Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “data intensive computing”

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 1,765 records · Page 98
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Accelerating Discovery of Atomistic Defects via Machine Learning

The quantification of defects such as vacancies in crystalline structures is a cornerstone of materials science research. Traditional efforts often rely on manual detection, a process that is time-intensive, prone to human error, and challenging to scale. Here we leverage machine learning (ML) methods to identify and quantify vacancies within a crystalline lattice, aiming to expedite detection while improving accuracy. Additionally, we explore the transferability of these ML techniques, identifying characteristics of atomistic imaging data that complicate this task. We show how the integration of ML can drive innovation, providing a powerful tool that will play an increasingly crucial role in the future of materials science.

2D materials↗
DOE OSTI2014

First-principles simulation of the optical response of bulk and thin-film α-quartz irradiated with an ultrashort intense laser pulse

A computational method based on a first-principles multiscale simulation has been used for calculating the optical response and the ablation threshold of an optical material irradiated with an ultrashort intense laser pulse. The method employs Maxwell's equations to describe laser pulse propagation and time-dependent density functional theory to describe the generation of conduction band electrons in an optical medium. Optical properties, such as reflectance and absorption, were investigated for laser intensities in the range 10{sup 10} W/cm{sup 2} to 2 × 10{sup 15} W/cm{sup 2} based on the theory of generation and spatial distribution of the conduction band electrons. The method was applied to investigate the changes in the optical reflectance of α-quartz bulk, half-wavelength thin-film, and quarter-wavelength thin-film and to estimate their ablation thresholds. Despite the adiabatic local density approximation used in calculating the exchange–correlation potential, the reflectance and the ablation threshold obtained from our method agree well with the previous theoretical and experimental results. The method can be applied to estimate the ablation thresholds for optical materials, in general. The ablation threshold data can be used to design ultra-broadband high-damage-threshold coating structures.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2020

Reducing Communication in Graph Neural Network Training

Graph Neural Networks (GNNs) are powerful and flexible neural networks that use the naturally sparse connectivity information of the data. GNNs represent this connectivity as sparse matrices, which have lower arithmetic intensity and thus higher communication costs compared to dense matrices, making GNNs harder to scale to high concurrencies than convolutional or fully-connected neural networks. Here, we introduce a family of parallel algorithms for training GNNs and show that they can asymptotically reduce communication compared to previous parallel GNN training methods. We implement these algorithms, which are based on 1D, 1. 5D, 2D, and 3D sparse-dense matrix multiplication, using torch.distributed on GPU-equipped clusters. Our algorithms optimize communication across the full GNN training pipeline. We train GNNs on over a hundred GPUs on multiple datasets, including a protein network with over a billion edges.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2014

Genomic Instability and Radiation Risk in Molecular Pathways to Colon Cancer

Colon cancer is caused by multiple genomic alterations which lead to genomic instability (GI). GI appears in molecular pathways of microsatellite instability (MSI) and chromosomal instability (CIN) with clinically observed case shares of about 15–20% and 80–85%. Radiation enhances the colon cancer risk by inducing GI, but little is known about different outcomes for MSI and CIN. Computer-based modelling can facilitate the understanding of the phenomena named above. Comprehensive biological models, which combine the two main molecular pathways to colon cancer, are fitted to incidence data of Japanese a-bomb survivors. The preferred model is selected according to statistical criteria and biological plausibility. Imprints of cell-based processes in the succession from adenoma to carcinoma are identified by the model from age dependences and secular trends of the incidence data. Model parameters show remarkable compliance with mutation rates and growth rates for adenoma, which has been reported over the last fifteen years. Model results suggest that CIN begins during fission of intestinal crypts. Chromosomal aberrations are generated at a markedly elevated rate which favors the accelerated growth of premalignant adenoma. Possibly driven by a trend of Westernization in the Japanese diet, incidence rates for the CIN pathway increased notably in subsequent birth cohorts, whereas rates pertaining to MSI remained constant. An imbalance between number of CIN and MSI cases began to emerge in the 1980s, whereas in previous decades the number of cases was almost equal. The CIN pathway exhibits a strong radio-sensitivity, probably more intensive in men. Among young birth cohorts of both sexes the excess absolute radiation risk related to CIN is larger by an order of magnitude compared to the MSI-related risk. Observance of pathway-specific risks improves the determination of the probability of causation for radiation-induced colon cancer in individual patients, if their exposure histories are known.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2014

