Normalized Root Mean Research Articles

Performance evaluation of CMIP6 models for application to hydrological modelling studies – A case study of Australia

Accurate estimation of climate change impacts on catchment hydrology is essential for effective future water management. The efficacy of such estimations is dependent on proper climate model selection. In this study, an attempt was made to formulate a methodology for climate model selection, evaluating eight climate models from the sixth phase of the Coupled Model Intercomparison Project (CMIP6). The models were assessed for their ability to simulate variables used in hydrological studies and large-scale atmospheric circulation influencing rainfall in Australia. Five statistical indicators Root Mean Square Error (RMSE), Spatial Correlation (SC), Percentage Bias (Pbias), Normalized Root Mean Square Error (NRMSE), and Nash-Sutcliffe Efficiency (NSE) were used to evaluate the performance, and the models were ranked through Compromise Programming (CP), a multiple criteria decision making technique. Results show that HadGEM3-GC31-LL performed well in most of the categories considered and was top top-ranked model overall followed by GFDL-ESM4, CESM2-CAM6-RT, and CanESM5 for Australia. Conversely, MIROC6 consistently ranked lower in most of the categories. In the context of simulating hydrological variables, CESM2-CAM6-RT, HadGEM3-GC31-LL, and GFDL-ESM4 emerged as the top three models. The robustness of the proposed methodology suggests its applicability for model selection, making it a replicable approach for climate change impact assessment studies in diverse regions.

A new methodology for determining the central pressure waveform from peripheral measurement using Fourier-based machine learning

Radial applanation tonometry is a well-established technique for hemodynamic monitoring and is becoming popular in affordable non-invasive wearable healthcare electronics. To assess the central aortic pressure using radial-based measurements, there is an essential need to develop mathematical approaches to estimate the central pressure waveform. In this study, we propose a new Fourier-based machine learning (F-ML) methodology to transfer non-invasive radial pressure measurements to the central pressure waveform. To test the method, collection of tonometry recordings of the radial and carotid pressure measurements are used from the Framingham Heart Study (2640 individuals, 55 % women) with mean (range) age of 66 (40–91) years. Method-derived estimates are significantly correlated with the measured ones for three major features of the pressure waveform (systolic blood pressure, r=0.97, p < 0.001; diastolic blood pressure, r=0.99, p < 0.001; and mean blood pressure, r=0.99, p < 0.001). In all cases, the Bland-Altman analysis shows negligible bias in the estimations and error is bounded to 5.4 mmHg. Findings also suggest that the F-ML approach reconstructs the shape of the central pressure waveform accurately with the average normalized root mean square error of 5.5 % in the testing population which is blinded to all stages of machine learning development. The results show that the F-ML transfer function outperforms the conventional generalized transfer function, particularly in terms of reconstructing the shape of the central pressure waveform morphology. The proposed F-ML transfer function can provide accurate estimates for the central pressure waveform, and ultimately expand the usage of non-invasive devices for central hemodynamic assessment.

Neural network models and shapley additive explanations for a beam-ring structure

A combination of neural network modeling, SHapley Additive exPlanations (SHAP) and model simplification is proposed and applied to the system of a beam-ring structure in this paper. Based on the long short-term memory based encoder-decoder (LSTM ED), a new framework (P_LSTM ED) is presented to improve prediction accuracy. Several iterative methods are selected to optimize the neural network model under the new framework with LSTM EDs. The normalized root mean square error (NRMSE) is regarded as an evaluation metric for the prediction accuracy of neural network models. The SHAP method provides interpretability for the optimal neural network model and there are two applications through the SHAP analysis. One is a modification strategy for inputs of the neural network model, the other is a model simplification method for state equations of the nonlinear system. Numerical simulation shows that the P_LSTM ED has higher prediction accuracy while considering efficiency. The neural network model with the modification strategy applied achieves a maximum accuracy improvement of 24.15 % with less training time. The effectiveness and generality of model simplification are verified by numerical simulation comparison between simplified and original equations.

