Grid Search Research Articles

Convolutional LSTM model for cine image prediction of abdominal motion.

Objective.In this study, we tackle the challenge of latency in magnetic resonance linear accelerator (MR-Linac) systems, which compromises target coverage accuracy in gated real-time radiotherapy. Our focus is on enhancing motion prediction precision in abdominal organs to address this issue. We developed a convolutional long short-term memory (convLSTM) model, utilizing 2D cine magnetic resonance (cine-MR) imaging for this purpose.Approach.Our model, featuring a sequence-to-one architecture with six input frames and one output frame, employs structural similarity index measure (SSIM) as loss function. Data was gathered from 17 cine-MRI datasets using the Philips Ingenia MR-sim system and an Elekta Unity MR-Linac equivalent sequence, focusing on regions of interest (ROIs) like the stomach, liver, pancreas, and kidney. The datasets varied in duration from 1 to 10 min.Main results.The study comprised three main phases: hyperparameter optimization, individual training, and transfer learning with or without fine-tuning. Hyperparameters were initially optimized to construct the most effective model. Then, the model was individually applied to each dataset to predict images four frames ahead (1.24-3.28 s). We evaluated the model's performance using metrics such as SSIM, normalized mean square error, normalized correlation coefficient, and peak signal-to-noise ratio, specifically for ROIs with target motion. The average SSIM values achieved were 0.54, 0.64, 0.77, and 0.66 for the stomach, liver, kidney, and pancreas, respectively. In the transfer learning phase with fine-tuning, the model showed improved SSIM values of 0.69 for the liver and 0.78 for the kidney, compared to 0.64 and 0.37 without fine-tuning.Significance. The study's significant contribution is demonstrating the convLSTM model's ability to accurately predict motion for multiple abdominal organs using a Unity-equivalent MR sequence. This advancement is key in mitigating latency issues in MR-Linac radiotherapy, potentially improving the precision and effectiveness of real-time treatment for abdominal cancers.

Optimization of 2024-T3 Aluminum Alloy Friction Stir Welding Using Random Forest, XGBoost, and MLP Machine Learning Techniques.

This study optimized friction stir welding (FSW) parameters for 1.6 mm thick 2024T3 aluminum alloy sheets. A 3 × 3 factorial design was employed to explore tool rotation speeds (1100 to 1300 rpm) and welding speeds (140 to 180 mm/min). Static tensile tests revealed the joints' maximum strength at 87% relative to the base material. Hyperparameter optimization was conducted for machine learning (ML) models, including random forest and XGBoost, and multilayer perceptron artificial neural network (MLP-ANN) models, using grid search. Welding parameter optimization and extrapolation were then carried out, with final strength predictions analyzed using response surface methodology (RSM). The ML models achieved over 98% accuracy in parameter regression, demonstrating significant effectiveness in FSW process enhancement. Experimentally validated, optimized parameters resulted in an FSW joint efficiency of 93% relative to the base material. This outcome highlights the critical role of advanced analytical techniques in improving welding quality and efficiency.

Abstract 4915: Identification and validation of novel prognostic genetic markers in HER2-negative advanced gastric cancer (AGC) by artificial intelligence (AI) deep learning and machine learning algorithm

