Model Development Dataset Research Articles

Resampling and data augmentation for short-term PV output prediction based on an imbalanced sky images dataset using convolutional neural networks

Integrating photovoltaics (PV) into electricity grids is challenged by potentially large fluctuations in power generation. In recent years, sky image-based PV output prediction using convolutional neural networks (CNNs) has emerged as a promising approach to forecasting fluctuations. A key challenge is imbalanced sky image datasets: because of the geography of solar PV system installations, sky image datasets are often rich in sunny condition data but deficient in cloudy condition data. This imbalance contrasts with the fact that model errors are dominated by cloudy condition performance. In this study, we attempt to remedy this by exploring the enrichment and augmentation of an imbalanced sky images dataset for two PV output prediction tasks: nowcasting (predicting concurrent PV output) and forecasting (predicting 15-minute-ahead future PV output). We empirically examine the efficacy of using different resampling and data augmentation approaches to create a rebalanced dataset for model development. A three-stage greedy search is used to determine the optimal resampling approach, data augmentation techniques and over-sampling rate. The results show that for the nowcast problem, resampling and data augmentation can effectively enhance the model performance, reducing overall root mean squared error (RMSE) by an average of 4%, or a 15 std. (standard deviation) of improvement compared to the variability of the baseline model. In contrast, the treatment RMSE for the forecast problem nearly always overlaps the baseline performance at the ± 2 std. level. The optimal resampling approach expands on the original dataset by over-sampling the minority cloudy data, with the best results from large over-sampling rate (e.g., 4 ~ 6 times over-sampling of cloudy images).

Identifying anti-TNF response biomarkers in ulcerative colitis using a diffusion-based signalling model.

Resistance to anti-TNF therapy in subgroups of ulcerative colitis (UC) patients is a major challenge and incurs significant treatment costs. Identification of patients at risk of nonresponse to anti-TNF is of major clinical importance. To date, no quantitative computational framework exists to develop a complex biomarker for the prognosis of UC treatment. Modelling patient-wise receptor to transcription factor (TF) network connectivity may enable personalized treatment. We present an approach for quantitative diffusion analysis between receptors and TFs using gene expression data. Key TFs were identified using pandaR. Network connectivities between immune-specific receptor-TF pairs were quantified using network diffusion in UC patients and controls. The patient-specific network could be considered a complex biomarker that separates anti-TNF treatment-resistant and responder patients both in the gene expression dataset used for model development and separate independent test datasets. The model was further validated in rheumatoid arthritis where it successfully discriminated resistant and responder patients to tocilizumab treatment. Our model may contribute to prognostic biomarkers that may identify treatment-resistant and responder subpopulations of UC patients. Software is available at https://github.com/Amy3100/receptor2tfDiffusion. Supplementary data are available at Bioinformatics Advances online.

Ice-Crystal Icing Investigation on a Honeywell Uncertified Research Engine in an Altitude Simulation Icing Facility

AbstractA Honeywell Uncertified Research Engine was exposed to various ice-crystal conditions in the NASA Propulsion Systems Laboratory (PSL). Simulations using NASA's one-dimensional (1D) Icing Risk Analysis tool were used to determine potential inlet conditions that could lead to ice-crystal accretion along the inlet of the core flowpath and into the high-pressure compressor. Baseline conditions were established, and parameters were varied to observe accretion characteristics. Data were acquired at altitudes varying from 5 kft to 45 kft, at nominal ice particle median volumetric diameters from 20 µm to 100 µm, and total water contents of 1 g/m3 to 12 g/m3. Metal temperatures were acquired for the inlet guide vane and vane stators 1–2. In situ measurements of the particle size distribution were acquired upstream and downstream of the engine fan face in order to study particle breakup behavior. Cameras were installed in the engine to capture ice accretions at the leading edge of the fan stator, splitter lip, and inlet guide vane. The goal of this study was to understand the key parameters of accretion, acquire particle breakup data aft of the fan, and generate a unique icing dataset for model development. Significant particle breakup downstream of the fan in the bypass was observed. The metal temperatures on the inlet guide vanes (IGVs) and stators show a temperature increase with increasing particle size. Accretion behavior at the fan stator and splitter lip across was very similar. However, accretion decreased with increasing particle size at the IGVs.

Segmentation of the Prostate Transition Zone and Peripheral Zone on MR Images with Deep Learning.

