Training Set Research Articles

Δ-Machine Learning to Elevate DFT-Based Potentials and a Force Field to the CCSD(T) Level Illustrated for Ethanol.

Progress in machine learning has facilitated the development of potentials that offer both the accuracy of first-principles techniques and vast increases in the speed of evaluation. Recently, Δ-machine learning has been used to elevate the quality of a potential energy surface (PES) based on low-level, e.g., density functional theory (DFT) energies and gradients to close to the gold-standard coupled cluster level of accuracy. We have demonstrated the success of this approach for molecules, ranging in size from H3O+ to 15-atom acetyl-acetone and tropolone. These were all done using the B3LYP functional. Here, we investigate the generality of this approach for the PBE, M06, M06-2X, and PBE0 + MBD functionals, using ethanol as the example molecule. Linear regression with permutationally invariant polynomials is used to fit both low-level and correction PESs. These PESs are employed for standard RMSE analysis for training and test data sets, and then general fidelity tests such as energetics of stationary points, normal-mode frequencies, and torsional potentials are examined. We achieve similar improvements in all cases. Interestingly, we obtained significant improvement over DFT gradients where coupled cluster gradients were not used to correct the low-level PES. Finally, we present some results for correcting a recent molecular mechanics force field for ethanol and comment on the possible generality of this approach.

Robust unsupervised 5D seismic data reconstruction on regular and irregular grids

Seismic data reconstruction has become a central focus in seismic data processing, addressing challenges posed by sparse sampling due to physical and budgetary constraints. The advent of 5D acquisition methodologies marks a significant advancement in the quality and completeness of seismic data sets. Most traditional 5D reconstruction methods commonly use the fast Fourier transform (FFT), requiring regular grids and preliminary 4D binning before 5D interpolation. Discrete Fourier transform and nonequidistant FFT can honor the original irregular coordinates. However, when using exact locations, these methods become computationally expensive. We introduce an unsupervised deep-learning methodology to learn a continuous function across the sampling points in seismic data, facilitating reconstruction on regular and irregular grids. The network comprises a multilayer perceptron with linear layers and element-wise periodic activation functions. It excels at mapping the input coordinates to the corresponding seismic data amplitudes without relying on external training sets. The network’s intrinsic low-frequency bias is crucial in prioritizing acquiring self-similar features over high-frequency and incoherent ones during training. This characteristic mitigates incoherent noise in seismic data, such as random and erratic components. To assess the robustness of the unsupervised reconstruction technique, we conduct comprehensive evaluations using synthetic data examples sampled regularly and irregularly, as well as field-data examples with and without binning. The findings demonstrate the efficacy of our deep-learning framework in achieving resilient and accurate seismic data reconstruction across diverse sampling scenarios.

Few-Shot Anomaly Detection via Category-Agnostic Registration Learning.

Most existing anomaly detection (AD) methods require a dedicated model for each category. Such a paradigm, despite its promising results, is computationally expensive and inefficient, thereby failing to meet the requirements for real-world applications. Inspired by how humans detect anomalies, by comparing a query image to known normal ones, this article proposes a novel few-shot AD (FSAD) framework. Using a training set of normal images from various categories, registration, aiming to align normal images of the same categories, is leveraged as the proxy task for self-supervised category-agnostic representation learning. At test time, an image and its corresponding support set, consisting of a few normal images from the same category, are supplied, and anomalies are identified by comparing the registered features of the test image to its corresponding support image features. Such a setup enables the model to generalize to novel test categories. It is, to our best knowledge, the first FSAD method that requires no model fine-tuning for novel categories: enabling a single model to be applied to all categories. Extensive experiments demonstrate the effectiveness of the proposed method. Particularly, it improves the current state-of-the-art (SOTA) for FSAD by 11.3% and 8.3% on the MVTec and MPDD benchmarks, respectively. The source code is available at https://github.com/Haoyan-Guan/CAReg.