Abstract Introduction: Despite advances in anticancer treatments, there has been urgent need to identify novel prognostic markers for HER2-negative AGC. We propose AI based deep learning and machine learning pipeline to explore genetic markers that predict patient response to chemotherapy in HER2-negative AGC. Methods: We retrospectively assessed 179 stage IV HER2-negative AGC patients who have been treated with first-line chemotherapy in Yonsei Cancer Center, Korea between 2015 and 2021. In-house targeted sequencing panel data (CancerSCANTM and CancerMaster) were used to identify candidate prognostic markers. DeepSurv was mainly used to analyze the progression-free survival (PFS), which is a deep learning algorithm that can investigate prognostic roles. The SHapley Additive exPlanations (SHAP) method was applied to open the deep learning models’ black-box and calculate the importance ranking of gene features. Machine learning models-random survival forest, elastic-net, and lasso-were also investigated to identify potentially meaningful gene variants. To rank the gene features, we calculated variable importance (VIMP) for random survival forest, and penalized regression coefficients for elastic-net and lasso. All deep learning and machine learning models were trained using 5-folds cross-validation in 1,000 iterations of re-sampling bootstrapping with hyperparameter tuning by grid search. To classify responder or non-responder related genetic markers to chemotherapy, we used the signs of averaged penalize coefficients obtained from elastic-net and lasso. Based on the median linear predictor with candidate gene features, patients were classified into high-risk and low-risk group. Results: A total of 7,824 common variants were analyzed and 20 prognostic genetic markers were identified based on the top average values of SHAP, VIMP, and penalized regression coefficients. The top 10 non-responder genetic markers are included as follows: PARP4(p.S873N), CHUK(p.V268I), BARD1(359_365del), CREBBP(p.Y1125F), BARD1(p.P24S), PHLPP2(p.R1312Q), FAT3(p.Q3375R), CASP5(p.T106A), PHLPP2(p.I544V), and PTCH2(p.T988M). Otherwise, the top 10 responder genetic markers are defined as follows: IL7R(p.I66T), NOTCH4(L16delinsLL), FGFR3, BCL2A1(p.G82D), ZNF217(p.T548I), BRCA1(p.K1183R), FANCA(p.A412V), ADGRA2, CDKN2B, and FANCM(p.S175F). The high-risk group had worse PFS compared with low-risk group (median PFS 4.1 vs. 9.1 months; P-value <0.0001). Conclusion: This study showed the potential of AI deep learning and machine learning pipeline that employs an innovate prognostic genetic markers for HER2-negative AGC. Current ongoing analyses of larger cohort with more comprehensive clinical data will be presented. Citation Format: Sejung Park, Seok-Jae Heo, Choong-Kun Lee, Yaeji Lee, Woo Sun Kwon, Jingmin Che, Minseok Kim, Hyun Cheol Chung, Sun Young Rha. Identification and validation of novel prognostic genetic markers in HER2-negative advanced gastric cancer (AGC) by artificial intelligence (AI) deep learning and machine learning algorithm [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4915.

Optimization of computational intelligence approach for the prediction of glutinous rice dehydration.

Five computational intelligence approaches, namely Gaussian process regression (GPR), artificial neural network (ANN), decision tree (DT), ensemble of trees (EoT) and support vector machine (SVM), were used to describe the evolution of moisture during the dehydration process of glutinous rice. The hyperparameters of the models were optimized with three strategies: Bayesian optimization, grid search and random search. To understand the parameters that facilitate intelligence model adaptation to the dehydration process, global sensitivity analysis (GSA) was used to compute the impact of the input variables on the model output. The result shows that the optimum computational intelligence techniques include the 3-9-1 topology trained with Bayesian regulation function for ANN, Gaussian kernel function for SVM, Matérn covariance function combined with zero mean function for GPR, boosting method for EoT and 4 minimum leaf size for DT. GPR has the highest performance with R2 of 100% and 99.71% during calibration and testing of the model, respectively. GSA reveals that all the models significantly rely on the variation in time as the main factor that affects the model outputs. Therefore, the computational intelligence models, especially GPR, can be applied for an effective description of moisture evolution during small-scale and industrial dehydration of glutinous rice. © 2024 Society of Chemical Industry.

Classification of Medicinal Plant Leaves for Types and Diseases with Hybrid Deep Learning Methods

Leaf images are often used to detect plant diseases because most disease symptoms appear on the leaves. Analyzes performed by experts in the laboratory environment are expensive and time consuming. Therefore, there is a need for automated plant disease detection systems that are both economical and can help diagnose early symptoms more accurately. In this study, a deep learning-based methodology is presented for the classification of leaf diseases of plants, which are very similar in color, texture, vein and shape and cannot be noticed by non-experts, which are important for traditional medicine and pharmaceutical industry. In the model development process, 7 pre-learning deep learning algorithms and an image data set created from plant leaves in 10 categories were preferred. The proposed model classifies the plant type and diseased condition in the dataset. In the first step of training the model, different learning rates were tested with optimum hyperparameters. In the second part, a test accuracy rate of 98.69% was achieved with the DenseNet121 model, with increased data. At the last stage, after the edge detection processes, the test accuracy value of 67.92% was reached with the DenseNet 121 model.