MO-G-17A-03: MR-Based Cortical Bone Segmentation for PET Attenuation Correction with a Non-UTE 3D Fast GRE Sequence

Purpose: To determine the feasibility of identifying cortical bone on MR images with a short-TE 3D fast-GRE sequence for attenuation correction of PET data in PET/MR. Methods: A water-fat-bone phantom was constructed with two pieces of beef shank. MR scans were performed on a 3T MR scanner (GE Discovery™ MR750). A 3D GRE sequence was first employed to measure the level of residual signal in cortical bone (TE{sub 1}/TE{sub 2}/TE{sub 3}=2.2/4.4/6.6ms, TR=20ms, flip angle=25°). For cortical bone segmentation, a 3D fast-GRE sequence (TE/TR=0.7/1.9ms, acquisition voxel size=2.5×2.5×3mm{sup 3}) was implemented along with a 3D Dixon sequence (TE{sub 1}/TE{sub 2}/TR=1.2/2.3/4.0ms, acquisition voxel size=1.25×1.25×3mm{sup 3}) for water/fat imaging. Flip angle (10°), acquisition bandwidth (250kHz), FOV (480×480×144mm{sup 3}) and reconstructed voxel size (0.94×0.94×1.5mm{sup 3}) were kept the same for both sequences. Soft tissue and fat tissue were first segmented on the reconstructed water/fat image. A tissue mask was created by combining the segmented water/fat masks, which was then applied on the fast-GRE image (MRFGRE). A second mask was created to remove the Gibbs artifacts present in regions in close vicinity to the phantom. MRFGRE data was smoothed with a 3D anisotropic diffusion filter for noise reduction, after which cortical bone and air was separated using a threshold determined from the histogram. Results: There is signal in the cortical bone region in the 3D GRE images, indicating the possibility of separating cortical bone and air based on signal intensity from short-TE MR image. The acquisition time for the 3D fast-GRE sequence was 17s, which can be reduced to less than 10s with parallel imaging. The attenuation image created from water-fat-bone segmentation is visually similar compared to reference CT. Conclusion: Cortical bone and air can be separated based on intensity in MR image with a short-TE 3D fast-GRE sequence. Further research is required to optimize the strategy to reduce Gibbs artifacts.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2013

Hydrothermal reactions: From the synthesis of ligand to new lanthanide 3D-coordination polymers

The organic ligand 2,5-piperazinedione-1,4-diacetic acid (H{sub 2}PDA) was synthesized under hydrothermal conditions starting from the iminodiacetic acid and catalyzed by oxalic acid. The X-ray powder diffraction data indicates that the compound crystallizes in the P2{sub 1}/c monoclinic system as reported in the literature. The ligand was also characterized by elemental analysis, magnetic nuclear resonance, infrared spectroscopy and thermogravimetric analysis. Two new coordination networks based on lanthanide ions were obtained with this ligand using hydrothermal reaction. In addition to single-crystal X-ray diffraction, the compounds were characterized by infrared spectroscopy, thermogravimetric analysis, scanning electron microscopy and elemental analysis. Single-crystal XRD showed that the compounds are isostructural, crystallizing in P2{sub 1}/n monoclinic system with chemical formula [Ln(PDA){sub 1.5}(H{sub 2}O)](H{sub 2}O){sub 3} (Ln=Gd{sup 3+}(1) and Eu{sup 3+}(2)).The luminescence properties of both compounds were studied. In the compound (1), a broad emission band was observed at 479 nm, redshifted by 70 nm in comparison of the free ligand. In (2), the typical f–f transition was observed with a maximum peak at 618 nm, related with the red emission of the europium ions. Computational methods were performed to simulate the crystal structure of (2). The theoretical calculations of the intensity parameters are in good agreement with the experimental values. - Graphical abstract: Scheme of obtaining the ligand 2,5-piperazinedione-1,4-diacetic acid (H{sub 2}PDA) and two new isostructural 3D-coordination polymers [Ln(PDA){sub 1.5}(H{sub 2}O)](H{sub 2}O){sub 3} (Ln=Gd{sup 3+} and Eu{sup 3+}) by hydrothermal synthesis. Display Omitted - Highlights: • The ligand 2,5-piperazinedione-1,4-diacetic acid was synthetized using the hydrothermic method and characterized. • Two new 3D-coordination polymers with this ligand containing Gd{sup 3+} and Eu{sup 3+} ions were also obtained. • The compounds are isostructural and the typical luminescent properties were observed in both structures.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2016