Purpose To develop a deep learning model to delineate the transition zone (TZ) and peripheral zone (PZ) of the prostate on MR images. Materials and Methods This retrospective study was composed of patients who underwent a multiparametric prostate MRI and an MRI/transrectal US fusion biopsy between January 2013 and May 2016. A board-certified abdominal radiologist manually segmented the prostate, TZ, and PZ on the entire data set. Included accessions were split into 60% training, 20% validation, and 20% test data sets for model development. Three convolutional neural networks with a U-Net architecture were trained for automatic recognition of the prostate organ, TZ, and PZ. Model performance for segmentation was assessed using Dice scores and Pearson correlation coefficients. Results A total of 242 patients were included (242 MR images; 6292 total images). Models for prostate organ segmentation, TZ segmentation, and PZ segmentation were trained and validated. Using the test data set, for prostate organ segmentation, the mean Dice score was 0.940 (interquartile range, 0.930-0.961), and the Pearson correlation coefficient for volume was 0.981 (95% CI: 0.966, 0.989). For TZ segmentation, the mean Dice score was 0.910 (interquartile range, 0.894-0.938), and the Pearson correlation coefficient for volume was 0.992 (95% CI: 0.985, 0.995). For PZ segmentation, the mean Dice score was 0.774 (interquartile range, 0.727-0.832), and the Pearson correlation coefficient for volume was 0.927 (95% CI: 0.870, 0.957). Conclusion Deep learning with an architecture composed of three U-Nets can accurately segment the prostate, TZ, and PZ. Keywords: MRI, Genital/Reproductive, Prostate, Neural Networks Supplemental material is available for this article. © RSNA, 2021.

A Comprehensive Techno-eco-assessment of CO2 Enhanced Oil Recovery Projects Using a Machine-learning Assisted Workflow

Injection of CO2 gas to partially depleted oil reservoirs not only enhances hydrocarbon production but also take advances of the underground porous media for carbon storage purposes. Therefore, carbon utilization to enhanced oil recovery (EOR) projects become more and more attractive in the petroleum industry. Notably, the use of anthropogenic CO2 in EOR project could be challenging due to the capital and operational costs of carbon capture, transportation, and recycling. In the meantime, the US government offers encouraging tax policies, Section 45 Q, to oil field operators who inject a large volume of CO2 to produce oil. Thus, the project economics of the CO2-EOR project could be even more complex, especially when the oil price is low at this time. In this article, a robust machine-learning assisted CO2-EOR project optimization and design protocol is presented. The workflow aims at simultaneously optimizing the hydrocarbon recovery, carbon storage volume and project economics using a Pareto-front-based multiple-objective optimization (MOO) scheme. A solution repository containing various optimized development strategies will be structured. Moreover, the proposed workflow will investigate the project economics of the solutions by systematically considering impactive factors such as tax credits, project capital costs, oil prices, etc., and provide comprehensive recommendations to the field operators for decision-making purposes. This work employs data collected from the Farnsworth unit (FWU) in west Texas, US, as a field case to demonstrate the workflow. FWU is characterized as partially depleted oil sands undergoing water-alternative-CO2 (CO2-WAG) injection processes. A compositional numerical reservoir simulation model is established to investigate the fluid transportations dynamic of FWU. The numerical model is validated via a rigorous history-matching study using 55-year of primary and secondary (water flooding) recovery data, and 8-year of ternary recovery (CO2-WAG) data. It is worth emphasizing that the history-matched model successfully incorporates eight relative permeability curves to various spatial regions of the field. Such relative permeability curves are measured via laboratory investigations using representative core samples collected from the fields (hydraulic flow units). The history matched numerical model can be used to forecast the hydrocarbon production and CO2 storage volume using different CO2-WAG designs. In order to obtain finer optimization solutions, the field operational parameters, such as water injection rate, gas injection period, water injection periods, production well specifications, etc. of each individual well are considered as control parameters. In this way, the field operators can get detailed guidance to operate each individual well. However, the imposing of MOO requires computationally intensive procedures by totally relying on the high-fidelity numerical simulator, which engages the motivation to employ machine-learning-based proxies in the workflow. In this work, supported vector machine regression (SVR) combined with the Gaussian kernel is utilized to mimic the high-fidelity numerical model. The hyper-parameters of the SVR are optimized using Bayesian Optimization to achieve a better generalization performance. The SVR couples with Multi-objective Particle Swarm Optimization (MOPSO) protocol to structure the Pareto-front solution repository. To further eliminate the uncertainties introduced by the proxy error margins, a self-adapting scheme is integrated into the workflow to justify the Pareto-front solutions. This scheme will automatically add new sampling data to the training dataset for surrogate model development, making the optimization results more correct. Furthermore, this paper will carry comprehensive techno-eco-assessments of the optimized solutions considering the vital economic factors including the carbon capture and transporting costs, field operational cost, tax credits, oil price, etc. With the help of the proxy models, fast uncertainty analysis can be conducted to obtain stochastic results of the project net present value, rate of return and payback period. Field operators can make their decisions to design CO2-WAG projects based on both technical and economical perspectives.

