Search NASA⌕ Search

SEARCH · Search NASA

Results for “timing error evaluation”

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 685 records · Page 38

NASA Tech Briefs, December 2011

Topics covered include: 1) SNE Industrial Fieldbus Interface; 2) Composite Thermal Switch; 3) XMOS XC-2 Development Board for Mechanical Control and Data Collection; 4) Receiver Gain Modulation Circuit; 5) NEXUS Scalable and Distributed Next-Generation Avionics Bus for Space Missions; 6) Digital Interface Board to Control Phase and Amplitude of Four Channels; 7) CoNNeCT Baseband Processor Module; 8) Cryogenic 160-GHz MMIC Heterodyne Receiver Module; 9) Ka-Band, Multi-Gigabit-Per-Second Transceiver; 10) All-Solid-State 2.45-to-2.78-THz Source; 11) Onboard Interferometric SAR Processor for the Ka-Band Radar Interferometer (KaRIn); 12) Space Environments Testbed; 13) High-Performance 3D Articulated Robot Display; 14) Athena; 15) In Situ Surface Characterization; 16) Ndarts; 17) Cryo-Etched Black Silicon for Use as Optical Black; 18) Advanced CO2 Removal and Reduction System; 19) Correcting Thermal Deformations in an Active Composite Reflector; 20) Umbilical Deployment Device; 21) Space Mirror Alignment System; 22) Thermionic Power Cell To Harness Heat Energies for Geothermal Applications; 23) Graph Theory Roots of Spatial Operators for Kinematics and Dynamics; 24) Spacesuit Soft Upper Torso Sizing Systems; 25) Radiation Protection Using Single-Wall Carbon Nanotube Derivatives; 26) PMA-PhyloChip DNA Microarray to Elucidate Viable Microbial Community Structure; 27) Lidar Luminance Quantizer; 28) Distributed Capacitive Sensor for Sample Mass Measurement; 29) Base Flow Model Validation; 30) Minimum Landing Error Powered-Descent Guidance for Planetary Missions; 31) Framework for Integrating Science Data Processing Algorithms Into Process Control Systems; 32) Time Synchronization and Distribution Mechanisms for Space Networks; 33) Local Estimators for Spacecraft Formation Flying; 34) Software-Defined Radio for Space-to-Space Communications; 35) Reflective Occultation Mask for Evaluation of Occulter Designs for Planet Finding; and 36) Molecular Adsorber Coating

Source record↗

Evidence Report: Risk of Inadequate Human-Computer Interaction

Human-computer interaction (HCI) encompasses all the methods by which humans and computer-based systems communicate, share information, and accomplish tasks. When HCI is poorly designed, crews have difficulty entering, navigating, accessing, and understanding information. HCI has rarely been studied in an operational spaceflight context, and detailed performance data that would support evaluation of HCI have not been collected; thus, we draw much of our evidence from post-spaceflight crew comments, and from other safety-critical domains like ground-based power plants, and aviation. Additionally, there is a concern that any potential or real issues to date may have been masked by the fact that crews have near constant access to ground controllers, who monitor for errors, correct mistakes, and provide additional information needed to complete tasks. We do not know what types of HCI issues might arise without this "safety net". Exploration missions will test this concern, as crews may be operating autonomously due to communication delays and blackouts. Crew survival will be heavily dependent on available electronic information for just-in-time training, procedure execution, and vehicle or system maintenance; hence, the criticality of the Risk of Inadequate HCI. Future work must focus on identifying the most important contributing risk factors, evaluating their contribution to the overall risk, and developing appropriate mitigations. The Risk of Inadequate HCI includes eight core contributing factors based on the Human Factors Analysis and Classification System (HFACS): (1) Requirements, policies, and design processes, (2) Information resources and support, (3) Allocation of attention, (4) Cognitive overload, (5) Environmentally induced perceptual changes, (6) Misperception and misinterpretation of displayed information, (7) Spatial disorientation, and (8) Displays and controls.

Kritina Holden↗

Evaluation of KU-band SPA and EA-1 software, task no. 2

The software for selected space shuttle components are examined for potential problem areas and possible suboptimal coding techniques. Errors discovered in the procedure termination routine are discussed. The technique used to store status flags is found not to be as efficient as possible, from both the standpoints of memory usage and processing time. Self-test routines are examined with the intent of providing a more detailed description of multiplexer/demultiplexer and display and control outputs. An example of a reconfiguration is presented with a block diagram to indicate the critical paths. A cross-reference table of status flags is given to aid in tracing the functional activity of software modules.