Clinical Experience and Evaluation of Patient Treatment Verification With a Transit Dosimeter

Purpose: To prospectively evaluate a protocol for transit dosimetry on a patient population undergoing intensity modulated radiation therapy (IMRT) and to assess the issues in clinical implementation of electronic portal imaging devices (EPIDs) for treatment verification. Methods and Materials: Fifty-eight patients were enrolled in the study. Amorphous silicon EPIDs were calibrated for dose and used to acquire images of delivered fields. Measured EPID dose maps were back-projected using the planning computed tomographic (CT) images to calculate dose at prespecified points within the patient and compared with treatment planning system dose offline using point dose difference and point γ analysis. The deviation of the results was used to inform future action levels. Results: Two hundred twenty-five transit images were analyzed, composed of breast, prostate, and head and neck IMRT fields. Patient measurements demonstrated the potential of the dose verification protocol to model dose well under complex conditions: 83.8% of all delivered beams achieved the initial set tolerance level of Δ{sub D} of 0 ± 5 cGy or %Δ{sub D} of 0% ± 5%. Importantly, the protocol was also sensitive to anatomic changes and spotted that 3 patients from 20 measured prostate patients had undergone anatomic change in comparison with the planning CT. Patient data suggested an EPID-reconstructed versus treatment planning system dose difference action level of 0% ± 7% for breast fields. Asymmetric action levels were more appropriate for inversed IMRT fields, using absolute dose difference (−2 ± 5 cGy) or summed field percentage dose difference (−6% ± 7%). Conclusions: The in vivo dose verification method was easy to use and simple to implement, and it could detect patient anatomic changes that impacted dose delivery. The system required no extra dose to the patient or treatment time delay and so could be used throughout the course of treatment to identify and limit systematic and random errors in dose delivery for patient groups.

62 RADIOLOGY AND NUCLEAR MEDICINE↗
NASA NTRS2008

Hybrid LES/RANS Simulation of Transverse Sonic Injection into a Mach 2 Flow

A computational study of transverse sonic injection of air and helium into a Mach 1.98 cross-flow is presented. A hybrid large-eddy simulation / Reynolds-averaged Navier-Stokes (LES/RANS) turbulence model is used, with the two-equation Menter baseline (Menter-BSL) closure for the RANS part of the flow and a Smagorinsky-type model for the LES part of the flow. A time-dependent blending function, dependent on modeled turbulence variables, is used to shift the closure from RANS to LES. Turbulent structures are initiated and sustained through the use of a recycling / rescaling technique. Two higher-order discretizations, the Piecewise Parabolic Method (PPM) of Colella and Woodward, and the SONIC-A ENO scheme of Suresh and Huyhn are used in the study. The results using the hybrid model show reasonably good agreement with time-averaged Mie scattering data and with experimental surface pressure distributions, even though the penetration of the jet into the cross-flow is slightly over-predicted. The LES/RANS results are used to examine the validity of commonly-used assumptions of constant Schmidt and Prandtl numbers in the intense mixing zone downstream of the injection location.