Predictive ability of four small‐scale quality tests for dough rheological properties and baking quality in hard red spring wheat

AbstractBackground and ObjectivesPredictive models for use in early generation hard spring wheat breeding were evaluated for Farinograph and baking quality. Flour samples were prepared on the Quadrumat Jr. and Buhler mill from 48 genotypes grown at four locations in North Dakota.FindingsModels developed from 13 predictor variables from GlutoPeak, Glutomatic, Near Infrared Reflectance (NIR), and Solvent Retention Capacity (SRC) predicted Farinograph water absorption (ABS) and baking absorption (BA) with cross‐validated prediction accuracies of 0.71 and 0.58, respectively. For Farinograph peak time (PT), Farinograph Stability (STAB), mixing time (MT), and loaf volume (LV), the prediction accuracies were high in the model development dataset (r2 ranging from 0.60 to 0.76). However, in the cross‐validation set, the prediction accuracies only ranged from 0.30 to 0.48.ConclusionsGlutoPeak parameters can be used to predict ABS and BA in early breeding stages. Prediction of stability and peak time showed potential, and models should be improved to better predict mixing time and loaf volume.Significance and noveltyResults from a multi‐location trial involving a large number of genotypes are needed to validate the prediction accuracy for end‐use traits in hard red spring wheat. Comparing results from different mill types and diverse genotypes ensures broad applicability.

Abstract P01: Development of a TMB classifier on NSCLC H&E images

Abstract A number of immunotherapy treatments in cancer can be guided by a patient’s tumor mutation burden (TMB) status. However typical TMB determination is performed via next generation sequencing which is problematic due to its high cost, low availability worldwide and operational challenges. The ability to screen patients for TMB determination using alternative methodologies is of interest, particularly in the use of H&E slides which are readily available for most cancer patients. Bio-AI Health has developed a machine learning (ML) classifier model capable of identifying TMB status in patients using H&E slides only. Our automated model is unique in that it uses an enriched signal from samples consisting of tumor region only while generating unique visualizers that provide an insight into the performance and mode of action of the model itself. This visualizer allows the pathologist a unique view into the ML performance therefore eliminating the black box concept and providing a crucial QC step to ensure proper algorithm performance. This test can increase the sensitivity of detection providing significant clinical utility. The dataset for model development was obtained from TCGA cohort and TMB status was calculated using whole exome sequencing data. We further developed automated ML segmentation models to accurately segment the tissue into multiple compartments and filter our all necrotic and non-neoplastic tissue regions. Taking into account only tumor regions, the classifier ML model uses an ensemble approach which integrates the prediction of multiple convolutional network probabilities and achieves an accuracy of 0.76 with a sensitivity of .79 and F1 score of .82 towards the classification of the tissue as TMB high or Low. Citation Format: Alan Jerusalmi. Development of a TMB classifier on NSCLC H&E images [abstract]. In: Proceedings of the AACR Virtual Meeting: COVID-19 and Cancer; 2021 Feb 3-5. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(6_Suppl):Abstract nr P01.

Proposing a machine-learning based method to predict stillbirth before and during delivery and ranking the features: nationwide retrospective cross-sectional study