Multi-omics analysis and response prediction of PD-1 monoclonal antibody containing regimens in patients with relapsed/refractory diffuse large B-cell lymphoma

Patients with relapsed/refractory (r/r) diffuse large B-cell lymphoma (DLBCL) show varied responses to PD-1 monoclonal antibody (mAb) containing regimens. The mechanisms and predictive biomarkers for the efficacy of this regimen are unclear. This study retrospectively collected r/r DLBCL patients who received PD-1 mAb and rituximab regimens as salvage therapy. Clinical and genomic features were collected, and mechanisms were explored by multiplex immunofluorescence and digital spatial profiling. An artificial neural network (ANN) model was constructed to predict the response. Between October 16th, 2018 and May 4th, 2023, 50 r/r DLBCL patients were collected, 29 were response patients and 21 were non-response patients. CREBBP (p = 0.029) and TP53 (p = 0.015) alterations were statistically higher in non-response patients. Patients with PD-L1 CPS ≥ 5 were correlated with a longer overall survival (OS) than those with PD-L1 CPS < 5 (median OS: not reached vs. 9.7 months, hazard ratio [HR]: 3.8, 95% confidence interval [CI] 0.64–22.44, p = 0.016). Immune-related pathways were activated in response patients. The proportion and spatial organization of tumor-infiltrating immune cells affect the response. PD-L1 CPS level, age, and alterations of TP53, MYD88, CREBBP, EP300, GNA13 were used to build an ANN predictive model that showed high prediction efficiency (training set area under curve [AUC] of 0.97 and test set AUC of 0.94). The proportion and spatial distribution of tumor-infiltrating immune cells may be related to the function of immune-related pathways, thereby influencing the efficacy of PD-1 mAb containing regimens. The ANN predictive model showed potential value in predicting the responses of r/r DLBCL patients received PD-1 mAb and rituximab regimens.

Prediction of preoperative lymph-vascular space invasion and survival outcomes of cervical squamous cell carcinoma by utilizing 18F-FDG PET/CT imaging at early stage.

To establish nomograms for predicting preoperative lymph-vascular space invasion (LVSI) and survival outcomes of cervical squamous cell carcinoma (CSCC) based on PET/CT radiomics. One hundred and twenty-three patients with CSCC and LVSI status were enrolled retrospectively. Independent predictors of LVSI were identified through clinicopathological factors and PET/CT metabolic parameters. We extracted 1316 features from PET and CT volume of interest, respectively. Additionally, four models (PET-RS: radiomic signature of PET only; CT-RS: radiomic signature of CT only; PET/CT-RS + clinical data; PET/CT-RS: radiomic signature of PET and CT) were established to predict LVSI status. Calculation of radiomics scores of PET/CT was executed for assessment of the survival outcomes, followed by development of nomograms with radiomics (NR) or without radiomics (NWR). One hundred and twenty-three patients with pathologically confirmed CSCC had been categorized into two sets (training and testing sets). It was found that only maximum standardized uptake value (SUVmax) and squamous cell carcinoma antigen were independent predictors of LVSI. Meanwhile, the PET/CT-RS + clinical data outperformed the other three models in the training set [area under the curve (AUC): 0.91 vs. 0.861 vs. 0.81 vs. 0.814] and the testing set (AUC: 0.885 vs. 0.857 vs. 0.783 vs. 0.798). Additionally, SUVmax and LVSI had been demonstrated to be independent prognostic indicators for progression-free survival and overall survival. Decision curve analysis and calibration curve indicated that NRs were superior to NWRs. The survival outcomes were assessed. PET/CT-based radiomic signature nomogram enables a new method for preoperative prediction of LVSI and survival prognosis for patients with CSCC.

Machine learning analysis of a large set of homopolymers to predict glass transition temperatures

Glass transition temperature of polymers, Tg, is an important thermophysical property, which sometimes can be difficult to measure experimentally. In this regard, data-driven machine learning approaches are important alternatives to assess Tg values, in a high-throughput way. In this study, a large dataset of more than 900 polymers with reported glass transition temperature (Tg) was assembled from various public sources in order to develop a predictive model depicting the structure-property relationships. The collected dataset was curated, explored via cluster analysis, and then split into training and test sets for validation purposes and then polymer structures characterized by molecular descriptors. To find the models, several machine learning techniques, including multiple linear regression (MLR), k-nearest neighbor (k-NN), support vector machine (SVM), random forest (RF), gaussian processes for regression (GPR), and multi-layer perceptron (MLP) were explored. As result, a model with the subset of 15 descriptors accurately predicting the glass transition temperatures was developed. The electronic effect indices were determined to be important properties that positively contribute to the Tg values. The SVM-based model showed the best performance with determination coefficients (R2) of 0.813 and 0.770, for training and test sets, respectively. Also, the SVM model showed the lowest estimation error, RMSE = 0.062. In addition, the developed structure-property model was implemented as a web app to be used as an online computational tool to design and evaluate new homopolymers with desired glass transition profiles.