Dobbs, J.↗

GMAO Seasonal Forecast Ensemble Exploration

GMAO Sub/Seasonal prediction system (S2S) has recently been upgraded. A complete set (1981-2016) of 9-months hindcasts for the previous and current versions (S2S-1.0 and S2S-2.1 respectively) allows for the evaluation of the forecast skill and a study of various characteristics of the ensemble forecasts in particular. We compared the intra-seasonal, interannual and intra-ensemble SST variability of the two systems against the observed. Focusing on the ENSO SST indices, we analyzed the consistency of the forecasts ensembles by studying rank histograms and comparing the ensemble spread with the standard error of the estimate.The S2S-2.1 ensemble appears to be more consistent with observations in Niño1+2 region compared to S2S-1.0, while in the central equatorial Pacific ocean this measure is comparably good for both systems. The S2S-1.0 system tends to be under dispersive, while the new system is under dispersive only at very short lead times, but tends to be over dispersive at long leads and for forecasts verifying in spring in Niño 3.4 region.Overall, the new system has greater skill in predicting ENSO. The evaluation techniques tested here will be applied for testing of the next generation sub/seasonal forecast system under development.

Borovikov, Anna↗

Simulation and Optimizing Groundwater Investigations, Conceptual Design Analysis, Model Calibration and Remedial Design Using Parallel Numerical Methods: Theory and Application - 16578

Optimization of environmental restoration efforts at complex groundwater sites involves the development of a conceptual and often numerical flow and transport models. Typically, a limited number of simulations are done to assess the impact of parameter uncertainty, to evaluate remedial alternatives and develop a remedy, or to modify an existing remedy. A limited number of simulations are performed because each simulation can take upwards of several hours per simulation and until now the ability to conduct the 100's to 1000's of simulations needed to derive an optimal remedy was beyond the reach of the everyday practitioner. HGL has developed and implemented new parallel simulation and optimization methods in their software and has successfully deployed applications on high performance Cloud-based computing platforms. This advancement removes the impediment of burdensome and unpredictable runtime requirements these types of projects have typically exhibited using heuristic trial-and-error or even automated serial computation approaches. A parallel numerical solver has been added to the subsurface flow and transport simulator MODFLOW-SURFACT{sup TM} and has demonstrated reductions in solution time by as much as a factor of 9.8. This is accomplished by distributing the numerical computations of every time step over multiple processors. The integrated parallel simulation and optimization method uses the breadth and depth search approach; iteratively producing sets of tens or 100's of candidate solutions that are evaluated concurrently instead sequentially (i.e. one model evaluation at a time). Solutions within 80-100% of the globally optimal solution are regularly found expending only 5-10% of the usual computational effort using this new parallel optimization method (reducing computation effort by 10 to 20 times). Distributing optimal search runs across multiple processors additionally reduces calendar time to solutions. Practitioners evaluating multiple conceptual site models, model calibration by auto-calibration methods and remedial design by heuristics can enjoy significant reductions in model solution time by using the parallel numerical matrix solver in the physics-based simulator. Combining investigation and design analysis with breadth and depth parallel optimization methods, projects at the fringe of practical optimization, along with projects that were beyond the reach of formal optimization methods are now wholly feasible and tractable optimization problems. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Troubleshooting deep-learner training data problems using an evolutionary algorithm on Summit

Architectural and hyper-parameter design choices can influence deep-learner (DL) model fidelity but can also be affected by malformed training and validation data. However, practitioners may spend significant time refining layers and hyper-parameters before discovering that distorted training data was impeding training progress. We found that an evolutionary algorithm (EA) can be used to diagnose this kind of DL problem. An EA evaluated thousands of DL configurations on Summit that yielded no overall improvement in DL performance, which suggested problems with the training and validation data. We suspected that Contrast Limited Adaptive Histogram Equalization (CLAHE) enhancement that was applied to previously generated digital surface models (DSMs), for which we were training DLs to find errors, had damaged the training data. Subsequent runs with an alternative global normalization yielded significantly improved DL performance. However, the DL Intersection Over Union (IOU) still exhibited consistent sub-par performance, which suggested further problems with the training data and DL approach. Nonetheless, we were able to diagnose this problem within a 12-hour span via Summit runs, which prevented several weeks of unproductive trial-and-error DL configuration refinement, and allowed for a more timely convergence on an ultimately viable solution.