BackgroundStillbirth is defined as fetal loss in pregnancy beyond 28 weeks by WHO. In this study, a machine-learning based method is proposed to predict stillbirth from livebirth and discriminate stillbirth before and during delivery and rank the features.MethodA two-step stack ensemble classifier is proposed for classifying the instances into stillbirth and livebirth at the first step and then, classifying stillbirth before delivery from stillbirth during the labor at the second step. The proposed SE has two consecutive layers including the same classifiers. The base classifiers in each layer are decision tree, Gradient boosting classifier, logistics regression, random forest and support vector machines which are trained independently and aggregated based on Vote boosting method. Moreover, a new feature ranking method is proposed in this study based on mean decrease accuracy, Gini Index and model coefficients to find high-ranked features.ResultsIMAN registry dataset is used in this study considering all births at or beyond 28th gestational week from 2016/04/01 to 2017/01/01 including 1,415,623 live birth and 5502 stillbirth cases. A combination of maternal demographic features, clinical history, fetal properties, delivery descriptors, environmental features, healthcare service provider descriptors and socio-demographic features are considered. The experimental results show that our proposed SE outperforms the compared classifiers with the average accuracy of 90%, sensitivity of 91%, specificity of 88%. The discrimination of the proposed SE is assessed and the average AUC of ±95%, CI of 90.51% ±1.08 and 90% ±1.12 is obtained on training dataset for model development and test dataset for external validation, respectively. The proposed SE is calibrated using isotopic nonparametric calibration method with the score of 0.07. The process is repeated 10,000 times and AUC of SE classifiers using random different training datasets as null distribution. The obtained p-value to assess the specificity of the proposed SE is 0.0126 which shows the significance of the proposed SE.ConclusionsGestational age and fetal height are two most important features for discriminating livebirth from stillbirth. Moreover, hospital, province, delivery main cause, perinatal abnormality, miscarriage number and maternal age are the most important features for classifying stillbirth before and during delivery.

Landscape Similarity Analysis Using Texture Encoded Deep-Learning Features on Unclassified Remote Sensing Imagery

Convolutional neural networks (CNNs) are known for their ability to learn shape and texture descriptors useful for object detection, pattern recognition, and classification problems. Deeper layer filters of CNN generally learn global image information vital for whole-scene or object discrimination. In landscape pattern comparison, however, dense localized information encoded in shallow layers can contain discriminative information for characterizing changes across image local regions but are often lost in the deeper and non-spatial fully connected layers. Such localized features hold potential for identifying, as well as characterizing, process–pattern change across space and time. In this paper, we propose a simple yet effective texture-based CNN (Tex-CNN) via a feature concatenation framework which results in capturing and learning texture descriptors. The traditional CNN architecture was adopted as a baseline for assessing the performance of Tex-CNN. We utilized 75% and 25% of the image data for model training and validation, respectively. To test the models’ generalization, we used a separate set of imagery from the Aerial Imagery Dataset (AID) and Sentinel-2 for model development and independent validation. The classical CNN and the Tex-CNN classification accuracies in the AID were 91.67% and 96.33%, respectively. Tex-CNN accuracy was either on par with or outcompeted state-of-the-art methods. Independent validation on Sentinel-2 data had good performance for most scene types but had difficulty discriminating farm scenes, likely due to geometric generalization of discriminative features at the coarser scale. In both datasets, the Tex-CNN outperformed the classical CNN architecture. Using the Tex-CNN, gradient-based spatial attention maps (feature maps) which contain discriminative pattern information are extracted and subsequently employed for mapping landscape similarity. To enhance the discriminative capacity of the feature maps, we further perform spatial filtering, using PCA and select eigen maps with the top eigen values. We show that CNN feature maps provide descriptors capable of characterizing and quantifying landscape (dis)similarity. Using the feature maps histogram of oriented gradient vectors and computing their Earth Movers Distances, our method effectively identified similar landscape types with over 60% of target-reference scene comparisons showing smaller Earth Movers Distance (EMD) (e.g., 0.01), while different landscape types tended to show large EMD (e.g., 0.05) in the benchmark AID. We hope this proposal will inspire further research into the use of CNN layer feature maps in landscape similarity assessment, as well as in change detection.

A Comparison of Turbine Mass Flow Models Based on Pragmatic Identification Data Sets for Turbogenerator Model Development

The use of turbogenerators as a means of waste heat recovery has gained interest from industry and academia in recent years. Accurate mass flow models of turbogenerators are required for assessing their impact on overall engine performance during design–development. This paper presents the results of an experimental model identification program for a commercially available turbogenerator. The experimental data was categorised into training and validation data sets. Training data sets were selected using a simulation based method to bound data within a region representative of turbocharger turbine operation. A comprehensive review of promulgated models is presented, and the replication and extrapolation performance with respect to the experimental data sets is assessed. A new family of models is proposed which is applicable to a large class of radial flow turbines. Application of a systematic model selection process based on Akaike Information Criteria yields models from this family with improved performance. Furthermore, the robustness of the proposed family of models is assessed using published experimental data sets from a range of turbine designs, demonstrating the versatility of the proposed model family and model selection techniques.

Machine Learning Algorithms to Predict Recurrence within 10 Years after Breast Cancer Surgery: A Prospective Cohort Study.

