment is devised to circumvent discontinuities in the solution or the first derivative of the solutions that are admitted in transport calculations. The resulting method achieves spectral convergence on smooth problems, like a standard DG implementation. When applied to problems with nonsmooth sources that induce discontinuities, our moving mesh, uncollided source method returns a significantly more accurate solution than the standard DG method. On problems with smooth sources, we observe spectral convergence even in problems with wave fronts. In problems where the angular flux is inherently non-smooth, as in Ganapol's (2001) well known plane pulse benchmark, we do not observe an elevated order of accuracy when compared with static meshes, but there is a reduction in error that is nearly three orders of magnitude.

The widely used AZURV1 transport benchmarks package provides a suite of solutions to isotropic scattering transport problems with a variety of initial conditions (Ganapol 2001). Most of these solutions have an initial condition that is a Dirac delta function in space; as a result these benchmarks are challenging problems to use for verification tests in computer codes. Nevertheless, approximating a delta function in simulation often leads to low orders of convergence and the inability to test the convergence of high-order numerical methods. While there are examples in the literature of integration of these solutions as Green's functions for the transport operator to produce results for more easily simulated sources, they are limited in scope and briefly explained. For a sampling of initial conditions and sources, we present solutions for the uncollided and collided scalar flux to facilitate accurate testing of source treatment in numerical solvers. The solution for the uncollided scalar flux is found in analytic form for some sources. Since integrating the Green's functions is often nontrivial, discussion of integration difficulty and workarounds to find convergent integrals is included. Additionally, our uncollided solutions can be used as source terms in verification studies, in a similar way to the method of manufactured solutions.

Predicting accurate protein-ligand binding affinity is important in drug discovery but remains a challenge even with computationally expensive biophysics-based energy scoring methods and state-of-the-art deep learning approaches. Despite the recent advances in the deep convolutional and graph neural network based approaches, the model performance depends on the input data representation and suffers from distinct limitations. It is natural to combine complementary features and their inference from the individual models for better predictions. We present fusion models to benefit from different feature representations of two neural network models to improve the binding affinity prediction. We demonstrate effectiveness of the proposed approach by performing experiments with the PDBBind 2016 dataset and its docking pose complexes. The results show that the proposed approach improves the overall prediction compared to the individual neural network models with greater computational efficiency than related biophysics based energy scoring functions. We also discuss the benefit of the proposed fusion inference with several example complexes. The software is made available as open source at https://github.com/llnl/fast.

Topical intra-nasal sprays are amongst the most commonly prescribed therapeutic options for sinonasal diseases in humans. However, inconsistency and ambiguity in instructions show a lack of definitive knowledge on best spray use techniques. In this study, we have identified a new usage strategy for nasal sprays available over-the-counter, that registers an average 8-fold improvement in topical delivery of drugs at diseased sites, when compared to prevalent spray techniques. The protocol involves re-orienting the spray axis to harness inertial motion of particulates and has been developed using computational fluid dynamics simulations of respiratory airflow and droplet transport in medical imaging-based digital models. Simulated dose in representative models is validated through in vitro spray measurements in 3D-printed anatomic replicas using the gamma scintigraphy technique. This work breaks new ground in proposing an alternative user-friendly strategy that can significantly enhance topical delivery inside human nose. While these findings can eventually translate into personalized spray usage instructions and hence merit a change in nasal standard-of-care, this study also demonstrates how relatively simple engineering analysis tools can revolutionize everyday healthcare., Comment: 24 pages, 8 figures, 5 tables

Introduction: Topical intranasal drugs are widely prescribed for Chronic Rhinosinusitis (CRS), although delivery can vary with device type and droplet size. The study objective was to compare nebulized and sprayed droplet deposition in the paranasal sinuses and ostiomeatal complex (OMC) across multiple droplet sizes in CRS patients using computational fluid dynamics (CFD). Methods: Three-dimensional models of sinonasal cavities were constructed from computed tomography (CT) scans of three subjects with CRS refractory to medical therapy using imaging software. Assuming steady-state inspiratory airflow at resting rate, CFD was used to simulate 1-120 {\mu}m sprayed droplet deposition in the left and right sinuses and OMC with spray nozzle positioning as in current nasal spray use instructions. Zero-velocity nebulization simulations were performed for 1-30 {\mu}m droplet sizes, maximal sinus and OMC deposition fractions (MSDF) were obtained, and sizes that achieved at least 50% of MSDF were identified. Nebulized MSDF was compared to sprayed droplet deposition. We also validated CFD framework through in vitro experiments. Results: Among nebulized droplet sizes, 11-14 {\mu}m droplets achieved at least 50% of MSDF in all six sinonasal cavities. Five of six sinonasal cavities had greater sinus and OMC deposition with nebulized droplets than with sprayed droplets at optimal sizes. Conclusions: Nebulized droplets may target the sinuses and OMC more effectively than sprayed particles at sizes achieving best deposition. Further studies are needed to confirm our preliminary findings. Several commercial nasal nebulizers have average particle sizes outside the optimal nebulized droplet size range found here, suggesting potential for product enhancement., Comment: 32 pages, 9 figures