Machine Method Research Articles

Non-invasive Assessment of Human Epidermal Growth Factor Receptor 2 Expression in Gastric Cancer Based on Deep Learning: AComputed Tomography-based Multicenter Study.

The expression of human epidermal growth factor receptor 2 (HER2) in gastric cancer is closely associated with its treatment outcomes and prognosis. This study aims to develop and validate a HER2 prediction model based on computed tomography (CT). Additionally, the study evaluates the robustness of the proposed model. This retrospective study included 1059 patients from three hospitals (A, B, and C), where patients from hospitals A and B formed the training set (720 cases), and patients from hospital C served as the external test set (339 cases). Venous-phase CT radiomic features were extracted, normalized using the Z-score method, and simplified via principal component analysis. Feature selection was performed using recursive feature elimination (RFE), analysis of variance, Relief, and the Kruskal-Wallis (KW) test, followed by modeling using Lasso-regularized logistic regression and Support Vector Machine (SVM) methods. The models were evaluated and validated using the area under the curve (AUC) and decision curve analysis to determine the best-performing model. The positive proportions of HER2 expression were 8.60% (52/658) in the training set and 5.60% (19/320) in the test set. Eight distinct models were developed to predict HER2 expression. Among these, the model utilizing RFEand Lasso-regularized logistic regression (LR-Lasso) exhibited the highest predictive performance, with AUC values of 0.7874 (95% CI: 0.7346-0.8402) in the training set and 0.8033 (95% CI: 0.7288-0.8788) in the test set. Compared to other models, this model provided a greater net benefit on the decision curve analysis. These results suggest that the proposed model can be effectively applied to predict HER2 expression in patients. The HER2 prediction model demonstrated promising performance in predicting HER2 expression in gastric cancer patients.

Enhanced accuracy and stability in automated intra-pancreatic fat deposition monitoring of type 2 diabetes mellitus using Dixon MRI and deep learning.

Intra-pancreatic fat deposition (IPFD) is closely associated with the onset and progression of type 2 diabetes mellitus (T2DM). We aimed to develop an accurate and automated method for assessing IPFD on multi-echo Dixon MRI. In this retrospective study, 534 patients from two centers who underwent upper abdomen MRI and completed multi-echo and double-echo Dixon MRI were included. A pancreatic segmentation model was trained on double-echo Dixon water images using nnU-Net. Predicted masks were registered to the proton density fat fraction (PDFF) maps of the multi-echo Dixon sequence. Deep semantic segmentation feature-based radiomics (DSFR) and radiomics features were separately extracted on the PDFF maps and modeled using the support vector machine method with 5-fold cross-validation. The first deep learning radiomics (DLR) model was constructed to distinguish T2DM from non-diabetes and pre-diabetes by averaging the output scores of the DSFR and radiomics models. The second DLR model was then developed to distinguish pre-diabetes from non-diabetes. Two radiologist models were constructed based on the mean PDFF of three pancreatic regions of interest. The mean Dice similarity coefficient for pancreas segmentation was 0.958 in the total test cohort. The AUCs of the DLR and two radiologist models in distinguishing T2DM from non-diabetes and pre-diabetes were 0.868, 0.760, and 0.782 in the training cohort, and 0.741, 0.724, and 0.653 in the external test cohort, respectively. For distinguishing pre-diabetes from non-diabetes, the AUCs were 0.881, 0.688, and 0.688 in the training cohort, which included data combined from both centers. Testing was not conducted due to limited pre-diabetic patients. Intraclass correlation coefficients between radiologists' pancreatic PDFF measurements were 0.800 and 0.699 at two centers, suggesting good and moderate reproducibility, respectively. The DLR model demonstrated superior performance over radiologists, providing a more efficient, accurate and stable method for monitoring IPFD and predicting the risk of T2DM and pre-diabetes. This enables IPFD assessment to potentially serve as an early biomarker for T2DM, providing richer clinical information for disease progression and management.