Simple SummaryNo studies have discussed machine learning algorithms to predict recurrence within 10 years after breast cancer surgery. Artificial neural networks (ANN) model is superior to the other forecasting models in terms of accuracy in predicting recurrence within 10 years after breast cancer surgery. Surgeon volume was the best predictor of recurrence within 10 years after breast cancer surgery, followed by hospital volume and tumor stage. For patients who are candidates for breast cancer surgery or who have already received breast cancer surgery, these important predictors can also be used for education in the expected course of recovery and health outcomes. Integration of the machine learning algorithms applied in this study in other clinical decision-making tools would provide additional data that can be used to improve accuracy in predicting recurrence.No studies have discussed machine learning algorithms to predict recurrence within 10 years after breast cancer surgery. This study purposed to compare the accuracy of forecasting models to predict recurrence within 10 years after breast cancer surgery and to identify significant predictors of recurrence. Registry data for breast cancer surgery patients were allocated to a training dataset (n = 798) for model development, a testing dataset (n = 171) for internal validation, and a validating dataset (n = 171) for external validation. Global sensitivity analysis was then performed to evaluate the significance of the selected predictors. Demographic characteristics, clinical characteristics, quality of care, and preoperative quality of life were significantly associated with recurrence within 10 years after breast cancer surgery (p < 0.05). Artificial neural networks had the highest prediction performance indices. Additionally, the surgeon volume was the best predictor of recurrence within 10 years after breast cancer surgery, followed by hospital volume and tumor stage. Accurate recurrence within 10 years prediction by machine learning algorithms may improve precision in managing patients after breast cancer surgery and improve understanding of risk factors for recurrence within 10 years after breast cancer surgery.

Penalization and shrinkage methods produced unreliable clinical prediction models especially when sample size was small

ObjectivesWhen developing a clinical prediction model, penalization techniques are recommended to address overfitting, as they shrink predictor effect estimates toward the null and reduce mean-square prediction error in new individuals. However, shrinkage and penalty terms (‘tuning parameters’) are estimated with uncertainty from the development data set. We examined the magnitude of this uncertainty and the subsequent impact on prediction model performance. Study Design and SettingThis study comprises applied examples and a simulation study of the following methods: uniform shrinkage (estimated via a closed-form solution or bootstrapping), ridge regression, the lasso, and elastic net. ResultsIn a particular model development data set, penalization methods can be unreliable because tuning parameters are estimated with large uncertainty. This is of most concern when development data sets have a small effective sample size and the model's Cox-Snell R2 is low. The problem can lead to considerable miscalibration of model predictions in new individuals. ConclusionPenalization methods are not a ‘carte blanche’; they do not guarantee a reliable prediction model is developed. They are more unreliable when needed most (i.e., when overfitting may be large). We recommend they are best applied with large effective sample sizes, as identified from recent sample size calculations that aim to minimize the potential for model overfitting and precisely estimate key parameters.

Mortality prognostication scores do not predict long-term, health-related quality of life after burn: A burn model system national database study

ObjectiveDespite improved mortality rates after burn injury, many patients face significant long-term physical and psychosocial disabilities. We aimed to determine whether commonly used mortality prognostication scores predict long-term, health-related quality of life after burn injury. By doing so, we might add evidence to support goals of care discussions and facilitate shared decision-making efforts in the hours and days after a life-changing injury. MethodsWe used the multicenter National Institute of Disability, Independent Living and Rehabilitation Research Burn Model System database (1994–2019) to analyze SF-12 physical (PCS) and mental component (MCS) scores among survivors one year after major burn injury. Ninety percent of the observations were randomly assigned to a model development dataset. Multilevel, mixed-effects, linear regression models determined the relationship between revised Baux and Ryan Scores and SF-12 measures. Additionally, we tested a model with disaggregated independent and other covariates easily obtained around the time of index admission: age, sex, race, burn size, inhalation injury. Residuals from the remaining 10% of observations in the validation dataset were examined. ResultsThe analysis included 1606 respondents (median age 42 years, IQR 28–53 years; 70% male). Median burn size was 16% TBSA (IQR 6–30) and 13% of respondents sustained inhalation injury. Higher revised Baux and Ryan Scores and age, burn size, and inhalation injury were significantly correlated with lower PCS, but were not correlated with MCS. Female sex, black race, burn size, and inhalation injury correlated with lower MCS. All models poorly explained the variance in SF-12 scores (adjusted r2 0.01–0.12). ConclusionHigher revised Baux and Ryan Scores negatively correlated with long-term physical health, but not mental health, after burn injury. Regardless, the models poorly explained the variance in SF-12 scores one year after injury. More accurate models are needed to predict long-term, health-related quality of life and support shared decision-making during acute burn care.