Enhanced Two-Stream Bayesian Hyper Parameter Optimized 3D-CNN Inception-v3 Based Drop-ConvLSTM2D Deep Learning Model for Human Action Recognition

Human Action Recognition (HAR) has grown to be the toughest and most attractive concern in the domains of computer vision, communication between a person and the surroundings, and video surveillance. In variation to the conventional methods that usually make use of the Long Short Term Memory model (LSTM) for training, this work designed dropout variant Drop-ConvLSTM2D, to provide more effectiveness in regularization for deep Convolution Neural Networks (CNNs).In addition, to speed up the runtime performance of the Deep Learning model, Bayesian Hyper Parameter Optimization (BHPO) is also introduced to autonomously optimize, the hyperparameters of the trained architecture. In this study, a two-stream Bayesian Hyper Parameter optimized Drop-ConvLSTM2D model is designed for HAR to overcome the current research deficiencies. In one stream, an Inception-v3 model extracts the temporal characteristics from the optical frames which are generated through the dense flow process. In another stream, a 3D-CNN involves the mining of the spatial-temporal characteristics from the RGB frames. Finally, the features of Inception-v3 and 3D-CNN are fused using which the Drop-ConvLSTM2D model is trained to recognize human behavior. On perceptive public video datasets UCF-101, and HMDB51, the quantitative assessments are conducted on the Drop-ConvLSTM2D BHPO model. For all hyperparameters, the built model explicitly obtains optimized values in this process, which can save time and improve performance. The experimental outcome shows that with a precision of at least 3%, the designed model beats the traditional two-stream model.

Abstract 4928: REFINE Method: Novel strategy for signal enhancement

Abstract Fragmentomics has emerged as a powerful tool in the detection of early cancer malignancies. One of the challenges for fragmentomics is the low signal to noise ratio in early stage cancer samples. Fragmentomic sequencing data is also inherently noisy and susceptible to batch effects. Common strategies to address these challenges include using normalization methods, such as frequency or Z-score normalization, to scale data. However, these existing methods struggle to differentiate true signals from noise and are susceptible to experimental bias. To address the need for a strategy that removes noise and batch effects, Genece Health is employing a novel normalization technique, the REFINE method, to improve the results of fragmentomic cancer detection. Whole genome sequencing (Illumina) fragmentomic data were generated at an average 2.5x coverage from 3,669 normal and 1,346 cancer samples. Six different cancer types were represented: colorectal, esophageal, liver, lung, ovarian, and pancreatic. Fragment end-motifs and fragment sizes (FEMS) data were then extracted from the sequence data. REFINE first establishes a baseline FEMS signal from a representative and random panel of normals (PoN) comprised of 100 healthy samples. This baseline, which represents the normal background FEMS noise in healthy donors, is subsequently decomposed using truncated singular value decomposition. Next, REFINE then eliminates this decomposed baseline FEMS signal from the raw FEMS data in all other samples using linear regression. A convolutional neural network (CNN) was trained (2,206 normal and 1,000 cancer samples) using frequency normalized data versus REFINE normalized data using 5-fold cross validation. The performances of both models were then assessed using the remaining 1,463 normal with 346 cancer samples. With a training dataset specificity threshold of 90%, REFINE significantly improves the sensitivity performance of the CNN model >30% (from 50.4% to 83.0%). REFINE also improves the auROC from 0.753 to 0.944. We also show that increasing the size of the PoN to 812 samples and including hyper-parameter optimizations further improves the sensitivity to 85.9% (with an auROC of 0.955). The REFINE method demonstrates a dramatic enhancement in the detection of ctDNA from fragmentomic data, overcoming the limitations of conventional normalization techniques. REFINE effectively enhances the signal to noise ratio, facilitating more accurate cancer detection. Future work will focus on validating the REFINE method across diverse datasets and exploring its application in other areas of oncology. Citation Format: Mengchi Wang, Jin Mo Ahn, Junnam Lee, Dasom Kim, Eun-Hae Cho, Byung In Lee, Andrew Carson. REFINE Method: Novel strategy for signal enhancement [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4928.