Spatial downscaling of rainfall from general circulation models: A chaos theory-based framework

The present study proposes a new chaos theory-based local approximation technique for the spatial downscaling of rainfall from GCM outputs. The approach involves: (1) reconstruction of relevant climate data in the state space using an embedding technique; and (2) use of a neighbor search procedure in the reconstructed space and local approximation for downscaling. The approach is implemented for downscaling of monthly rainfall at a spatial resolution of 0.25° × 0.25° across India. The effectiveness of the approach is first demonstrated using the NCEP/NCAR reanalysis dataset (1979–2014) and then tested using the outputs from five CMIP6 GCMs (CMCC-ESM2, CESM2-WACCM, CMCC-CM2-HR4, E3SM-1-1-ECA, and BCC-ESM1) (1961–2014). Eight climate variables (eastward wind, northward wind, relative humidity, specific humidity, surface temperature, air temperature, sea level pressure, and geopotential height) are considered as predictors, with the observed rainfall data from the India Meteorological Department (IMD) as the predictand. Three statistical measures, namely correlation coefficient (CC), Nash-Sutcliffe efficiency (NSE), and normalized root mean square error (NRMSE), are used for evaluation. The results indicate that the local approximation downscaling approach yields promising results for both reanalysis dataset and GCM outputs. Among the five GCMs, the CMCC-ESM2 model yields the best results. Comparison of the results from the chaos theory-based approach with those of the linear regression-based approach indicates better performance of the former. The proposed framework is general and can be applied for any climate variable, GCM, and region.

Mathematical modeling and performance evaluation of Ducted Horizontal-axis Helical Wind Turbines: Insights into aerodynamics and efficiency.

With the escalating demand for energy, there is a growing focus on decentralized, small-scale energy infrastructure. The success of new turbines in this context is notable. However, many of these turbines do not follow many of the basic ideas established to evaluate their performance, leaving no precise technique or mathematical model. This research developed a Ducted Horizontal-axis Helical Wind Turbine (DHAHWT). The DHAHWT is a duct-mounted helical savonius turbine with a venturi and diffuser to improve flow. Unlike a vertical axis helical savonius turbine, DHAHWT revolves roughly parallel to the wind, making it a horizontal turbine. This complicates mathematical and theoretical analysis. This study created a DHAHWT mathematical model. COMSOL simulations utilizing Menter's Shear Stress Transport model (SST) across an incoming velocity range of 1m/s to 4m/s were used to evaluate the turbine's interaction with the wind. MATLAB was used to train an artificial neural network (ANN) utilizing COMSOL data to obtain greater velocity data. The Mean Average Percentage Error (MAPE) and Root Mean Square Error (RMSE) of ANN data were found to be 3%, indicating high accuracy. Further, using advanced statistical methods the Pearson's correlation coefficient was calculated resulting in a better understanding of the relationship of between incoming velocity and velocity at different sections of the wind turbine. This study will shed light on the aerodynamics and working of DHAHWT.

A method to rapidly construct 3D canopy scenes for maize and their spectral response evaluation

Three-dimensional (3D) scenes of a maize canopy can be utilized to depict its vertical structure and serve as the foundation for a 3D radiative transfer model. Given the inherent heterogeneity of canopy structures, evaluating their spectral response is crucial for remote-sensing inversion algorithms. Yet existing methods to model 3D canopy scenes of crops lack the high efficiency required for large-scale breeding applications. Hence, this study aims to develop a method to rapidly construct 3D canopy scenes of maize based on key structural parameters—namely, leaf area, base angle, and inclination angle—across multiple cultivars and growth stages. The accuracy of this canopy scenes modeling method was validated by comparison with a fine-scale 3D model and a multi-growth stage reflectance dataset. The root mean square error (RMSE) and normalized root mean square error (NRMSE) between the fine-scale 3D model and the constructed 3D model were less than 0.021 and 6.4%, respectively. Moreover, the RMSE and NRMSE between the simulated reflectance and measured reflectance were less than 0.042 and 11.1%, respectively. Next, we integrated the 3D radiative transfer model with 3D scenes to analyze the spectral response of maize canopy structure. The findings revealed that the near-infrared band was affected more by leaf area than leaf base angle, while the opposite is observed for the red-edge band. Furthermore, the middle layer contributed more to the canopy reflectance than either the upper or lower layers. Notably, two maize scenes where two structural parameters simultaneously change could elicit the same canopy spectra. The proposed method thus offers the dual advantages of rapid modeling and high efficiency, making it highly applicable for maize and other similar crops.