P0338 Early identification of pediatric inflammatory bowel disease based on machine learning with non-invasive data

Abstract Background Early identification of pediatric inflammatory bowel disease (IBD) can improve long-term prognosis, yet significant diagnostic delays persist. This study aims to develop an effective early identification tool using machine learning with non-invasive testing. Methods This retrospective study collected data from 314 pediatric patients (109 with IBD and 205 without IBD) hospitalized in Fudan University Children’s Hospital, to serve as the training dataset. We developed machine learning classifiers using support vector machine, artificial neural network, extreme gradient boosting, decision tree, random forest, k-nearest neighbors, logistic model, and gradient boosting machine methods, incorporating easily obtainable clinical examination tools. Participant data included age, sex, and three groups of features across different dimensions: IBD symptoms (diarrhea lasting ≥1 month; blood in stool for ≥1 week; recurrent perianal abscesses or fistulas; delayed growth; abdominal pain lasting ≥1 month; >10% weight loss; first-degree family history; arthritis, uveitis, erythema nodosum without definitive rheumatologic diagnosis; recurrent aphthous ulcers; unexplained fever), inflammatory biomarkers (elevated fecal calprotectin; elevated serum inflammatory markers; anemia; hypoalbuminemia; ANCA positivity), and transabdominal bowel ultrasound parameters (Limberg level >1 indicating bowel wall thickening with vascularity; mesenteric fat wrapping; bowel wall stratification disorder or loss; lymphadenopathy). We evaluated the performance of each feature group and their combinations, optimized hyperparameters, internal validation, and performed sequential validation with 66 prospectively recruited pediatric patients from the same center. The final model was selected based on optimal and stable performance in both the development and external validation cohorts, with Shapley values used to interpret feature importance. Results The support vector machine outperformed other algorithms, achieving an area under the curve of 0.95 in both internal and external validation datasets. The five most important features based on Shapley values were Limberg level >1, elevated ESR, elevated fecal calprotectin, recurrent perianal disease, and elevated CRP. Additionally, combining all three feature dimensions outperformed any single or paired combination. The probability of IBD classification showed moderate correlation with clinical, endoscopic, and histologic scores . Conclusion The machine learning model based on non-invasive data has the potential to become a low-cost and time-efficient tool for early identification of IBD. Ultrasound examination plays a significant role in distinguishing children with IBD from those without.

Finite Element Analysis of the Excavation Stability of Deep and Large Ventilation Shafts of Zimuyan Tunnel Using the Raise Boring Machine Method in a Karst Area

Finite Element Analysis of the Excavation Stability of Deep and Large Ventilation Shafts of Zimuyan Tunnel Using the Raise Boring Machine Method in a Karst Area

Sentiment Analysis of the Influence of the Korean Wave in Indonesia using the Naive Bayes Method and Support Vector Machine

This study analyzes public sentiment towards the influence of the Korean wave in Indonesia using the Naive Bayes and Support Vector Machine (SVM) methods. The Korean wave, as a popular cultural phenomenon from South Korea, has had a significant influence on various aspects of Indonesian society. The dataset consists of 6,327 tweets obtained through a crawling process on social media X, with 80% data divided for training and 20% for testing. The pre-processing process includes cleaning, case folding, tokenizing, stopwords, and stemming. Data imbalance in sentiment distribution is overcome by the SMOTE technique. The test results show that the SVM model has the highest accuracy of 89%, outperforming the Naive Bayes model with an accuracy of 81%. Performance evaluation using precision, recall, and F1-score shows that SVM is more consistent in classifying positive and negative sentiments. Data visualization is done using bar charts and word clouds to illustrate the main patterns and themes in discussions related to the Korean wave in Indonesia. However, this study has limitations, such as data is only taken from one social media platform, so the results are less representative of public opinion as a whole. Nevertheless, this study provides new insights into how Indonesian society responds to popular culture phenomena online. These findings can also be utilized by policymakers to support the development of creative industries based on popular culture.