97 MATHEMATICS AND COMPUTING↗

A 3D global-to-local deformable mesh model based registration and anatomy-constrained segmentation method for image guided prostate radiotherapy

Purpose: In the external beam radiation treatment of prostate cancers, successful implementation of adaptive radiotherapy and conformal radiation dose delivery is highly dependent on precise and expeditious segmentation and registration of the prostate volume between the simulation and the treatment images. The purpose of this study is to develop a novel, fast, and accurate segmentation and registration method to increase the computational efficiency to meet the restricted clinical treatment time requirement in image guided radiotherapy. Methods: The method developed in this study used soft tissues to capture the transformation between the 3D planning CT (pCT) images and 3D cone-beam CT (CBCT) treatment images. The method incorporated a global-to-local deformable mesh model based registration framework as well as an automatic anatomy-constrained robust active shape model (ACRASM) based segmentation algorithm in the 3D CBCT images. The global registration was based on the mutual information method, and the local registration was to minimize the Euclidian distance of the corresponding nodal points from the global transformation of deformable mesh models, which implicitly used the information of the segmented target volume. The method was applied on six data sets of prostate cancer patients. Target volumes delineated by the same radiation oncologist on the pCT and CBCT were chosen as the benchmarks and were compared to the segmented and registered results. The distance-based and the volume-based estimators were used to quantitatively evaluate the results of segmentation and registration. Results: The ACRASM segmentation algorithm was compared to the original active shape model (ASM) algorithm by evaluating the values of the distance-based estimators. With respect to the corresponding benchmarks, the mean distance ranged from -0.85 to 0.84 mm for ACRASM and from -1.44 to 1.17 mm for ASM. The mean absolute distance ranged from 1.77 to 3.07 mm for ACRASM and from 2.45 to 6.54 mm for ASM. The volume overlap ratio ranged from 79% to 91% for ACRASM and from 44% to 80% for ASM. These data demonstrated that the segmentation results of ACRASM were in better agreement with the corresponding benchmarks than those of ASM. The developed registration algorithm was quantitatively evaluated by comparing the registered target volumes from the pCT to the benchmarks on the CBCT. The mean distance and the root mean square error ranged from 0.38 to 2.2 mm and from 0.45 to 2.36 mm, respectively, between the CBCT images and the registered pCT. The mean overlap ratio of the prostate volumes ranged from 85.2% to 95% after registration. The average time of the ACRASM-based segmentation was under 1 min. The average time of the global transformation was from 2 to 4 min on two 3D volumes and the average time of the local transformation was from 20 to 34 s on two deformable superquadrics mesh models. Conclusions: A novel and fast segmentation and deformable registration method was developed to capture the transformation between the planning and treatment images for external beam radiotherapy of prostate cancers. This method increases the computational efficiency and may provide foundation to achieve real time adaptive radiotherapy.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Vector-model-supported approach in prostate plan optimization

Lengthy time consumed in traditional manual plan optimization can limit the use of step-and-shoot intensity-modulated radiotherapy/volumetric-modulated radiotherapy (S&S IMRT/VMAT). A vector model base, retrieving similar radiotherapy cases, was developed with respect to the structural and physiologic features extracted from the Digital Imaging and Communications in Medicine (DICOM) files. Planning parameters were retrieved from the selected similar reference case and applied to the test case to bypass the gradual adjustment of planning parameters. Therefore, the planning time spent on the traditional trial-and-error manual optimization approach in the beginning of optimization could be reduced. Each S&S IMRT/VMAT prostate reference database comprised 100 previously treated cases. Prostate cases were replanned with both traditional optimization and vector-model-supported optimization based on the oncologists' clinical dose prescriptions. A total of 360 plans, which consisted of 30 cases of S&S IMRT, 30 cases of 1-arc VMAT, and 30 cases of 2-arc VMAT plans including first optimization and final optimization with/without vector-model-supported optimization, were compared using the 2-sided t-test and paired Wilcoxon signed rank test, with a significance level of 0.05 and a false discovery rate of less than 0.05. For S&S IMRT, 1-arc VMAT, and 2-arc VMAT prostate plans, there was a significant reduction in the planning time and iteration with vector-model-supported optimization by almost 50%. When the first optimization plans were compared, 2-arc VMAT prostate plans had better plan quality than 1-arc VMAT plans. The volume receiving 35 Gy in the femoral head for 2-arc VMAT plans was reduced with the vector-model-supported optimization compared with the traditional manual optimization approach. Otherwise, the quality of plans from both approaches was comparable. Vector-model-supported optimization was shown to offer much shortened planning time and iteration number without compromising the plan quality.