Performance prediction of a ground source heat pump system using denoised long short-term memory neural network optimised by fast non-dominated sorting genetic algorithm-II

The ground source heat pump (GSHP) is a commonly employed technique that utilises geothermal resources to heat or cool buildings, offering an advantageous alternative for reducing conventional energy consumption. Considering that the output energy of GSHP system is a critical criterion for evaluating the geothermal energy extraction, this study proposes an improved long short-term memory (LSTM) model for predicting the energy output of geothermal heat exchangers (GHE) EGHE-PT and the electrical energy consumption EEC-PT of a GSHP system. The model is trained and validated using a comprehensive real-time monitoring dataset gathered over a four-year period from a three-story residential house situated in Cleveland, Ohio, USA. To process the raw data, the wavelet denoising method is employed, while the fast non-dominated sorting genetic algorithm-II (NSGA-II) is employed to automatically determine the optimal hyperparameters for the LSTM model. Comparative analyses with alternative prediction models demonstrate the superior performance of the Denoised-LSTM-NSGA-II model. The results indicate that the Denoised-LSTM-NSGA-II model yields reasonable prediction for the two performance indicators: EGHE-PT and EEC-PT (with the R2 of 0.91 and 0.89, respectively). Further examination reveals that outdoor temperature holds a significantly high importance rank within the Denoised-LSTM-NSGA-II model. This implies that the proposed method can achieve an acceptable level of accuracy by only utilising weather data from the preceding 5 days. The aforementioned findings present a potentially cost-effective approach for predicting the performance of GSHP system based on limited monitoring data.

Comparative Study of Various Hyperparameter Tuning on Random Forest Classification With SMOTE and Feature Selection Using Genetic Algorithm in Software Defect Prediction

Software defect prediction is necessary for desktop and mobile applications. Random Forest defect prediction performance can be significantly increased with the parameter optimization process compared to the default parameter. However, the parameter tuning step is commonly neglected. Random Forest has numerous parameters that can be tuned, as a result manually adjusting parameters would diminish the efficiency of Random Forest, yield suboptimal results and it will take a lot of time. This research aims to improve the performance of Random Forest classification by using SMOTE to balance the data, Genetic Algorithm as selection feature, and using hyperparameter tuning to optimize the performance. Apart from that, it is also to find out which hyperparameter tuning method produces the best improvement on the Random Forest classification method. The dataset used in this study is NASA MDP which included 13 datasets. The method used contains SMOTE to handle imbalance data, Genetic Algorithm feature selection, Random Forest classification, and hyperparameter tuning methods including Grid Search, Random Search, Optuna, Bayesian (with Hyperopt), Hyperband, TPE and Nevergrad. The results of this research were carried out by evaluating performance using accuracy and AUC values. In terms of accuracy improvement, the three best methods are Nevergrad, TPE, and Hyperband. In terms of AUC improvement, the three best methods are Hyperband, Optuna, and Random Search. Nevergrad on average improves accuracy by about 3.9% and Hyperband on average improves AUC by about 3.51%. This study indicates that the use of hyperparameter tuning improves Random Forest performance and among all the hyperparameter tuning methods used, Hyperband has the best hyperparameter tuning performance with the highest average increase in both accuracy and AUC. The implication of this research is to increase the use of hyperparameter tuning in software defect prediction and improve software defect prediction performance.

Medical Image Classifications Using Convolutional Neural Networks: A Survey of Current Methods and Statistical Modeling of the Literature

In this review, we compiled convolutional neural network (CNN) methods which have the potential to automate the manual, costly and error-prone processing of medical images. We attempted to provide a thorough survey of improved architectures, popular frameworks, activation functions, ensemble techniques, hyperparameter optimizations, performance metrics, relevant datasets and data preprocessing strategies that can be used to design robust CNN models. We also used machine learning algorithms for the statistical modeling of the current literature to uncover latent topics, method gaps, prevalent themes and potential future advancements. The statistical modeling results indicate a temporal shift in favor of improved CNN designs, such as a shift from the use of a CNN architecture to a CNN-transformer hybrid. The insights from statistical modeling point that the surge of CNN practitioners into the medical imaging field, partly driven by the COVID-19 challenge, catalyzed the use of CNN methods for detecting and diagnosing pathological conditions. This phenomenon likely contributed to the sharp increase in the number of publications on the use of CNNs for medical imaging, both during and after the pandemic. Overall, the existing literature has certain gaps in scope with respect to the design and optimization of CNN architectures and methods specifically for medical imaging. Additionally, there is a lack of post hoc explainability of CNN models and slow progress in adopting CNNs for low-resource medical imaging. This review ends with a list of open research questions that have been identified through statistical modeling and recommendations that can potentially help set up more robust, improved and reproducible CNN experiments for medical imaging.