Hybridization in Metal Wire Additive Manufacturing: A Case Study of an Impeller

Additive manufacturing (AM) has evolved to enable the direct production of functional components through the hybridization of additive and subtractive processes. In metal wire AM, hybridization is key, encompassing process integration (addition/subtraction), energy source combinations (arc/laser), kinematic options (3/4/5 axis), and slicing techniques (planar/conformal). This paper focuses on these hybridization methods, with a unified system designed for single-machine setups, improving efficiency and accuracy. This study presents a detailed exploration of these hybridization levels through the fabrication of a complex 5-axis geometry—an impeller. The impeller was manufactured with hybridization using various levels and subsequently compared with manufacturing processes like additive manufacturing with interlayer machining and traditional machining methods. The hybrid approach significantly reduced the manufacturing time for the selected impeller geometry from 3536 min to 792 min (saving 77.6% manufacturing time) and minimized material waste to 9.3%, compared with 74.07% in traditional machining. This demonstrates a more efficient, precise, and cost-effective method to optimize metal wire AM for producing complex metal components, advancing capabilities and applications.

Milk Composition Is Predictive of Low Milk Supply Using Machine Learning Approaches.

Background/Objectives: The causes of low milk supply are multifactorial, including factors such as gene mutations, endocrine disorders, and infrequent milk removal. These factors affect the functional capacity of the mammary gland and, potentially, the concentrations of milk components. This study aimed to investigate the differences in milk composition between mothers with low and normal milk supply and develop predictive machine learning models for identifying low milk supply. Methods: Twenty-four-hour milk production measurements were conducted using the test-weigh method. An array of milk components was measured in 58 women with low milk supply (<600 mL/24 h) and 106 with normal milk supply (≥600 mL/24 h). Machine learning algorithms were employed to develop prediction models integrating milk composition and maternal and infant characteristics. Results: Among the six machine learning algorithms tested, deep learning and gradient boosting machines methods had the best performance metrics. The best-performing model, incorporating 14 milk components and maternal and infant characteristics, achieved an accuracy of 87.9%, an area under the precision-recall curve (AUPRC) of 0.893, and an area under the receiver operating characteristic curve (AUC) of 0.917. Additionally, a simplified model, optimised for clinical applicability, maintained a reasonable accuracy of 78.8%, an AUPRC of 0.776, and an AUC of 0.794. Conclusions: These findings demonstrate the potential of machine learning models to predict low milk supply with high accuracy. Integrating milk composition and maternal and infant characteristics offers a practical approach to identify women at risk of low milk supply, facilitating timely interventions to support breastfeeding and ensure adequate infant nutrition.

Experimental study on the influence of lapping plate materials on the quality of both sides cylindrical rollers machining

The both-sides machining method can obtain high precision cylindrical rollers, but there is a drawback that the lapping plate is easy to wear, which restricts further improvement in the quality of rollers machining. Aiming to solve this problem, a both-sides machining method using hard ceramic lapping plate is proposed. Friction and wear experiments with different lapping plate materials, along with the corresponding comparative machining experiments, demonstrated the superior performance of Al2O3 ceramic lapping plate in terms of roundness (0.23 μm), cylindricity (0.58 μm), straightness (0.489 μm), surface roughness (11.3 nm). By studying the influence of performance indicators, the optimal process parameters for cylindrical rollers machining were obtained. The durability of the Al2O3 ceramic lapping plate was also analyzed. The study fully confirm that the hard ceramic significantly enhances lapping plate wear resistance, machining precision and consistency, which provides a new approach to ultra-precision machining of cylindrical rollers.