Comparison of the Tree-Based Machine Learning Algorithms to Cox Regression in Predicting the Survival of Oral and Pharyngeal Cancers: Analyses Based on SEER Database

Simple SummaryFormulating accurate survival prediction models of oral and pharyngeal cancers (OPCs) is important, as they might impact the decisions of clinicians and patients. Improving the quality of these clinical prediction modelling studies can benefit the reliability of the developed models and facilitate their implementations in clinical practice. Given the growing trend on the application of machine learning methods in cancer research, we present the use of popular tree-based machine learning algorithms and compare them to the standard Cox regression as an aim to predict OPCs survival. The predictive models discussed here are based on a large cancer registry dataset incorporating various prognosis factors and different forms of bias. The comparable predictive performance between Cox and tree-based models suggested that these machine learning algorithms provide non-parametric alternatives to Cox regression and are of clinical use for estimating the survival probability of OPCs patients.This study aims to demonstrate the use of the tree-based machine learning algorithms to predict the 3- and 5-year disease-specific survival of oral and pharyngeal cancers (OPCs) and compare their performance with the traditional Cox regression. A total of 21,154 individuals diagnosed with OPCs between 2004 and 2009 were obtained from the Surveillance, Epidemiology, and End Results (SEER) database. Three tree-based machine learning algorithms (survival tree (ST), random forest (RF) and conditional inference forest (CF)), together with a reference technique (Cox proportional hazard models (Cox)), were used to develop the survival prediction models. To handle the missing values in predictors, we applied the substantive model compatible version of the fully conditional specification imputation approach to the Cox model, whereas we used RF to impute missing data for the ST, RF and CF models. For internal validation, we used 10-fold cross-validation with 50 iterations in the model development datasets. Following this, model performance was evaluated using the C-index, integrated Brier score (IBS) and calibration curves in the test datasets. For predicting the 3-year survival of OPCs with the complete cases, the C-index in the development sets were 0.77 (0.77, 0.77), 0.70 (0.70, 0.70), 0.83 (0.83, 0.84) and 0.83 (0.83, 0.86) for Cox, ST, RF and CF, respectively. Similar results were observed in the 5-year survival prediction models, with C-index for Cox, ST, RF and CF being 0.76 (0.76, 0.76), 0.69 (0.69, 0.70), 0.83 (0.83, 0.83) and 0.85 (0.84, 0.86), respectively, in development datasets. The prediction error curves based on IBS showed a similar pattern for these models. The predictive performance remained unchanged in the analyses with imputed data. Additionally, a free web-based calculator was developed for potential clinical use. In conclusion, compared to Cox regression, ST had a lower and RF and CF had a higher predictive accuracy in predicting the 3- and 5-year OPCs survival using SEER data. The RF and CF algorithms provide non-parametric alternatives to Cox regression to be of clinical use for estimating the survival probability of OPCs patients.

1042P Objective response rate as an intermediate surrogate endpoint to predict overall survival at 12 months

Immune checkpoint inhibitors (ICI) have a unique mechanism of action that differs from chemotherapy. Despite the poor correlation between ORR and overall survival (OS), 12-month OS (OS12) ratio is a good intermediate surrogate for OS at trial level. We aimed to develop and validate a model using ORR as a predictor for OS12 in ICI trials. We performed an electronic search to identify eligible studies. Randomized controlled trials (RCTs) reported between January 2020-June 2019 formed the model development dataset. Correlation between ORR and OS12 were examined both at trial and ICI arm levels. Using the ICI arm data, a linear regression model was fitted using the ordinary least-squares approach, adjusting for tumour types, with ORR as a predictor for OS12. We then extracted data from an independent dataset of RCTs and single-arm ICI studies, reported from June 2019 to Jan 2020, to validate the performance of our prediction model. We identified 63 eligible trials, with 78 ICI treatment arms (58 RCTs, 20 single-arm). Thirteen trials were doublet ICI studies. The tumours types were predominantly non-small cell lung cancer (30%) and melanoma (13%). Other main tumours types included gastric/oesophageal (10%), renal cell (9%), head and neck squamous cell (8%), urothelial carcinoma (8%) and others (17%). Fifty ICI arms (from 43 RCTs) formed the model development dataset. The validation dataset comprised of 28 ICI arms (8 RCTs, 20 single arm studies). At arm level, the correlation between ORR and OS12 within ICI arms was strong, r=0.80. Complete response rate correlates poorly with OS12 (r= 0.50). The ORR-OS12 prediction model was OS12 = (1.012324 x ORR) + 0.3006825 + (0 x melanoma) + (0.0021079 x NSCLC) – (0.038521 x other tumors). When this model was applied in the validation dataset, the fitted line of scatterplot of ORR-predicted OS12 revealed close calibration with the observed OS12, r= 0.63. At trial level, the correlation between ORR risk ratio and OS HR was poor, r=0.58. Our prediction model using ORR can be used to predict for OS12 in early phase ICI trials, regardless of tumour type, overcoming the issue of poor correlation between ORR and OS. This may help support decisions for phase III testing based on ORR in phase II trials of ICI.