The combined machine learning model SMOTER-GA-RF for methane yield prediction during anaerobic digestion of straw lignocellulose based on random forest regression

Simulating anaerobic digestion (AD) using a machine learning (ML) is important for guiding methane production in practice. But the performance of the ML is affected by operational conditions and microbial types and quantities, which lead to low accuracy of the traditional model. In this study, a combined ML algorithm SMOTER-GA-RF was constructed by expanding the number of genomic data, introducing oversampling technique (SMOTER) and genetic algorithm (GA) to increase prediction accuracy, which the test set R2 increased 36.22%. The normalized root mean square error and percent bias of the validation set (D2) were 8.92% (<10%) and 4.13% (<15%), indicating the accuracy of prediction. Simultaneously, the predictive accuracy of the validation sets (V1–V5) in the presence of Gaussian white noise remained high, confirming the robustness of the SMOTER-GA-RF. Furthermore, similar ML models were collected and compared, and it was found that SMOTER-GA-RF constructed in this paper had best effect. Finally, important factors affecting the AD performance of straw were analyzed using SHAP. The hydraulic retention time was controlled at approximately 40 d, volatile fatty acid should be controlled within 1000 mg/L, and the relative abundances of the important archaea Methanoculleus, Methanobacterium and Methanosarcina in the second stage of AD were controlled at 5–10%, <18%, and <5%.

Estimation and analysis of GNSS specific-observation biases using constrained rank-deficient virtual observation algorithm

ABSTRACT This paper initially introduces two algorithms for the constrained rank-deficient indirect adjustment model and validates their equivalence. Subsequently, using DCBs provided by the Chinese Academy of Sciences (CAS) and the German Aerospace Center (DLR) as observed values, and CAS-provided OSBs as reference values, the accuracy of the virtual observation algorithm is verified. The results indicate that the average Root Mean Square (RMS) values between the estimated OSB value obtained based on CAS’s DCB and the reference value is mostly below 0.1 ns. The average RMS values between the OSB estimates derived from DLR’s DCBs for GPS, GLONASS, BDS, and Galileo compared to the reference values are 0.31 ns, 0.37 ns, 0.55 ns, and 0.24 ns, respectively, demonstrating a good level of consistency. This further validates the accuracy of the algorithm.

Wind turbine gearbox oil temperature feature extraction and condition monitoring based on energy flow

Supervisory Control and Data Acquisition (SCADA) data is widely used for wind turbine gearbox condition monitoring (WTGCM) due to its easy access and low cost, thus ensuring the safe and reliable operation of wind turbine gearboxes. However, existing WTGCM methods based on SCADA overlook the fundamental thermal physics and sensor failure. To effectively address the challenges of weak feature generalization and numerous false alarms, we propose a highly robust WTGCM method based on energy flow. Firstly, based on the principles of thermal balance and wind energy conversion, a feature extraction method considering the physical change process of oil temperature is proposed. Secondly, the performance of different feature extraction methods is quantitatively analyzed using data evaluation criteria. The reasons why different feature extraction methods have different effects are revealed through dimensionality reduction analysis. Finally, we compare the proposed method with other feature extraction methods using four models to validate the excellent performance of the proposed method in the early fault warning. The results demonstrate that the proposed method reduces the average root mean square error by 18.21% during the healthy operation of the wind turbine gearboxes. Moreover, the proposed method performs excellently in early warning of mechanical and sensor faults, significantly reducing the number of false alarms. The average warning time for mechanical faults is 32.56 h earlier.