Rolling the dice for better deep learning performance: A study of randomness techniques in deep neural networks

This paper presents a comprehensive empirical investigation into the interactions between various randomization techniques in Deep Neural Networks (DNNs) and their impact on learning performance. It is well-established that injecting randomness into the training process of DNNs, through various approaches, at different stages, is often beneficial for reducing overfitting and improving generalization. Nonetheless, the interactions between randomness techniques such as weight noise, dropout, and many others remain poorly understood. Consequently, it is challenging to determine which methods can be effectively combined to optimize DNN performance. To address this issue, we categorize the existing randomness techniques into four key types: injection of noise/randomness at the data, model structure, optimization or learning stage. We use this classification to identify gaps in the current coverage of potential mechanisms for the introduction of randomness, leading to proposing two new techniques: adding noise to the loss function and random masking of the gradient updates.In our empirical study, we employ a Particle Swarm Optimizer (PSO) for hyperparameter optimization (HPO) to explore the space of possible configurations to determine where and how much randomness should be injected to maximize DNN performance. We assess the impact of various types and levels of randomness for DNN architectures across standard computer vision benchmarks: MNIST, FASHION-MNIST, CIFAR10, and CIFAR100. Across more than 30000 evaluated configurations, we perform a detailed examination of the interactions between randomness techniques and their combined impact on DNN performance. Our findings reveal that randomness through data augmentation and in weight initialization are the main contributors to performance improvement. Additionally, correlation analysis demonstrates that different optimizers, such as Adam and Gradient Descent with Momentum, prefer distinct types of randomization during the training process. A GitHub repository with the complete implementation and generated dataset is available.2

Research on optimization of key thermal parameters of the secondary loop of PWR based on improved BP neural network

The simulation and optimization time of traditional secondary loop thermal parameters is relatively long, and the computational cost is generally high. To solve this problem, we construct an accurate model of the full-power steady-state operation of the secondary loop of the PWR, and use meta-learning combined with Bayesian optimization to realize the adaptive optimization of BP neural network hyperparameters. The effectiveness of the method is verified by designing comparative experiments by taking thermal efficiency as the optimization goal and high-pressure heater end difference and other parameters as the optimization variables, and using the improved BP neural network to optimize the accurate model. The results show that the improved secondary loop thermal parameter prediction and optimization framework can quickly and accurately predict the thermal efficiency and optimize the thermal parameters, and has good generalization ability, which lays a foundation for improving the thermal efficiency of nuclear power plants.

Multi-source driven estimation of earthquake economic losses: A comprehensive and interpretable ensemble machine learning model

Rapid and precise quantification of economic losses post-earthquake is critical for crafting informed disaster management strategies by governmental and insurance entities. This study introduces an ensemble model, constituted by Support Vector Machines (SVM) and Extreme Gradient Boosting (XGBoost), tailored for quick and interpretative prediction of GDP-related seismic loss assessments, with Sichuan Province serving as the empirical backdrop. The ensemble approach, pre-trained for expediency, draws on a knowledge-driven selection of 30 features that span seismic risks, socio-economic variables, and exposure factors. Enhanced by feature scaling and Bayesian hyperparameter optimization, the model's efficacy is validated through 6 metrics such as R-Squared, MSE, and MAE etc. The incorporation of SHAP values further unravels the model's decision-making, providing transparency to the often opaque computations of machine learning. This methodology offers a scalable, interpretable framework that equips stakeholders with timely and accurate insights for risk mitigation, ultimately strengthening earthquake resilience.

An optimum hyperparameters of restnet-50 for orchid classification based on convolutional neural network

There are many types of orchids in Indonesia, such as Phalaenopsis Amabilis (Moon Orchid), Cattleya, etc. Because the shape and color of each orchid flower looks the same, a system is needed that can classify orchid flowers. In this research, we will use a system using a Convolutional Neural Network with ResNet50 architecture to classify orchid species. There are 4 types of orchids that will be used, namely Moon Orchids, xDoritaenopsis Orchids, Cattleya Orchids, and Coelogyne Pandurata Orchids (1000 datasets for each type). The aim of this research is to implement deep learning using the Convolutional Neural Network method combined with the ResNet50 architecture and identifying the types of orchid flowers and calculating accuracy when identifying orchid flower types. This research uses 4000 orchid image datasets, with a data split of 80:20 so that 800 images are used as training data and 200 as test data. ResNet50 uses a confusion matrix evaluation, namely Accuracy, Precision, Recall, Specificity and F1-score with epochs 10, 20, 30, 40. From the research that has been carried out, it produces the highest accuracy on Test Data with the 30th epoch, reaching 98.87%. and the lowest accuracy on Test Data with the 10th epochs which produces an accuracy of 97.75%.