Diagnostic Model for In-Hospital Bleeding in Patients with Acute ST-Segment Elevation Myocardial Infarction: Algorithm Development and Validation.

BackgroundBleeding complications in patients with acute ST-segment elevation myocardial infarction (STEMI) have been associated with increased risk of subsequent adverse consequences.ObjectiveThe objective of our study was to develop and externally validate a diagnostic model of in-hospital bleeding.MethodsWe performed multivariate logistic regression of a cohort for hospitalized patients with acute STEMI in the emergency department of a university hospital. Participants: The model development data set was obtained from 4262 hospitalized patients with acute STEMI from January 2002 to December 2013. A set of 6015 hospitalized patients with acute STEMI from January 2014 to August 2019 were used for external validation. We used logistic regression analysis to analyze the risk factors of in-hospital bleeding in the development data set. We developed a diagnostic model of in-hospital bleeding and constructed a nomogram. We assessed the predictive performance of the diagnostic model in the validation data sets by examining measures of discrimination, calibration, and decision curve analysis (DCA).ResultsIn-hospital bleeding occurred in 112 of 4262 participants (2.6%) in the development data set. The strongest predictors of in-hospital bleeding were advanced age and high Killip classification. Logistic regression analysis showed differences between the groups with and without in-hospital bleeding in age (odds ratio [OR] 1.047, 95% CI 1.029-1.066; P<.001), Killip III (OR 3.265, 95% CI 2.008-5.31; P<.001), and Killip IV (OR 5.133, 95% CI 3.196-8.242; P<.001). We developed a diagnostic model of in-hospital bleeding. The area under the receiver operating characteristic curve (AUC) was 0.777 (SD 0.021, 95% CI 0.73576-0.81823). We constructed a nomogram based on age and Killip classification. In-hospital bleeding occurred in 117 of 6015 participants (1.9%) in the validation data set. The AUC was 0.7234 (SD 0.0252, 95% CI 0.67392-0.77289).ConclusionsWe developed and externally validated a diagnostic model of in-hospital bleeding in patients with acute STEMI. The discrimination, calibration, and DCA of the model were found to be satisfactory.Trial RegistrationChiCTR.org ChiCTR1900027578; http://www.chictr.org.cn/showprojen.aspx?proj=45926

Detecting shielded explosives by coupling prompt gamma neutron activation analysis and deep neural networks

Prompt Gamma Neutron Activation Analysis is a nuclear-based technique that can be used in explosives detection. It relies on bombarding unknown samples with neutrons emitted from a neutron source. These neutrons interact with the sample nuclei emitting the gamma spectrum with peaks at specific energies, which are considered a fingerprint for the sample composition. Analyzing these peaks heights will give information about the unknown sample material composition. Shielding the sample from gamma rays or neutrons will affect the gamma spectrum obtained to be analyzed, providing a false indication about the sample constituents, especially when the shield is unknown. Here we show how using deep neural networks can solve the shielding drawback associated with the prompt gamma neutron activation analysis technique in explosives detection. We found that the introduced end-to-end framework was capable of differentiating between explosive and non-explosive hydrocarbons with accuracy of 95% for the previously included explosives in the model development data set. It was also, capable of generalizing with accuracy 80% over the explosives which were not included in the model development data set. Our results show that coupling prompt gamma neutron activation analysis with deep neural networks has a good potential for high accuracy explosives detection regardless of the shield presence.