Boles, John A.↗
DOE OSTI2016

Ghost fringe removal techniques using Lissajous data presentation

A VISAR (Velocity Interferometer System for Any Reflector) is a Dopplervelocity interferometer which is an important optical diagnostic in shockwave experiments at the national laboratories, used to measureequation of state(EOS) of materials under extreme conditions. Unwanted reflection of laser light from target windows can produce an additional component to the VISAR fringe record that can distort and obscure the true velocity signal. When accurately removing this so-called ghost artifact component is essential for achieving high accuracy EOSmeasurements, especially when the true light signal is only weakly reflected from the shock front. Independent of the choice of algorithm for processing the raw data into a complex fringe signal, we have found it beneficial to plot this signal as a Lissajous and seek the proper center of this path, even under time varying intensity which can shift the perceived center. Moreover, the ghost contribution is then solved by a simple translation in the complex plane that recenters the Lissajous path. For continuous velocity histories, we find that plotting the fringe magnitude vs nonfringing intensity and optimizing linearity is an invaluable tool for determining accurate ghost offsets. For discontinuous velocity histories, we have developed graphically inspired methods which relate the results of two VISARs having different velocity per fringe proportionalities or assumptions of constant fringe magnitude to find the ghost offset. The technique can also remove window reflection artifacts in generic interferometers, such as in the metrology of surfaces.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2014

Routing-as-a-Service (RaaS): A Framework for Tenant-Directed Route Control in Data Center

In a multi-tenant data center environment, the current paradigm for route control customization involves a labor-intensive ticketing process where tenants submit route control requests to the landlord. This results in tight coupling between tenants and the landlord, extensive human resource deployment, and long ticket resolution time. We propose Routing-as-a-Service (RaaS), a framework for tenant-directed route control in data centers. Here, we show that RaaS-based implementation provides a route control platform where multiple tenants can perform route control independently with little administrative involvement, and the landlord can set the overall network policies. RaaS-based solutions can run on commercial off-the-shelf (COTS) hardware and leverage existing technologies, so it can be implemented in existing networks without major infrastructural overhaul. We present the design of RaaS, introduce its components, and evaluate a prototype based on RaaS.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Dynamics retrieval from stochastically weighted incomplete data by low-pass spectral analysis

Time-resolved serial femtosecond crystallography (TR-SFX) provides access to protein dynamics on sub-picosecond timescales, and with atomic resolution. Due to the nature of the experiment, these datasets are often highly incomplete and the measured diffracted intensities are affected by partiality. To tackle these issues, one established procedure is that of splitting the data into time bins, and averaging the multiple measurements of equivalent reflections within each bin. This binning and averaging often involve a loss of information. Here, we propose an alternative approach, which we call low-pass spectral analysis (LPSA). In this method, the data are projected onto the subspace defined by a set of trigonometric functions, with frequencies up to a certain cutoff. This approach attenuates undesirable high-frequency features and facilitates retrieving the underlying dynamics. A time-lagged embedding step can be included prior to subspace projection to improve the stability of the results with respect to the parameters involved. Subsequent modal decomposition allows to produce a low-rank description of the system's evolution. Using a synthetic time-evolving model with incomplete and partial observations, we analyze the LPSA results in terms of quality of the retrieved signal, as a function of the parameters involved. We compare the performance of LPSA to that of a range of other sophisticated data analysis techniques. We show that LPSA allows to achieve excellent dynamics reconstruction at modest computational cost. Finally, we demonstrate the superiority of dynamics retrieval by LPSA compared to time binning and merging, which is, to date, the most commonly used method to extract dynamical information from TR-SFX data.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2021

A NICER View of the Massive Pulsar PSR J0740+6620 Informed by Radio Timing and XMM-Newton Spectroscopy