61 RADIATION PROTECTION AND DOSIMETRY↗

Generative Adversarial Network for Synthetic Time Series Data Generation in Smart Grids

The availability of fine grained time series data is a pre-requisite for research in smart-grids. While data for transmission systems is relatively easily obtainable, issues related to data collection, security and privacy hinder the widespread public availability/accessibility of such datasets at the distribution system level. This has prevented the larger research community from effectively applying sophisticated machine learning algorithms to significantly improve the distribution-level accuracy of predictions and increase the efficiency of grid operations. Synthetic dataset generation has proven to be a promising solution for addressing data availability issues in various domains such as computer vision, natural language processing and medicine. However, its exploration in the smart grid context remains unsatisfactory. Previous works have tried to generate synthetic datasets by modeling the underlying system dynamics: an approach which is difficult, time consuming, error prone and often times infeasible in many problems. In this work, we propose a novel data-driven approach to synthetic dataset generation by utilizing deep generative adversarial networks (GAN) to learn the conditional probability distribution of essential features in the real dataset and generate samples based on the learned distribution. To evaluate our synthetically generated dataset, we measure the maximum mean discrepancy (MMD) between real and synthetic datasets as probability distributions, and show that their sampling distance converges. To further validate our synthetic dataset, we perform common smart grid tasks such as k-means clustering and short-term prediction on both datasets. Experimental results show the efficacy of our synthetic dataset approach: the real and synthetic datasets are indistinguishable by solely examining the output of these tasks.

14 SOLAR ENERGY↗

Method to Extract System-Independent Material Properties From Dual-Energy X-Ray CT

In 2016, we published a method for processing dual-energy computed tomography (DECT) data called system-independent rho-e/Z-e (ρ e /Z e ) or SIRZ. Using data from multiple DECT systems and spectra, SIRZ estimated the electron density ρ e and effective atomic number Z e (based on published X-ray cross sections), for a set of known specimens. However, the decomposition process required complex spectral modeling of the DECT system, which made SIRZ difficult to implement and automate. This paper describes the subsequent work on “SIRZ-2” to simplify the spectral modeling, automate the process, and improve its range and versatility. The SIRZ-2 basis functions are more accurate for the X-ray energy range (<;200 keV) and materials that are common to DECT applications in security and nondestructive characterization. To evaluate the SIRZ-2 performance, eight well-characterized specimens (with Z e of 6-20) were scanned with multiple spectral pairs (up to 200 keV) on seven different DECT systems, including a commercial airport luggage scanner. For these varied tests, the average SIRZ-2 relative errors for ρ e estimates were four times lower (0.7% compared to 3.0% for SIRZ), while average Z e relative errors were comparably low for both methods (<;1.5%). Average precision errors were also low (<;2.3% for both). SIRZ-2 produces accurate, repeatable, and system-independent material property estimates of (ρ e , Z e ) that can be directly compared across DECT systems and over time.

42 ENGINEERING↗

A bronchoscopic navigation system using bronchoscope center calibration for accurate registration of electromagnetic tracker and CT volume without markers