The prototype of the front-end electronics for STCF muon detector

The muon detector (MUD), serving as the outermost detector of the high-precision spectrometer in STCF, is used to provide muon identification in the presence of a significant pion background. The accuracy of muon identification relies heavily on excellent momentum resolution, which can be determined by their flight trajectory positions. Physical simulations of the measurement precision for reconstructed muons indicate that a spatial resolution of 2 cm is required. In the barrel MUD, a double-ended readout is required to determine the hit position, with a time resolution requirement of approximately 500 ps. To meet the readout requirements of the MUD, a comprehensive scheme for the front-end readout electronics of the STCF MUD is proposed, and a prototype of MUD readout electronics is developed. To validate the final 8-channel application-specific integrated circuit (ASIC) design, an 8-channel time-to-digital converter (TDC) is implemented using a field programmable gate array (FPGA). The results of the electronics tests demonstrate that the average root mean square (RMS) precision ranges from 14 to 16 ps for each channel within a 1∼20 ns time interval. In the joint test with the detector, the system achieves a single-channel RMS precision of 297 ps, with a detection efficiency exceeding 95.5%. All indicators meet project requirements. This validates that the prototype is capable of preliminary evaluation and parameter optimization of the STCF MUD.

Forecasting daily rainfall in a humid subtropical area: an innovative machine learning approach

ABSTRACT Hydrological modeling is one of the most complicated tasks in sustainable water resources management, particularly in terms of predicting rainfall. Predicting rainfall is critical to build a sustainable society in terms of hydropower operations, agricultural planning, and flood control. In this study, a hybrid model based on the integration of k-nearest neighbor (KNN), XGBoost (XGB), decision tree (DCT), and Random Forest (RF) has been developed and implemented for forecasting daily rainfall for the first time at Sydney airport, Australia. Daily rainfall, temperature, evaporation, and humidity have been selected as input parameters. Three statistical measurements, namely, root mean square error (RMSE), coefficient of determination (R2), mean absolute error (MAE), and Normalized Root Mean Square Error (NRMSE) have been utilized in order to check the accuracy of the proposed model. A sensitivity analysis was conducted, and the results indicated that for the purpose of prediction, the temperature, humidity, and evaporation were highly sensitive to the rainfall data. According to the results, the developed hybrid model was capable of predicting daily rainfall with high performance for both training and testing parts with RMSE = 0.124, R2 = 0.999, MAE = 0.007, NRMSE = 0.04 and RMSE = 1.246, R2 = 0.991, MAE = 0.109, NRMSE = 0.339, respectively.

Towards integration of LOTOS-EUROS high resolution simulations and heterogenous low-cost sensor observations

Official air quality monitoring networks are often scarce and unevenly spatially distributed. In recent years, the use of low-cost sensors next to official networks is increasing. These additional networks provide measurements of high spatial and temporal resolution and potentially reveal patterns and emissions sources that are hard to detect with conventional methods. In this work, the data assimilation method implemented around the LOTOS-EUROS chemistry transport model is employed to assimilate measurements from heterogenous low-cost sensor networks around the city of Eindhoven in the Netherlands in November 2021. Three data assimilation experiments are performed and evaluated against a free run of the model. In the first one, measurements from the low-cost Innovatief Lucht Meetsysteem (ILM) network are exploited. In the second one, the citizen science network SamenMeten is used and in the third one, a combination of both datasets is applied. In the assimilation experiments at a domain around the city of Eindhoven, it is shown to be essential to use boundary conditions from an assimilation on a larger domain to account for the variability in pollution that originates from sources outside the domain of interest. Such an improvement in boundary conditions counts for a decrease in the initial free run negative biases of 45% for PM10 and 23% for PM2.5 in the city of Eindhoven. The assimilation of low-cost measurements in the region after the correction of the boundary conditions decreases the absolute PM10 biases in the 3 independent official stations over the city of Eindhoven further from −4.4 μg m−3 to about 0.8 μg m−3 averaged over the three experiments. Also, the correlation coefficient is increased from 0.75 to 0.89 and the normalized root mean square is decreased from 0.47 to 0.25. We conclude that the improved boundary conditions and assimilation of observations from dense low-cost networks are able to improve the LOTOS-EUROS simulations at urban scale.