We report on Bayesian estimation of the radius, mass, and hot surface regions of the massive millisecond pulsar PSR J0740+6620, conditional on pulse-profile modeling of Neutron Star Interior Composition Explorer X-ray Timing Instrument event data. We condition on informative pulsar mass, distance, and orbital inclination priors derived from the joint North American Nanohertz Observatory for Gravitational Waves and Canadian Hydrogen Intensity Mapping Experiment/Pulsar wideband radio timing measurements of Fonseca et al. We use XMM-Newton European Photon Imaging Camera spectroscopic event data to inform our X-ray likelihood function. The prior support of the pulsar radius is truncated at 16 km to ensure coverage of current dense matter models. We assume conservative priors on instrument calibration uncertainty. We constrain the equatorial radius and mass of PSR J0740+6620 to be ${12.39}_{-0.98}^{+1.30}$ km and ${2.072}_{-0.066}^{+0.067}$ M ⊙ respectively, each reported as the posterior credible interval bounded by the 16% and 84% quantiles, conditional on surface hot regions that are non-overlapping spherical caps of fully ionized hydrogen atmosphere with uniform effective temperature; a posteriori, the temperature is ${\mathrm{log}}_{10}(T\,[{\rm{K}}])={5.99}_{-0.06}^{+0.05}$ for each hot region. All software for the X-ray modeling framework is open-source and all data, model, and sample information is publicly available, including analysis notebooks and model modules in the Python language. Furthermore, our marginal likelihood function of mass and equatorial radius is proportional to the marginal joint posterior density of those parameters (within the prior support) and can thus be computed from the posterior samples.

79 ASTRONOMY AND ASTROPHYSICS↗
NASA NTRS1986

Further studies with data collected by NASA's airborne Doppler lidar in Oklahoma in 1981

Continued study of the lidar data collected in 1981 has resulted in significant new improvements in the analysis techniques reported by Bluestein et al. (1985) and McCaul (1985). Through comparison of fore- and aft-derived scalar fields of intensity and spectral width, the self-consistency of the lidar moment estimates was assessed. Reflectivity estimates were found to be quite stable and reliable, while spectral widths were prone to become noisy if signal to noise ratio (SNR) fell below 12 dB. In addition, spectral widths contained a significant component due to radial velocity gradients in areas along gust fronts, and these components were different along the fore and aft lines of sight. Significant improvement in agreement between the fore and aft fields of spectral width was obtained by estimating the radial velocity gradient component and then removing it from the raw measured widths to yield only the turbulent portion of the contribution to width. Additional analyses showed that lidar-derived vorticity estimates were consistent with several approximate models of vorticity growth along gust front zones, and with the hypothesis that Helmholtz instability could have been responsible for vortices seen along part of the gust front of 30 June 1981. Computations of divergence transverse to axes through an isolated cumulus congestus indicated that the strongest convergence tended to lie along an axis parallel to the congestus. This and the results of other additional analyses seem to suggest that the lidar winds do indeed accurately reflected the basic features of the real wind field.

Bluestein, H. B.↗
DOE OSTI2022

In-process monitoring and prediction of droplet quality in droplet-on-demand liquid metal jetting additive manufacturing using machine learning