Purpose: Various bronchoscopic navigation systems are developed for diagnosis, staging, and treatment of lung and bronchus cancers. To construct electromagnetically navigated bronchoscopy systems, registration of preoperative images and an electromagnetic tracker must be performed. This paper proposes a new marker-free registration method, which uses the centerlines of the bronchial tree and the center of a bronchoscope tip where an electromagnetic sensor is attached, to align preoperative images and electromagnetic tracker systems. Methods: The chest computed tomography (CT) volume (preoperative images) was segmented to extract the bronchial centerlines. An electromagnetic sensor was fixed at the bronchoscope tip surface. A model was designed and printed using a 3D printer to calibrate the relationship between the fixed sensor and the bronchoscope tip center. For each sensor measurement that includes sensor position and orientation information, its corresponding bronchoscope tip center position was calculated. By minimizing the distance between each bronchoscope tip center position and the bronchial centerlines, the spatial alignment of the electromagnetic tracker system and the CT volume was determined. After obtaining the spatial alignment, an electromagnetic navigation bronchoscopy system was established to real-timely track or locate a bronchoscope inside the bronchial tree during bronchoscopic examinations. Results: The electromagnetic navigation bronchoscopy system was validated on a dynamic bronchial phantom that can simulate respiratory motion with a breath rate range of 0–10 min{sup −1}. The fiducial and target registration errors of this navigation system were evaluated. The average fiducial registration error was reduced from 8.7 to 6.6 mm. The average target registration error, which indicates all tracked or navigated bronchoscope position accuracy, was much reduced from 6.8 to 4.5 mm compared to previous registration methods. Conclusions: An electromagnetically navigated bronchoscopy system was constructed with accurate registration of an electromagnetic tracker and the CT volume on the basis of an improved marker-free registration approach that uses the bronchial centerlines and bronchoscope tip center information. The fiducial and target registration errors of our electromagnetic navigation system were about 6.6 and 4.5 mm in dynamic bronchial phantom validation.

60 APPLIED LIFE SCIENCES↗

Experiences Detecting Defective Hardware in Exascale Supercomputers

In May 2022, the newest supercomputer to top the TOP 500 list was Frontier at Oak Ridge National Laboratory, demonstrating the capability of computing more than 1.1 quintillion (1018) floating-point calculations every second. Driving this ground-breaking rate of computing is Frontier’s more than 37,000 graphics processing units (GPUs) and 9,408 central processing units (CPUs). In total, Frontier contains more than 60 million parts. At this scale, the smallest margin of error may generate hundreds of hardware errors across the system. These errors are capable of directly hindering world-class science performed on Frontier if not found. In this work, we describe and evaluate two strategies for finding hardware-level faults in Frontier’s 9,408 compute nodes. There are two strategies developed: the first uses the Slurm scheduler to scavenge available compute time to run the node screen, the second builds upon the lessons learned in the first strategy and enforces a weekly screen of each node. Using June 2023 as a case study, we find that the first scheduling strategy consumed more than ten times the resources as the second scheduling strategy, but successfully detected five hardware defects in Frontier. We summarize the lessons learned while developing and running a node screen on the world’s first exascale supercomputer.

Hagerty, Nick↗

Classical eight- and lower-order Runge-Kutta-Nystroem formulas with stepsize control for special second-order differential equations

The formulas include a stepsize control procedure, based on a complete coverage of the leading term of the truncation error in x. The formulas require fewer evaluations per stop than other Runge-Kutta-Nystrom formulas if the latter are operated by using the standard procedure for stepsize control. An example is presented. With results being of the same accuracy, Runge-Kutta-Nystrom formulas discussed save 50 percent or more computer time compared with other Runge-Kutta-Nystrom formulas.

Fehlberg, E.↗

A study of stall deterrent systems for general aviation aircraft

Stall deterrent concepts for general aviation aircraft have been investigated using simulation studies and flight test experiments. It was found that the simulator was suitable for the development of deterrent system concepts, but the simulator was unacceptable for pilot evaluation of system effectiveness under typical stall/spin accident conditions. A Cessna 319 was outfitted with sensors, servoactuators, and analog control logic necessary to investigate a wide range of stall deterrent systems. It was found that an acoustic stall sensor and an error control law were sufficient for stall deterrence. The pitch intervention control system prevented aircraft stall for all aircraft configurations and pilot inputs. The variable up elevator stop concept was found to be effective for slow decelerations to aircraft stall. The stall deterrent systems were evaluated by four professional pilots and three low time pilots.

Bennett, A. G.↗

Software-implemented fault insertion: An FTMP example

This report presents a model for fault insertion through software; describes its implementation on a fault-tolerant computer, FTMP; presents a summary of fault detection, identification, and reconfiguration data collected with software-implemented fault insertion; and compares the results to hardware fault insertion data. Experimental results show detection time to be a function of time of insertion and system workload. For the fault detection time, there is no correlation between software-inserted faults and hardware-inserted faults; this is because hardware-inserted faults must manifest as errors before detection, whereas software-inserted faults immediately exercise the error detection mechanisms. In summary, the software-implemented fault insertion is able to be used as an evaluation technique for the fault-handling capabilities of a system in fault detection, identification and recovery. Although the software-inserted faults do not map directly to hardware-inserted faults, experiments show software-implemented fault insertion is capable of emulating hardware fault insertion, with greater ease and automation.