Predicting herbage biomass on small‐scale farms by combining sward height with different aggregations of weather data

AbstractAccurate predictions of herbage biomass are important for efficient grazing management. Small‐scale farms face challenges using remote sensing technologies due to insufficient resources. This limitation hinders their ability to develop machine learning‐based prediction models. An alternative is to adopt less expensive measurement methods and readily available data such as weather data. This study aimed to examine how different temporal aggregations of weather data combined with compressed sward height (CSH) affect the prediction performance. We considered weather features based on different numbers of weather variables, statistical functions, weather events, and periods. Between 2019 and 2021, data were collected from 11 organic dairy farms in Germany. Herbage biomass exhibited high variability (coefficient of variation [CV] = 0.65). Weather data were obtained from on‐farm and nearby public stations. Prediction models were learned on a training set (n = 291) and evaluated on a test set (n = 125). Random forest models performed better than models based on artificial neural networks and support vector regression. Representing weather data by a single feature for leaf wetness reduced the root mean square error (RMSE) by 12.1% (from 536 to 471 kg DM ha−1, where DM is dry matter) and increased the R2 by 0.109 (from 0.518 to 0.627). Adding features based on multiple variables, functions, events, and periods resulted in a further reduction in RMSE by 15.9% (R2 = 0.737). Overall, different aggregations of weather data enhanced the accuracy of CSH‐based models. These aggregations do not cause additional effort for data collection and, therefore, should be integrated into CSH‐based models for small‐scale farms.

TSOTSALearning at LLMs4OL Tasks A and B : Combining Rules to Large Language Model for Ontology Learning

This paper presents our contribution to the Large Language Model For Ontology Learning (LLMs4OL) challenge hosted by ISWC conference. The challenge involves extracting and classifying various ontological components from multiple datasets. The organizers of the challenge provided us with the train set and the test set. Our goal consists of determining in which conditions foundation models such as BERT can be used for ontologies learning. To achieve this goal, we conducted a series of experiments on various datasets. Initially, GPT-4 was tested on the wordnet dataset, achieving an F1-score of 0.9264. Subsequently, we performed additional experiments on the same dataset using BERT. These experiments demonstrated that by combining BERT with rule-based methods, we achieved an F1-score of 0.9938, surpassing GPT-4 and securing the first place for term typing on the Wordnet dataset.

Utilizing artificial neural network system to predict the residual valve after endoscopic posterior urethral valve ablation.

To build, train, and assess the artificial neural network (ANN) system in estimatingthe residual valve rate after endoscopic valve ablationand compare the data obtained with conventional analysis. In a retrospective cross-sectional study between June 2010 and December 2020, 144 children with a history of posteriorurethral valve (PUV) who underwent endoscopic valve ablation were enrolled in the study. MATLAB software was used to design and train the network in a feed-forward backpropagationerror adjustment scheme. Preoperative and postoperative data from 101 patients (70%) (training set) were utilized to assessthe impact and relative significance of the necessity for repeatedablation. The validated suitably trained ANN was used to predict repeated ablation in the next 33 patients (22.9%) (test set)whose preoperative data were serially input into the system. To assess system accuracy in forecasting the requirement for repeatablation, projected values were compared to actual outcomes. The likelihood of predicting the residual valve was calculatedusing a three-layered backpropagating deep ANN using preoperativeand postoperative information. Of 144 operated cases, 33 (22.9%) had residual valvesand needs to repeated ablation. The ANN accuracy, sensitivity, and specificity for predicting the residual valve were 90.75%,92.73%, and 73.19%, respectively. Younger age at surgery, hyperechogenicity of the renal parenchyma, presence of vesicoureteralreflux (VUR), and grade of reflux before surgery were among the most significant characteristics that affectedpostoperative outcome variables, the need for repeated ablation, and were given the highest relative weight by the ANN system. Conclusions: The ANN is an integrated data-gathering tool for analyzing and finding relationships among variables as a complex non-linear statistical model. The results indicate that ANN is a valuable tool for outcome prediction of the residualvalveafter endoscopic valve ablation in patients with PUV.