Abstract In droplet-on-demand liquid metal jetting (DoD-LMJ) additive manufacturing, complex physical interactions govern the droplet characteristics, such as size, velocity, and shape. These droplet characteristics, in turn, determine the functional quality of the printed parts. Hence, to ensure repeatable and reliable part quality it is necessary to monitor and control the droplet characteristics. Existing approaches for in-situ monitoring of droplet behavior in DoD-LMJ rely on high-speed imaging sensors. The resulting high volume of droplet images acquired is computationally demanding to analyze and hinders real-time control of the process. To overcome this challenge, the objective of this work is to use time series data acquired from an in-process millimeter-wave sensor for predicting the size, velocity, and shape characteristics of droplets in DoD-LMJ process. As opposed to high-speed imaging, this sensor produces data-efficient time series signatures that allows rapid, real-time process monitoring. We devise machine learning models that use the millimeter-wave sensor data to predict the droplet characteristics. Specifically, we developed multilayer perceptron-based non-linear autoregressive models to predict the size and velocity of droplets. Likewise, a supervised machine learning model was trained to classify the droplet shape using the frequency spectrum information contained in the millimeter-wave sensor signatures. High-speed imaging data served as ground truth for model training and validation. These models captured the droplet characteristics with a statistical fidelity exceeding 90%, and vastly outperformed conventional statistical modeling approaches. Thus, this work achieves a practically viable sensing approach for real-time quality monitoring of the DoD-LMJ process, in lieu of the existing data-intensive image-based techniques.

Gaikwad, Aniruddha (ORCID:0000000285642621)↗
DOE OSTI2014

A region-appearance-based adaptive variational model for 3D liver segmentation

Purpose: Liver segmentation from computed tomography images is a challenging task owing to pixel intensity overlapping, ambiguous edges, and complex backgrounds. The authors address this problem with a novel active surface scheme, which minimizes an energy functional combining both edge- and region-based information. Methods: In this semiautomatic method, the evolving surface is principally attracted to strong edges but is facilitated by the region-based information where edge information is missing. As avoiding oversegmentation is the primary challenge, the authors take into account multiple features and appearance context information. Discriminative cues, such as multilayer consecutiveness and local organ deformation are also implicitly incorporated. Case-specific intensity and appearance constraints are included to cope with the typically large appearance variations over multiple images. Spatially adaptive balancing weights are employed to handle the nonuniformity of image features. Results: Comparisons and validations on difficult cases showed that the authors’ model can effectively discriminate the liver from adhering background tissues. Boundaries weak in gradient or with no local evidence (e.g., small edge gaps or parts with similar intensity to the background) were delineated without additional user constraint. With an average surface distance of 0.9 mm and an average volume overlap of 93.9% on the MICCAI data set, the authors’ model outperformed most state-of-the-art methods. Validations on eight volumes with different initial conditions had segmentation score variances mostly less than unity. Conclusions: The proposed model can efficiently delineate ambiguous liver edges from complex tissue backgrounds with reproducibility. Quantitative validations and comparative results demonstrate the accuracy and efficacy of the model.

62 RADIOLOGY AND NUCLEAR MEDICINE↗
DOE OSTI2024

Patch2Self2: Self-supervised Denoising on Coresets via Matrix Sketching

Diffusion MRI (dMRI) non-invasively maps brain white matter yet necessitates denoising due to low signal-to-noise ratios. Patch2Self (P2S) employing self-supervised techniques and regression on a Casorati matrix effectively denoises dMRI images and has become the new de-facto standard in this field. P2S however is resource intensive both in terms of running time and memory usage as it uses all voxels (n) from all-but-one held-in volumes (d-1) to learn a linear mapping Phi : \mathbb R ^ n x(d-1) \mapsto \mathbb R ^ n for denoising the held-out volume. The increasing size and dimensionality of higher resolution dMRI acquisitions can make P2S infeasible for large-scale analyses. This work exploits the redundancy imposed by P2S to alleviate its performance issues and inspect regions that influence the noise disproportionately. Specifically this study makes a three-fold contribution: (1) We present Patch2Self2 (P2S2) a method that uses matrix sketching to perform self-supervised denoising. By solving a sub-problem on a smaller sub-space so called coreset we show how P2S2 can yield a significant speedup in training time while using less memory. (2) We present a theoretical analysis of P2S2 focusing on determining the optimal sketch size through rank estimation a key step in achieving a balance between denoising accuracy and computational efficiency. (3) We show how the so-called statistical leverage scores can be used to interpret the denoising of dMRI data a process that was traditionally treated as a black-box. Experimental results on both simulated and real data affirm that P2S2 maintains denoising quality while significantly enhancing speed and memory efficiency achieved by training on a reduced data subset.