A gas sensor based on free-standing SWCNT film for selective recognition of toxic and flammable gases under thermal cycling protocols

The widespread adoption of e-nose devices based on chemiresistive materials has been hindered by issues related to sensor device complexity and reliability, specifically sensor drift, necessitating frequent recalibration and retraining of pattern recognition models. This study introduces a method for thermocycling a single sensor based on a free-standing network of single-walled carbon nanotubes (SWCNTs) to acquire signal patterns for selective analyte detection. Additionally, it employs a data filtering technique to compensate for the sensor drift. A free-standing SWCNT film, only a few nanometers thick, is thermally cycled via Joule heating between room temperature and 120 °C. Under these conditions, the sensitivity was tested towards NO2, H2S, and acetone vapors (10–25 ppm) in the mixture with dry air. Signal patterns produced through thermocycling were processed using CatBoost and LSTM algorithms. The accuracy of detection reached 90 % in the classification task, and the average root mean squared error of analyte concentration detection in the multioutput regression task was below 4 ppm. By combining original sensor design, thermocycling, signal filtering for drift compensation, and advanced pattern recognition models, this work contributes to overcoming the challenges in multivariate sensing systems, paving the way for practical applications of the more reliable chemiresistive sensors.

CPSM: A Dynamic Simulation Model for Cucumber Productivity in Solar Greenhouse Based on the Principle of Effective Accumulated Temperature

The Cucumber Productivity Simulation Model (CPSM) was developed to precisely predict the dynamic process of cucumber productivity in a solar greenhouse. This research conducted a variety of sowing experiments and collected data on cucumber productivity and meteorological conditions from 2013 to 2015 and 2018 to 2020. Employing the principles of least squares, the relationship between cucumber productivity indicators and effective accumulated temperature (EAT) was fitted, determining key crop parameters and constructing the CPSM. Validation of the model was conducted using independent experimental data, evaluating its simulation accuracy. The results indicate that (1) CPSM can dynamically and meticulously simulate the formation process of different productivity indicators in cucumber. Normalized root mean square errors (NRMSE) ranged from 0.44% to 19.64%, and mean relative errors (MRE) ranged from 0.31% to 17.23% across different productivity indicator models. The models for organ water content, maximum root length, specific leaf area, and organ fresh weight distribution index demonstrated high simulation accuracy, while others showed relatively high accuracy. (2) Simulation accuracy varied with indicators and varieties. 19 indicators (34.55%) exhibited high simulation accuracy and 30 indicators (54.55%) showed relatively high accuracy. The JY35 variety (10.44 ± 8.49%) outperformed the JS206 variety (13.44 ± 8.50%) in terms of simulation accuracy. The JY35 variety had 39 superior productivity indicators (70.91%) while the JS206 variety had sixteen (29.09%). CPSM utilizes easily accessible temperature data as its input, allowing for precise and detailed simulation of productivity indicators for cucumber production in solar greenhouses. This research lays a theoretical foundation and provides technical support for guiding intelligent production management, efficient utilization of agricultural resources, and climate change productivity assessment in solar greenhouse cucumber production.

An improved neural operator framework for large-scale CO2 storage operations

Geologic carbon storage (GCS) is a practical solution to mitigate the impact of climate change and achieve net-zero carbon emissions. However, due to complex subsurface characteristics and operational constraints, GCS can potentially induce seismic events and leakage to groundwater resources. Moreover, predicting subsurface response in pressure and saturation changes during GCS requires high-fidelity models with high computational times. Machine learning (ML) offers a promising solution to alleviate these challenges. Yet, ML-driven approaches in subsurface physics face significant obstacles, such as the substantial computational resources required for training large-scale models and the accuracy requirement as a reliable surrogate model. To tackle these issues, we propose an improved neural operator (INO) that has improved a DeepONet architecture to handle a realistic reservoir model efficiently, with computational efficiency and high accuracy.Our proposed INO framework has three key benefits: precision, adaptability, and cost-efficiency. For precision, it achieves an average root mean square error (RMSE) of 0.05% (1.5 psi) compared to the average reservoir pressure (3,150 psi) and a RMSE value of 0.016 in void fraction for CO2 saturation among testing cases. Although it is sensitive to high-error areas with steep state variable gradients, proper domain decomposition, ensemble methods, and/or integrating domain-specific knowledge could improve this limitation. For adaptability, the INO framework can be trained with a subset of the computational domain during each backpropagation to achieve better training adaptability, even for problems with high-dimensional inputs (i.e., heterogeneous fields). For example, it can be trained with random subsampling of only 0.2% of the full domain but can predict pressure and saturation in any space and time within the entire domain. For cost efficiency, training procedures with a subset of the whole computational domain significantly improve computational time. For example, for 90 training cases with 1.73 million cells and 50-time steps, the overall training time was about 2.5 h using a single GPU. A trained INO model takes 1 s per evaluation over 1.73 million and 50 time stamps, much faster than a high-fidelity simulator. Overall, the proposed INO approach with these improved benefits will enable us to perform large-scale GCS applications, enhancing the potential of ML models in the subsurface and other multiphysics problems.