A-231 A modified Sampson LDL-C equation to improve concordance with the reference method at low levels of LDL-C

Abstract Background Patients at high risk for Atherosclerotic Cardiovascular Disease (ASCVD), who do not achieve adequate lowering of LDL-cholesterol (LDL-C) with maximally tolerated statins or who are statin-intolerant, are often treated with proprotein convertase subtilisin/kexin (type 9) serine protease inhibitor (PCSK9-inh) therapy. The LDL-C reference method, Beta-Quantitation (BQ), is not routinely performed in these patients. Because low values are measured less accurately with current methods, a variety of equations have been developed to calculate LDL-C from the standard lipid panel test results (total cholesterol (TC), triglycerides (TG), and high-density lipoprotein cholesterol (HDL-C)). We further modified the Sampson LDL-C (S-LDL-C) equation by a novel approach by combining logistic regression (LR) and linear regression to minimize errors when compared to BQ for low LDL-C. Methods A dataset of BQ results from Mayo Medical Labs was divided randomly into a training set (TG ≤ 1500 mg/dL, TC ≤ 1050 mg/dL, n=19,639) and a validation set (TG ≤ 800 mg/dL, BQ-LDL-C ≤ 130 mg/dL, n=12,910). The training dataset was analyzed by LR, using BQ-LDL-C to classify patients as being above or below the 70 mg/dL cut-point. Dependent variables in the LR model were the S-LDL-C, HDL-C, TC, TG and an interaction term (TG x NonHDL-C). The sum of the exponential terms in the trained LR equation for the dependent variables were then fitted by least squares linear regression to match BQ-LDL-C to develop the modified S-LDL-C equation (mS-LDL-C). We used the validation dataset to compare Friedewald (F-LDL-C), extended Martin (eM-LDL-C), Sampson (S-LDL-C) and modified Sampson (mS-LDL-C) equations by Cohen’s kappa coefficient for agreement with BQ-LDL-C and normalized Matthew’s correlation coefficient (MCC), which considers any imbalances in the number of false high versus false low results. Results The mS-LDL-C equation (mS-LDL-C = (TC/0.9423 - HDL-C/0.9582 - TG/9.397 - ((TGxNonHDL-C)/1305) + TG2/8569) - 5.688), primarily differs from the S-LDL-C equation in the value of the coefficients that include TG as a variable. For S-LDL-C ≤100 mg/dL, mS-LDL-C had the best concordance with BQ-LDL-C in terms of the total percent of patients misclassified as either high or low (mS-LDL-C: 21.4%, S-LDL-C: 24.4%, eM-LDL-C: 25.4% F-LDL-C: 30.8%). mS-LDL-C also had the best kappa score when compared to BQ (mS-LDL-C: 0.833, S-LDL-C: 0.810, eM-LDL-C: 0.789, F-LDL-C: 0.743). MCC was also the highest for mS-LDL-C (mS-LDL-C: 0.916, S-LDL-C: 0.906, eM-LDL-C: 0.894, F-LDL-C: 0.879). Finally, mS-LDL-C also showed the best concordance with BQ-LDL-C when an alternative lower LDL-C therapeutic decision cut-point of 55 mg/dL is used (kappa of 0.766 versus 0.724) but showed an increasingly poorer performance than S-LDL-C for higher LDL-C whenever it exceeded 150 mg/dL. Conclusions A strategy to maximize the overall accuracy of estimated LDL-C could provide an automatic reflex in a laboratory information system to the mS-LDL-C equation whenever S-LDL-C is &amp;lt;100 mg/dL. By doing so, the use of the more accurate mS-LDL-C equation for low LDL-C should improve PCSK9-inh therapeutic decision making over the other LDL-C equations.

Is an Electronic Nose Able to Predict Clinical Response following Neoadjuvant Treatment of Rectal Cancer? A Prospective Pilot Study