Fadnavis, Shreyas↗
DOE OSTI2025

Leveraging Large Language Models for Real-World Data Evidence: A Framework for Automated Treatment Extraction and Data Harmonization

Background: The ability to comprehensively collect treatment information from cancer patient medical records would enable studies to evaluate real-world benefits and risks tied to specific treatments. Currently, it is difficult to system- atically collect high-quality treatment information because it is often stored in unstructured text. Manually extracting and standardizing drug and regimen data is time-intensive. Recent advances in large language models (LLMs) offer a potential solution for automated extraction of structured treatment information from clinical text. Objective: This study systematically evaluates the utility of four LLMs from the Llama family for automated extraction of oncology treatment information from clinical text. This information can guide researchers using cancer registry data to provide insights into cancer care and outcomes beyond clinical trials. Methods: Four instruction-tuned Llama models with varying parameter counts (1B, 3B, 8B, and 70B) were evaluated for their ability to extract treatment information from clinical documents. A unified oncology knowledge base integrating seven major public data sources was developed to standardize and normalize extracted entities—a critical step for harmonizing data from diverse sources. Extracted treatment data were compared against expert-annotated ground truth. Model performance was assessed using accuracy metrics (Precision, Recall, F1-Score) and opera- tional feasibility metrics, including processing speed and structural compliance of the output. Results: A strong positive correlation was observed between model size and extraction accuracy. F1-score improved from 0.609 for the 1B model to 0.710 (3B), 0.807 (8B), and 0.828 (70B). While larger models demonstrated superior accuracy and compliance, they incurred higher computational costs. The modest performance difference between 8B and 70B suggests diminishing returns with increasing model size. Conclusions: LLMs represent a viable technology for automating oncology treatment extraction. The 8B-parameter model emerged as a highly effective option, balancing high accuracy and computational efficiency. Selecting an appropriate LLM for deployment in cancer registries involves a trade-off between desired accuracy and available operational resources. Harmonizing extracted entities with the oncology knowledge base facilitates standardized integration into common data models, enhancing data quality for real-world evidence analyses.

artificial intelligence↗
DOE OSTI2014

SU-E-J-103: Setup Errors Analysis by Cone-Beam CT (CBCT)-Based Imaged-Guided Intensity Modulated Radiotherapy for Esophageal Cancer

Purpose: To quantify setup errors by pretreatment kilovolt cone-beam computed tomography(KV-CBCT) scans for middle or distal esophageal carcinoma patients. Methods: Fifty-two consecutive middle or distal esophageal carcinoma patients who underwent IMRT were included this study. A planning CT scan using a big-bore CT simulator was performed in the treatment position and was used as the reference scan for image registration with CBCT. CBCT scans(On-Board Imaging v1. 5 system, Varian Medical Systems) were acquired daily during the first treatment week. A total of 260 CBCT scans was assessed with a registration clip box defined around the PTV-thorax in the reference scan based on(nine CBCTs per patient) bony anatomy using Offline Review software v10.0(Varian Medical Systems). The anterior-posterior(AP), left-right(LR), superiorinferior( SI) corrections were recorded. The systematic and random errors were calculated. The CTV-to-PTV margins in each CBCT frequency was based on the Van Herk formula (2.5Σ+0.7σ). Results: The SD of systematic error (Σ) was 2.0mm, 2.3mm, 3.8mm in the AP, LR and SI directions, respectively. The average random error (σ) was 1.6mm, 2.4mm, 4.1mm in the AP, LR and SI directions, respectively. The CTV-to-PTV safety margin was 6.1mm, 7.5mm, 12.3mm in the AP, LR and SI directions based on van Herk formula. Conclusion: Our data recommend the use of 6 mm, 8mm, and 12 mm for esophageal carcinoma patient setup in AP, LR, SI directions, respectively.

62 RADIOLOGY AND NUCLEAR MEDICINE↗