A Plastic Contamination Image Dataset for Deep Learning Model Development and Training

The removal of plastic contamination in cotton lint is an issue of top priority for the U.S. cotton industry. One of the main sources of plastic contamination appearing in marketable cotton bales is plastic used to wrap cotton modules on cotton harvesters. To help mitigate plastic contamination at the gin, automatic inspection systems are needed to detect and control removal systems. Due to significant cost constraints in the U.S. cotton ginning industry, the use of low-cost color cameras for detection of plastic contamination has been successfully adopted. However, some plastics of similar color to background are difficult to detect when utilizing traditional machine learning algorithms. Hence, current detection/removal system designs are not able to remove all plastics and there is still a need for better detection methods. Recent advances in deep learning convolutional neural networks (CNNs) show promise for enabling the use of low-cost color cameras for detection of objects of interest when placed against a background of similar color. They do this by mimicking the human visual detection system, focusing on differences in texture rather than color as the primary detection paradigm. The key to leveraging the CNNs is the development of extensive image datasets required for training. One of the impediments to this methodology is the need for large image datasets where each image must be annotated with bounding boxes that surround each object of interest. As this requirement is labor-intensive, there is significant value in these image datasets. This report details the included image dataset as well as the system design used to collect the images. For acquisition of the image dataset, a prototype detection system was developed and deployed into a commercial cotton gin where images were collected for the duration of the 2018–2019 ginning season. A discussion of the observational impact that the system had on reduction of plastic contamination at the commercial gin, utilizing traditional color-based machine learning algorithms, is also included.

Electrical Energy Demand Forecasting Model Development and Evaluation with Maximum Overlap Discrete Wavelet Transform-Online Sequential Extreme Learning Machines Algorithms

To support regional electricity markets, accurate and reliable energy demand (G) forecast models are vital stratagems for stakeholders in this sector. An online sequential extreme learning machine (OS-ELM) model integrated with a maximum overlap discrete wavelet transform (MODWT) algorithm was developed using daily G data obtained from three regional campuses (i.e., Toowoomba, Ipswich, and Springfield) at the University of Southern Queensland, Australia. In training the objective and benchmark models, the partial autocorrelation function (PACF) was first employed to select the most significant lagged input variables that captured historical fluctuations in the G time-series data. To address the challenges of non-stationarities associated with the model development datasets, a MODWT technique was adopted to decompose the potential model inputs into their wavelet and scaling coefficients before executing the OS-ELM model. The MODWT-PACF-OS-ELM (MPOE) performance was tested and compared with the non-wavelet equivalent based on the PACF-OS-ELM (POE) model using a range of statistical metrics, including, but not limited to, the mean absolute percentage error (MAPE%). For all of the three datasets, a significantly greater accuracy was achieved with the MPOE model relative to the POE model resulting in an MAPE = 4.31% vs. MAPE = 11.31%, respectively, for the case of the Toowoomba dataset, and a similarly high performance for the other two campuses. Therefore, considering the high efficacy of the proposed methodology, the study claims that the OS-ELM model performance can be improved quite significantly by integrating the model with the MODWT algorithm.

A probability model to detect carbonate cementation in sandstones and other enigmatic wireline facies

Zones of carbonate cementation identified within lower Paaratte Formation sandstones of the eastern Otway Basin, Victoria, southeastern Australia, can be quantitatively detected by a wireline log-based probability model. Though trained on these zones, the same model appears to accurately predict carbonate cementation within Late Cretaceous-to-Eocene reservoir sandstones of the Latrobe Group supersequence of the Gippsland Basin, 100s of km to the east. Predicted carbonate cementation matches published evidence (a regional petrographic study) and provides a plausible interpretation for corresponding sections of Formation Micro-Imager data acquired more recently. These Latrobe Group sandstones are thought to have once been pervasively cemented prior to development of the secondary porosity responsible for providing reservoirs to the main petroleum system of the basin. However, model predictions imply that discrete, heavily cemented zones remain, which must have acted, and must still act, as local obstructions to reservoir fluid migration. These zones would also likely react with carbonic acid formed at plume fronts generated by dedicated CO2 storage operations of the future. The cemented zones that have been predicted are sparse, spatially sporadic and cannot be resolved by 3D seismic reflection survey data at present-day reservoir depths. These predictions therefore emphasise the difficulty in mapping the distribution of carbonate cemented zones. However, they provide calibration data for future mapping systems.Two different tests of the probability log model were undertaken. One benchmarks its predictions against those of a neural network trained using the original model development dataset from the Otway Basin. Predictive performance of the neural network model is significantly worse than that of the probability log model when making predictions using wireline data acquired at a well in the Gippsland Basin. The second test demonstrates that the general probability log modelling approach is amenable to classifying other cryptic ‘electro-facies’, for example those representing enigmatic or ambiguous rock types not easily interpreted from common wireline log data using standard theory. This test presents a case study by which the probability log modelling approach predicts the presence of both basic and evolved, felsic igneous sill intrusions within the stratigraphic succession of the Faroe-Shetland Basin of NW Europe.