Czeck, Edward W.↗

Airborne gamma radiation soil moisture measurements over short flight lines

Results are presented on airborne gamma radiation measurements of soil moisture condition, carried out along short flight lines as part of the First International Satellite Land Surface Climatology Project Field Experiment (FIFE). Data were collected over an area in Kansas during the summers of 1987 and 1989. The airborne surveys, together with ground measurements, provide the most comprehensive set of airborne and ground truth data available in the U.S. for calibrating and evaluating airborne gamma flight lines. Analysis showed that, using standard National Weather Service weights for the K, Tl, and Gc radiation windows, the airborne soil moisture estimates for the FIFE lines had a root mean square error of no greater than 3.0 percent soil moisture. The soil moisture estimates for sections having acquisition time of at least 15 sec were found to be reliable.

Peck, Eugene L.↗

SU-G-BRA-05: Application of a Feature-Based Tracking Algorithm to KV X-Ray Fluoroscopic Images Toward Marker-Less Real-Time Tumor Tracking

Purpose: To detect target position on kV X-ray fluoroscopic images using a feature-based tracking algorithm, Accelerated-KAZE (AKAZE), for markerless real-time tumor tracking (RTTT). Methods: Twelve lung cancer patients treated with RTTT on the Vero4DRT (Mitsubishi Heavy Industries, Japan, and Brainlab AG, Feldkirchen, Germany) were enrolled in this study. Respiratory tumor movement was greater than 10 mm. Three to five fiducial markers were implanted around the lung tumor transbronchially for each patient. Before beam delivery, external infrared (IR) markers and the fiducial markers were monitored for 20 to 40 s with the IR camera every 16.7 ms and with an orthogonal kV x-ray imaging subsystem every 80 or 160 ms, respectively. Target positions derived from the fiducial markers were determined on the orthogonal kV x-ray images, which were used as the ground truth in this study. Meanwhile, tracking positions were identified by AKAZE. Among a lot of feature points, AKAZE found high-quality feature points through sequential cross-check and distance-check between two consecutive images. Then, these 2D positional data were converted to the 3D positional data by a transformation matrix with a predefined calibration parameter. Root mean square error (RMSE) was calculated to evaluate the difference between 3D tracking and target positions. A total of 393 frames was analyzed. The experiment was conducted on a personal computer with 16 GB RAM, Intel Core i7-2600, 3.4 GHz processor. Results: Reproducibility of the target position during the same respiratory phase was 0.6 +/− 0.6 mm (range, 0.1–3.3 mm). Mean +/− SD of the RMSEs was 0.3 +/− 0.2 mm (range, 0.0–1.0 mm). Median computation time per frame was 179 msec (range, 154–247 msec). Conclusion: AKAZE successfully and quickly detected the target position on kV X-ray fluoroscopic images. Initial results indicate that the differences between 3D tracking and target position would be clinically acceptable.

60 APPLIED LIFE SCIENCES↗

Boosting Noise2Inverse via enhanced model selection for denoising computed tomography data

Synchrotron-based x-ray tomographic imaging enables the examination of the internal structure of materials at high spatial and temporal resolution. Experimental constraints can impose dose and time limits on the measurements, introducing a higher level of noise and artifacts in the reconstructed images. Deep learning has emerged as a powerful tool to remove noise from reconstructed images. Recently, the Noise2Inverse method was designed specifically for denoising reconstructed images without requiring paired noisy and clean images. This method creates multiple statistically independent reconstructions used to pair the data in which training involves transforming one reconstruction into the other, and vice versa. Originally designed to be used after a fixed number of epochs, we see in practice that this approach may not produce the optimal model and may unnecessarily waste computational resources. Therefore, we propose an alternative method of identifying the best model during training that aligns with the Noise2Inverse method. During validation, we compare the model output of the multiple reconstructions among each other. We hypothesize that the best model is the one that produces images with the highest similarity, implying a convergence in the predicted material properties and absorption values. To compare model outputs, we consider the absolute error, square error, structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), and cosine similarity. We evaluate our method on two simulated tomography datasets and two, real-world, low-contrast, high-energy x-ray tomography datasets. We show our approach is more effective at determining the best model, up to an increase of 12.50% and 12.53% in SSIM and PSNR, respectively, while only requiring a fifth of the training time compared to the original approach.

CT↗