Spatio-temporal prediction of photovoltaic power based on a broad learning system and an improved backtracking search optimization algorithm

The accuracy of photovoltaic (PV) power forecasting techniques relies not only on high-quality spatiotemporal data but also on an efficient feature-mining methodology. In this study, a spatiotemporal power forecasting model based on the broad learning system (BLS) and the improved backtracking search optimization algorithm (IBSOA) is proposed. The objective is to enhance the accuracy of PV power predictions while reducing the time-intensive training process associated with an extensive set of broad learning system parameters. The spatiotemporal attributes of historical data from multiple PV sites are clustered using a self-organizing map. The clustering analysis explores the spatiotemporal correlation among five photovoltaic (PV) power stations for each season between 2017 and 2018. Subsequently, the IBSOA is employed to optimize the hyperparameters of the BLS model, particularly the mapping and enhancement nodes. By utilizing hyperparameter optimization, a BSOA-based broad learning model is introduced to achieve superior accuracy. The results are assessed using the proposed method in comparison with three popular optimization algorithms: 1) genetic algorithm (GA), 2) bird swarm algorithm (BSA), and 3) backtracking search optimization algorithm (BSOA). All scenarios are validated and compared using PV plant data from the DKA center in Australia. The root-mean-square error (RMSE) indicators of the proposed prediction method are consistently lower than the worst-case scenario in each season, decreasing by 3.2283 kW in spring, 3.9159 kW in summer, 1.3425 kW in autumn, and 1.4058 kW in winter. Similarly, the mean absolute percentage error (MAPE) exhibits a reduction compared to the worst case, with a decreases of 0.882% in spring, 1.2399% in summer, 1.803% in autumn, and 1.087% in winter. The comprehensive results affirm that the proposed method surpasses alternative optimization techniques, delivering high-quality power forecasts for the given case study.

Enhancing breast cancer screening with urinary biomarkers and Random Forest supervised classification: A comprehensive investigation

ObjectivesUrinary sex hormones are investigated as potential biomarkers for the early detection of breast cancer, aiming to evaluate their relevance and applicability, in combination with supervised machine-learning data analysis, toward the ultimate goal of extensive screening. MethodsSex hormones were determined on urine samples collected from 250 post-menopausal women (65 healthy - 185 with breast cancer, recruited among the clinical patients of Candiolo Cancer Institute FPO-IRCCS (Torino, Italy). Two analytical procedures based on UHPLC-MS/HRMS were developed and comprehensively validated to quantify 20 free and conjugated sex hormones from urine samples. The quantitative data were processed by seven machine learning algorithms. The efficiency of the resulting models was compared. ResultsAmong the tested models aimed to relate urinary estrogen and androgen levels and the occurrence of breast cancer, Random Forest (RF) proved to underscore all the other supervised classification approaches, including Partial Least Squares – Discriminant Analysis (PLS-DA), in terms of effectiveness and robustness. The final optimized model built on only five biomarkers (testosterone-sulphate, alpha-estradiol, 4-methoxyestradiol, DHEA-sulphate, and epitestosterone-sulphate) achieved an approximate 98% diagnostic accuracy on replicated validation sets. To balance the less-represented population of healthy women, a Synthetic Minority Oversampling TEchnique (SMOTE) data oversampling approach was applied. ConclusionsBy means of tunable hyperparameters optimization, the RF algorithm showed great potential for early breast cancer detection, as it provides clear biomarkers ranking and their relative efficiency, allowing to ground the final diagnostic model on a restricted selection five steroid biomarkers only, as desirable for noninvasive tests with wide screening purposes.

Evaluation of Whole Brain Intravoxel Incoherent Motion (IVIM) Imaging.