Same therapy, same calcium mobilization? Exploring calcium exchange across body compartments using a patient-specific predictive model.

Comprehensive, patient-specific models are essential to study calcium deposition and mobilization during dialysis. We aim to develop tools to support clinical prescriptions with a more accurate approach for the prediction of calcium mobilization while also considering major electrolytes and catabolites. We modified a multi-solute model predicting patient-specific dialysis response by incorporating a calcium buffer to represent bone exchanges. Data from four centers, involving 127 patients with six sessions each, were utilized. For each patient, three sessions were allocated for model training (ID123), while the remaining sessions were for validation (PRED456). The normalized root mean square error (nRMSE%) was used to evaluate both descriptive and predictive accuracy. Correlations between initial data and calcium exchanges were also assessed. The overall nRMSE% for ID123 was 3.92%. For PRED456, it was 3.46% (ranging from a minimum of 1.17% for [Na+] to a maximum of 6.62% for [urea]). The median nRMSE% for plasma calcium varied between 1.13 and 8.32 for SHD sessions, depending on whether Ca_dialysis fluid (Cad) was ≥ or <1.50 mmol/L, respectively. For HDF sessions, the range was between 2.90 and 5.89. A significant and moderate correlation was found between overall calcium removal and the buffer balance. The most robust correlation observed was between the amount of calcium administered via post-dilution fluid and the overall calcium removal in the dialysis filter. Identical therapy settings do not uniformly affect calcium mobilization, and our approach offers insight into calcium distribution across body compartments. This understanding will enhance clinical prescription practices.

What does visual snow look like? Quantification by matching a simulation.

The primary symptom of visual snow syndrome (VSS) is the unremitting perception of small, flickering dots covering the visual field. VSS is a serious but poorly understood condition that can interfere with daily tasks. Several studies have provided qualitative data about the appearance of visual snow, but methods to quantify the symptom are lacking. Here, we developed a task in which participants with VSS adjusted parameters of simulated visual snow on a computer monitor until the simulation matched their internal visual snow. On each trial, participants (n = 31 with VSS) modified the size, density, update speed, and contrast of the simulation. Participants' settings were highly reliable across trials (intraclass correlation coefficients > 0.89), and they reported that the task was effective at stimulating their visual snow. On average, visual snow was very small (less than 2 arcmin in diameter), updated quickly (mean temporal frequency = 18.2 Hz), had low density (mean snow elements vs. background = 2.87%), and had low contrast (average root mean square contrast = 2.56%). Our task provided a quantitative assessment of visual snow percepts, which may help individuals with VSS communicate their experience to others, facilitate assessment of treatment efficacy, and further our understanding of the trajectory of symptoms, as well as the neural origins of VSS.

Technical Note: Swing golden angle - A navigator-interleaved golden angle trajectory with eddy current suppression - Application in free-running cardiac MRI.