Introduction: A watch-and-wait strategy for patients with rectal cancer who achieve a clinical complete response after neoadjuvant (chemo) radiotherapy is a valuable alternative to rectal resection. In this pilot study, we explored the use of an electronic nose to predict response to neoadjuvant therapy by analyzing breath-derived volatile organic compounds. Materials and Methods: A pilot study was performed between 2020 and 2022 on patients diagnosed with intermediate- or high-risk rectal cancer who were scheduled for neoadjuvant therapy. Breath samples were collected before and after (chemo) radiotherapy. A machine-learning model was developed to predict clinical response using curatively treated rectal cancer patients as controls. Results: For developing the machine-learning model, a total of 99 patients were included: 45 patients with rectal cancer and 54 controls. In the training set, the model successfully discriminated between patients with and without rectal cancer, with a sensitivity and specificity of 0.80 and 0.65, respectively, and an accuracy of 0.72. In the test set, the model predicted partial or (near) complete response with a sensitivity and specificity of 0.64 and 0.47, respectively, and an accuracy of 0.58. The AUC of the ROC curve was 0.63. Conclusions: The prediction model developed in this pilot study lacks the ability to accurately differentiate between partial and (near) complete responders with an electronic nose. Machine-learning studies demand a substantial number of patients and operate in a rapidly evolving field. Therefore, the prevalence of disease and duration of a study are crucial considerations for future research.

Galileo-an Artificial Intelligence tool for evaluating pre-implantation kidney biopsies.

Pre-transplant procurement biopsy interpretation is challenging, also because of the low number of renal pathology experts. Artificial intelligence (AI) can assist by aiding pathologists with kidney donor biopsyassessment. Herein we present the "Galileo" AI tool, designed specifically to assist the on-call pathologist with interpreting pre-implantation kidney biopsies. A multicenter cohort of whole slide images acquired from core-needle and wedge biopsies of the kidney was collected. A deep learning algorithm was trained to detect the main findings evaluated in the pre-implantation setting (normal glomeruli, globally sclerosed glomeruli, ischemic glomeruli, arterioles and arteries). The model obtained on the Aiforia Create platform was validated on an external dataset by three independent pathologists to evaluate the performance of the algorithm. Galileo demonstrated a precision, sensitivity, F1 score and total area error of 81.96%, 94.39%, 87.74%, 2.81% and 74.05%, 71.03%, 72.5%, 2% in the training and validation sets, respectively. Galileo was significantly faster than pathologists, requiring 2min overall in the validation phase (vs 25, 22 and 31min by 3 separate human readers, p < 0.001). Galileo-assisted detection of renal structures and quantitative information was directly integrated in the final report. The Galileo AI-assisted tool shows promise inspeeding up pre-implantation kidney biopsy interpretation, as well as inreducing inter-observer variability. This toolmay represent a starting point for further improvements based on hardendpoints such as graft survival.

Immunogenic cell death signatures from on-treatment tumor specimens predict immune checkpoint therapy response in metastatic melanoma

Melanoma is a highly malignant form of skin cancer that typically originates from abnormal melanocytes. Despite significant advances in treating metastatic melanoma with immune checkpoint blockade (ICB) therapy, a substantial number of patients do not respond to this treatment and face risks of recurrence and metastasis. This study collected data from multiple datasets, including cohorts from Riaz et al., Gide et al., MGH, and Abril-Rodriguez et al., focusing on on-treatment samples during ICB therapy. We used the single-sample gene set enrichment analysis (ssGSEA) method to calculate immunogenic cell death scores (ICDS) and employed an elastic network algorithm to construct a model predicting ICB efficacy. By analyzing 18 ICD gene signatures, we identified 9 key ICD gene signatures that effectively predict ICB treatment response for on-treatment metastatic melanoma specimens. Results showed that patients with high ICD scores had significantly higher response rates to ICB therapy compared to those with low ICD scores. ROC analysis demonstrated that the AUC values for both the training and validation sets were around 0.8, indicating good predictive performance. Additionally, survival analysis revealed that patients with high ICD scores had longer progression-free survival (PFS). This study used an elastic network algorithm to identify 9 ICD gene signatures related to the immune response in metastatic melanoma. These gene features can not only predict the efficacy of ICB therapy but also provide references for clinical decision-making. The results indicate that ICD plays an important role in metastatic melanoma immunotherapy and that expressing ICD signatures can more accurately predict ICB treatment response and prognosis for on-treatment metastatic melanoma specimens, thus providing a basis for personalized treatment.