Intravoxel Incoherent Motion (IVIM) imaging provides non-invasive perfusion measurements, eliminating the need for contrast agents. This work explores the feasibility of IVIM imaging in whole brain perfusion studies, where an isotropic 1 mm voxel is widely accepted as a standard. This study follows the validity of a time-limited, precise, segmentation-ready whole-brain IVIM protocol suitable for clinical reality. To assess the influence of SNR on the estimation of S0, f, D*, and D IVIM parameters, a series of measurements and simulations were performed in MATLAB for the following three estimation techniques: segmented grid search, segmented curve fitting, and one-step curve fitting, utilizing known "ground truth" and noised data. Scanner-specific SNR was estimated based on a healthy subject IVIM MRI study in a 3T scanner. Measurements were conducted for 25.6 × 25.6 × 14.4 cm FOV with a 256 × 256 in-plane resolution and 72 slices, resulting in 1 × 1 × 2 mm voxel size. Simulations were performed for 36 SNR levels around the measured SNR value. For a single voxel grid, the search algorithm mean relative error Ŝ0, f^, D^*, and D^ of at the expected SNR level were 5.00%, 81.91%, 76.31%, and 18.34%, respectively. Analysis has shown that high-resolution IVIM imaging is possible, although there is significant variation in both accuracy and precision, depending on SNR and the chosen estimation method.

Probabilistic state estimation in district heating grids using deep neural network

Probabilistic state estimation is critical for operating and controlling district heating grids efficiently. However, computational bottlenecks of traditional solvers limit the feasibility of uncertainty-aware Bayesian estimation. This paper proposes using deep neural networks (DNNs) to enable fast and accurate posterior estimation. Fully-connected neural networks (FCNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs) are evaluated as candidate approximators of the physical model. Markov chain Monte Carlo sampling in the heat exchange space is leveraged to generate posterior samples. Experiments on a benchmark heating grid demonstrate FCNNs can efficiently learn the mapping from heat exchanges to network states. A FCNN trained on 20 training epochs after hyperparameter optimization provides the best approximation accuracy and uncertainty estimates, outperforming prior methods based on Deep Neural Networks. The results highlight the potential of data-driven deep learning models for probabilistic state estimation. The proposed framework could enable real-time uncertainty-aware control and decision-making for future intelligent district heating grids.

Machine Learning-Based Figure of Merit Model of SIPOS Modulated Drift Region for U-MOSFET.

This paper presents a machine learning-based figure of merit model for superjunction (SJ) U-MOSFET (SSJ-UMOS) with a modulated drift region utilizing semi-insulating poly-crystalline silicon (SIPOS) pillars. This SJ drift region modulation is achieved through SIPOS pillars beneath the trench gate, focusing on optimizing the tradeoff between breakdown voltage (BV) and specific ON-resistance (RON,sp). This analytical model considers the effects of electric field modulation, charge-coupling, and majority carrier accumulation due to additional SIPOS pillars. Gaussian process regression is employed for the figure of merit (FOM = BV2/RON,sp) prediction and hyperparameter optimization, ensuring a reasonable and accurate model. A methodology is devised to determine the optimal BV-RON,sp tradeoff, surpassing the SJ silicon limit. The paper also delves into a discussion of optimal structural parameters for drift region, oxide thickness, and electric field modulation coefficients within the analytical model. The validity of the proposed model is robustly confirmed through comprehensive verification against TCAD simulation results.

Source mechanism of kHz microseismic events recorded in multiple boreholes at the first EGS Collab testbed

Continuous microseismic monitoring using three-component (3C) accelerometers deployed in multiple boreholes allows for tracking the detailed evaluation of mesoscale (∼10 m scale) fracture growth during the fracture stimulation experiments at the first Enhanced Geothermal Systems (EGS) Collab testbed. Building on a well-constrained microseismic event catalog, we invert for moment tensor of the events to better understand the fracture geometry and stress orientations. However, it is challenging because of the unknown orientation of 3C accelerometers and low signal-to-noise-ratio nature of high-frequency (several kHz) monitoring. To address these challenges, we first perform the hodogram analysis on the continuous active-source seismic monitoring (CASSM) data to determine the orientations of the 18 3C accelerometers. We then apply the principal component analysis (PCA) to the observed microseismic waveforms to improve the signal-to-noise ratios. We perform a grid search for the full moment tensor by fitting the PCA-denoised waveforms at a frequency range of 5 to 8 kHz. The moment tensor results show both the creation of hydraulic fractures and the reactivation of natural fractures during the hydraulic stimulations. Our stress inversion based on the inverted moment tensors reveals the alteration of stress regime caused by hydraulic fracture stimulations.