Circuit Privacy for FHEW/TFHE-Style Fully Homomorphic Encryption in Practice

A fully homomorphic encryption (FHE) scheme allows a client to encrypt and delegate its data to a server that performs computation on the encrypted data that the client can then decrypt. While FHE gives confidentiality to clients' data, it does not protect the server's input and computation. Nevertheless, FHE schemes are still helpful in building delegation protocols that reduce communication complexity, as the ciphertext's size is independent of the size of the computation performed on them. We can further extend FHE by a property called circuit privacy, which guarantees that the result of computing on ciphertexts reveals no information on the computed function and the inputs of the server. Thereby, circuit private FHE gives rise to round optimal and communication efficient secure two-party computation protocols. Unfortunately, despite significant efforts and much work put into the efficiency and practical implementations of FHE schemes, very little has been done to provide useful and practical FHE supporting circuit privacy. In this work, we address this gap and design the first randomized bootstrapping algorithm whose single invocation sanitizes a ciphertext and, consequently, serves as a tool to provide circuit privacy. We give an extensive analysis, propose parameters, and provide a C++ implementation of our scheme. Our bootstrapping can sanitize a ciphertext to achieve circuit privacy at an 80-bit statistical security level in between 1.3 and 0.9 seconds, depending which Gaussian sampling algorithm is used, and whether the parameter set targets a fast Fourier or a number theoretic transform-based implementation. In addition, we can perform non-sanitized bootstrapping in around 0.27 or 0.14 seconds. Crucially, we do not need to increase the parameters to perform computation before or after sanitization takes place. For comparison's sake, we revisit the Ducas-Stehlé washing machine method. In particular, we give a tight analysis, estimate efficiency, review old, and provide new parameters.

Comparison of Extreme Learning Machine (ELM) and Multi-Support Vector Machine (Multi-SVM) Methods in Herbal Plants Identification

Comparison of Extreme Learning Machine (ELM) and Multi-Support Vector Machine (Multi-SVM) Methods in Herbal Plants Identification

Prediction of sentiment polarity in restaurant reviews using an ordinal regression approach based on evolutionary XGBoost

Prediction of sentiment polarity in restaurant reviews using an ordinal regression approach based on evolutionary XGBoost

Multi-Centered Pre-Treatment CT-Based Radiomics Features to Predict Locoregional Recurrence of Locally Advanced Esophageal Cancer After Definitive Chemoradiotherapy

Purpose: We constructed a prediction model to predict a 2-year locoregional recurrence based on the clinical features and radiomic features extracted from the machine learning method using computed tomography (CT) before definite chemoradiotherapy (dCRT) in locally advanced esophageal cancer. Patients and methods: A total of 264 patients (156 in Beijing, 87 in Tianjin, and 21 in Jiangsu) were included in this study. All those locally advanced esophageal cancer patients received definite radiotherapy and were randomly divided into five subgroups with a similar number and divided into training groups and validation groups by five cross-validations. The esophageal tumor and extratumoral esophagus were segmented to extract radiomic features from the gross tumor volume (GTV) drawn by radiation therapists before radiotherapy, and six clinical features associated with prognosis were added. T stage, N stage, M stage, total TNM stage, GTV, and GTVnd volume were included to construct a prediction model to predict the 2-year locoregional recurrence of patients after definitive radiotherapy. Results: A total of 264 patients were enrolled from August 2012 to April 2018, with a median age of 62 years and 81% were males. The 2-year locoregional recurrence rate was 52.6%, and the 2-year overall survival rate was 45.6%. About 66% of patients received concurrent chemotherapy. In total, we extracted 786 radiomic features from CT images and the Principal Component Analysis (PCA) method was used to screen out the maximum 30 features. Finally, the Support Vector Machine (SVM) method was used to construct the integrated prediction model combining radiomics and clinical features. In the five training groups for predicting locoregional recurrence, the mean value of C-index was 0.9841 (95%CI, 0.9809–0.9873), and in the five validation groups, the mean value was 0.744 (95%CI, 0.7437–0.7443). Conclusions: The integrated radiomics model could predict the 2-year locoregional recurrence after dCRT. The model showed promising results and could help guide treatment decisions by identifying high-risk patients and enabling strategies to prevent early recurrence.