Golden angle (GA) radial trajectory is advantageous for dynamic magnetic resonance imaging (MRI). Recently, several advanced algorithms have been developed based on navigator-interleaved GA trajectory to realize free-running cardiac MRI. However, navigator-interleaved GA trajectory suffers from the eddy-current effect, which reduces the image quality. This work aims to integrate the navigator-interleaved GA trajectory with clinical cardiac MRI acquisition, with the minimum eddy-current artifacts. The ultimate goal is to realize a high-quality free-running cardiac imaging technique. In this paper, we propose a new "swing golden angle" (swingGA) radial profile order. SwingGA samples the k-space by rotating back and forth at the generalized golden ratio interval, with smoothly interleaved navigator readouts. The sampling efficiency and angle increment distributions were investigated by numerical simulations. Static phantom imaging experiments were conducted to evaluate the eddy current effect, compared with cartesian, golden angle radial (GA), and tiny golden angle (tGA) trajectories. Furthermore, 12 heart-healthy subjects (aged 21-25years) were recruited for free-running cardiac imaging with different sampling trajectories. Dynamic images were reconstructed by a low-rank subspace-constrained algorithm. The image quality was evaluated by signal-to-noise-ratio and spectrum analysis in the heart region, and compared with traditional clinical cardiac MRI images. SwingGA pattern achieves the highest sampling efficiency (mSE>0.925) and the minimum azimuthal angle increment (mAD<1.05). SwingGA can effectively suppress eddy currents in static phantom images, with the lowest normalized root mean square error (nRMSE) values among radial trajectories. For the in-vivo cardiac images, swingGA enjoys the highest SNR both in the blood pool and myocardium, and contains the minimum level of high-frequency artifacts. The free-running cardiac images have good consistency with traditional clinical cardiac MRI, and the swingGA sampling pattern achieves the best image quality among all sampling patterns. The proposed swingGA sampling pattern can effectively improve the sampling efficiency and reduce the eddy currents for the navigator-interleaved GA sequence. SwingGA is a promising sampling pattern for free-running cardiac MRI.

Time-division multiplexing (TDM) sequence removes bias in T2 estimation and relaxation-diffusion measurements.

To compare the performance of multi-echo (ME) and time-division multiplexing (TDM) sequences for accelerated relaxation-diffusion MRI (rdMRI) acquisition and to examine their reliability in estimating accurate rdMRI microstructure measures. The ME, TDM, and the reference single-echo (SE) sequences with six echo times (TE) were implemented using Pulseq with single-band (SB-) and multi-band 2 (MB2-) acceleration factors. On a diffusion phantom, the image intensities of the three sequences were compared, and the differences were quantified using the normalized root mean squared error (NRMSE). For the in-vivo brain scan, besides the image intensity comparison and T2-estimates, different methods were used to assess sequence-related effects on microstructure estimation, including the relaxation diffusion imaging moment (REDIM) and the maximum-entropy relaxation diffusion distribution (MaxEnt-RDD). TDM performance was similar to the gold standard SE acquisition, whereas ME showed greater biases (3-4× larger NRMSEs for phantom, 2× for in-vivo). T2 values obtained from TDM closely matched SE, whereas ME sequences underestimated the T2 relaxation time. TDM provided similar diffusion and relaxation parameters as SE using REDIM, whereas SB-ME exhibited a 60% larger bias in the <R2> map and on average 3.5× larger bias in the covariance between relaxation-diffusion coefficients. Our analysis demonstrates that TDM provides a more accurate estimation of relaxation-diffusion measurements while accelerating the acquisitions by a factor of 2 to 3.

Attention-based Imputation of Missing Values in Electronic Health Records Tabular Data.

The imputation of missing values (IMV) in electronic health records tabular data is crucial to enable machine learning for patient-specific predictive modeling. While IMV methods are developed in biostatistics and recently in machine learning, deep learning-based solutions have shown limited success in learning tabular data. This paper proposes a novel attention-based missing value imputation framework that learns to reconstruct data with missing values leveraging between-feature (self-attention) or between-sample attentions. We adopt data manipulation methods used in contrastive learning to improve the generalization of the trained imputation model. The proposed self-attention imputation method outperforms state-of-the-art statistical and machine learning-based (decision-tree) imputation methods, reducing the normalized root mean squared error by 18.4% to 74.7% on five tabular data sets and 52.6% to 82.6% on two electronic health records data sets. The proposed attention-based missing value imputation method shows superior performance across a wide range of missingness (10% to 50%) when the values are missing completely at random.