Coping with intensive training demands: a longitudinal investigation of the relationships between appraisal, emotion, coping effectiveness and engagement among elite fencers

ABSTRACT Training is, by essence, very challenging, both physically and mentally. The way athletes stay fully engaged in their intensive training programme and achieve their performance goals may depend on their capacity to deal effectively with such daily training demands. This study therefore aimed to further the understanding of the relationships between appraisal, emotion, coping effectiveness and engagement among elite fencers in relation to the demands of intensive training over the course of a competitive season. Twenty-three elite fencers (15 women, 8 men; age: 23.04 ± 3.54 years; fencing experience: 15.04 ± 3.74 years) completed an online questionnaire once a week, measuring perceived effort, perceived difficulty, pleasant and unpleasant emotions, athlete engagement, and coping effectiveness (i.e., 36 time-points). The results of the random coefficient regression models revealed the mediating role played by emotions in the dynamic relationships (a) between perceived difficulty and coping effectiveness (pleasant emotions: Sobel Test = −2.39, p < .05; unpleasant emotions: Sobel Test = −2.47, p < .05), and (b) between perceived effort and coping effectiveness (pleasant emotions: Sobel Test = 2.15, p < .05). Results also highlighted the partial mediating role played by coping effectiveness in the dynamic relationships (a) between perceived effort and engagement (Sobel Test = 2.27, p < .05), and (b) between perceived difficulty and engagement (Sobel Test = −2.16, p < .05). This study underlines the need to consider subjective appraisal during training over the course of a competitive season and its influence on psychological adaptation and engagement in intensive training settings.

A Hybrid Deep Learning Approach with Generative Adversarial Network for Credit Card Fraud Detection

Credit card fraud detection is a critical challenge in the financial industry, with substantial economic implications. Conventional machine learning (ML) techniques often fail to adapt to evolving fraud patterns and underperform with imbalanced datasets. This study proposes a hybrid deep learning framework that integrates Generative Adversarial Networks (GANs) with Recurrent Neural Networks (RNNs) to enhance fraud detection capabilities. The GAN component generates realistic synthetic fraudulent transactions, addressing data imbalance and enhancing the training set. The discriminator, implemented using various DL architectures, including Simple RNN, Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRUs), is trained to distinguish between real and synthetic transactions and further fine-tuned to classify transactions as fraudulent or legitimate. Experimental results demonstrate significant improvements over traditional methods, with the GAN-GRU model achieving a sensitivity of 0.992 and specificity of 1.000 on the European credit card dataset. This work highlights the potential of GANs combined with deep learning architectures to provide a more effective and adaptable solution for credit card fraud detection.

Fully Automated Detection of the Appendix Using U-Net Deep Learning Architecture in CT Scans

Background: The accurate segmentation of the appendix with well-defined boundaries is critical for diagnosing conditions such as acute appendicitis. The manual identification of the appendix is time-consuming and highly dependent on the expertise of the radiologist. Method: In this study, we propose a fully automated approach to the detection of the appendix using deep learning architecture based on the U-Net with specific training parameters in CT scans. The proposed U-Net architecture is trained on an annotated original dataset of abdominal CT scans to segment the appendix efficiently and with high performance. In addition, to extend the training set, data augmentation techniques are applied for the created dataset. Results: In experimental studies, the proposed U-Net model is implemented using hyperparameter optimization and the performance of the model is evaluated using key metrics to measure diagnostic reliability. The trained U-Net model achieved the segmentation performance for the detection of the appendix in CT slices with a Dice Similarity Coefficient (DSC), Volumetric Overlap Error (VOE), Average Symmetric Surface Distance (ASSD), Hausdorff Distance 95 (HD95), Precision (PRE) and Recall (REC) of 85.94%, 23.29%, 1.24 mm, 5.43 mm, 86.83% and 86.62%, respectively. Moreover, our model outperforms other methods by leveraging the U-Net’s ability to capture spatial context through encoder–decoder structures and skip connections, providing a correct segmentation output. Conclusions: The proposed U-Net model showed reliable performance in segmenting the appendix region, with some limitations in cases where the appendix was close to other structures. These improvements highlight the potential of deep learning to significantly improve clinical outcomes in appendix detection.

Mechanism and Data-Driven Fusion SOC Estimation

An accurate assessment of the state of charge (SOC) of electric vehicle batteries is critical for implementing frequency regulation and peak shaving. This study proposes mechanism- and data-driven SOC fusion calculation methods. First, a second-order Thevenin battery model is developed to obtain the physical parameters of the battery. Second, data from the Thevenin battery model and data from four standard cycling conditions in the electric vehicle industry are added to the dataset of the feed-forward neural network data-driven model to construct the test and training sets of the data-driven model. Finally, the error of the mechanism and data-driven fusion modeling method is quantitatively analyzed by comparing the estimation error of the method for the battery SOC at different temperatures with the accuracy of the data-driven SOC estimation method. The simulation results show that the root mean square error, the mean age absolute error, and the maximum error of mechanism and data-driven method for the estimation error of battery SOC are lower than those of the data-driven method by 0.9%, 0.65%, and 1.3%, respectively. The results show that the mechanism and data-driven fusion SOC estimation method has better generalization performance and higher SOC estimation accuracy.

Bone density measurement in patients with spinal metastatic tumors using chest quantitative CT deep learning model

ObjectiveThis study aims to develop a deep learning model using the 3DResUNet architecture to predict vertebral volumetric bone mineral density (vBMD) from Quantitative Computed Tomography (QCT) scans in patients with spinal metastatic tumors, enhancing osteoporosis screening capabilities. Methods749 patients with spinal metastatic tumors underwent QCT vertebral vBMD measurements. The dataset was randomly split into training (599 cases) and test sets (150 cases). The 3DResUNet model was trained for vBMD classification and prediction using QCT images processed with automated bone segmentation and ROI extraction. ResultsThe deep learning model demonstrated strong performance with Spearman correlation coefficients of 0.923 (training set) and 0.918 (test set) between predicted and QCT-measured vBMD values. Bland-Altman analysis showed a slight bias of −1.42 mg/cm3 (training set) and −1.14 mg/cm3 (test set) between the model predictions and QCT measurements. The model achieved an area under the curve (AUC) of 0.977 (training set) and 0.966 (test set) for diagnosing Osteoporosis based on vBMD. ConclusionThe developed deep learning model using 3DResUNet effectively predicts vertebral vBMD from QCT scans in patients with spinal metastatic tumors. It provides accurate and automated vBMD measurements, potentially facilitating widespread osteoporosis screening in clinical practice, mainly where DXA availability is limited.

The Application of Deep Learning in Sports Competition Data Prediction

In order to predict sports competition data, the author needs to implement the structure and related processes of the relevant competition victory and defeat prediction system, and specifically introduce and plan the implementation of each functional module. The data collection and storage module adopts Alibaba Cloud servers and combines Python to remotely and automatically collect data on a scheduled basis, according to the actual situation of game wins and losses, data cleaning and filtering are carried out, and multiple encoding forms are used to vectorize the data in order to find the best model. The data is divided according to the standard training and testing sets, and multiple classifiers are used for model training and saved locally for direct use next time; Test the above model using the training set; Compare the advantages and disadvantages of each vectorized encoding and classifier based on the final performance evaluation module. Based on the relevant experimental results, a detailed analysis was conducted to compare the advantages and disadvantages of each model, proving that introducing word vectors (word embeddings) into the competition data analysis system is worthwhile. We have obtained an excellent performance prediction model with a highest accuracy P of 0.825, a recall R of 0.729, and a corresponding F1 value of 0.774. For a prediction model that only knows the initial lineup allocation as a prerequisite, this already has sufficient practical guidance significance.

Application of a Machine Learning Predictive Model for Recurrent Acute Pancreatitis.

Acute pancreatitis is the main cause of hospitalization for pancreatic disease. Some patients tend to have recurrent episodes after experiencing an episode of acute pancreatitis. This study aimed to construct predictive models for recurrent acute pancreatitis (RAP). A total of 531 patients who were hospitalized for the first episode of acute pancreatitis at the Affiliated Hospital of Southwest Medical University from January 2018 to December 2019 were enrolled in the study. We confirmed whether the patients had a second episode until December 31, 2021, through an electronic medical record system and telephone or WeChat follow-up. Clinical and follow-up data of patients were collected and randomly allocated to the training and test sets at a ratio of 7:3. The training set was used to select the best model, and the selected model was tested with the test set. The area under the receiver operating characteristic curves, sensitivity, specificity, positive predictive value, negative predictive value, accuracy, decision curve, and calibration plots were used to assess the efficacy of the models. Shapley additive explanation values were used to explain the model. Considering multiple indices, XGBoost was the best model. The area under the receiver operating characteristic curves, accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of the XGBoost model in the test set were 0.779, 0.763, 0.883, 0.647, 0.341, and 0.922, respectively. According to the Shapley additive explanation values, drinking, smoking, higher levels of triglyceride, and the occurrence of ANC are associated with RAP. The XGBoost model shows good performance in predicting RAP, which may help identify high-risk patients.