Forecasting Real Estate Price via Support Vector Machine

This paper focus on the analysis of real estate price using support vector machine. The data set of a local estate price gives the details of the house including the price, date, bedroom numbers and postcode (location), and has been displayed as the line charts. There is comparison about different forecasting methods done by previous researchers including the simple moving average and ARIMA model. The data set is then fed to the model for machine learning and price prediction support vector machine. Moreover, R language is utilised to compute the support vector machine method by e1071 package and graph plotting of different kernel functions inside the support vector machine function. Then the graphs are analysed based on the accuracy of the correct estimation on the diagonal of the matrix generated. Possible improvements including the data collection and survey method are discussed for a higher precision and accuracy result for further study.

A control method of virtual DC machine based on bus power compensation

A control method of virtual DC machine based on bus power compensation

Artificial Intelligence Based Hybrid ASFO-ESVM for Load Demand Prediction in Micro Grid Energy Management

Predicting load demand is relevant when used in microgrid energy management systems to address issues such as nonlinear and dynamic consumption data. In this research, the author presents a fusion of Adaptive Sunflower Optimization (ASFO) and Enhanced Support Vector Machine (ESVM) methods to predict the load demand in micro grid environment. The ASFO algorithm enhances the efficiency of the ESVM through a fine-tuning meta-heuristic algorithm based on the sunflower natural organisms. This integration of ASFO and ESVM eliminates many of the drawbacks associated with the basic performance of the task, namely low speed of convergence, overtraining, and the presence of local minima in choosing the parameters. Some of the general parameters used in training and validating the model include load and meteorological data features involving, weather, temporal, load histories are the main contributors in the analysis. Comparisons with other ML algorithm ‘shave been made in respect of relative performance against established methods, such as Random Forest (RF) and Particle Swarm Optimization based with ESVM (PSO-ESVM). The findings infer that lower values of Mean Absolute Percentage Error (MAPE) and Symmetric Mean Absolute Percentage Error (SMAPE) and higher consistency index (d) are yielded by the proposed hybrid ASFO-ESVM model. For instance, even on working days of the week, the precision of the load forecasts was higher with the hybrid model than with the other options. The outcomes do prove that the proposed ASFO-ESVM model is very reliable and precise in its concerning aspect of load demand forecasting as it can be seen in the results obtained for different situations. Relatively, this work estimates a cost effective and feasible method for micro grid energy predictions which can enhance decisions in matters concerning power production, distribution, and control of energy. The study shows how these techniques are relevant towards the complexity and dynamism of the contemporary energy systems.

Sample selection using multi-task autoencoders in federated learning with non-IID data

Federated learning is a machine learning paradigm in which multiple devices collaboratively train a model under the supervision of a central server while ensuring data privacy. However, its performance is often hindered by redundant, malicious, or abnormal samples, leading to model degradation and inefficiency. To overcome these issues, we propose novel sample selection methods for image classification, employing a multi-task autoencoder to estimate sample contributions through loss and feature analysis. Our approach incorporates unsupervised outlier detection, using one-class support vector machine (OCSVM), isolation forest (IF), and adaptive loss threshold (AT) methods managed by a central server to filter noisy samples on clients. We also propose a multi-class deep support vector data description (SVDD) loss controlled by a central server to enhance feature-based sample selection. We validate our methods on CIFAR10 and MNIST datasets across varying numbers of clients, non-IID distributions, and noise levels up to 40%. The results show significant accuracy improvements with loss-based sample selection, achieving gains of up to 7.02% on CIFAR10 with OCSVM and 1.83% on MNIST with AT. Additionally, our federated SVDD loss further improves feature-based sample selection, yielding accuracy gains of up to 0.99% on CIFAR10 with OCSVM. These results show the effectiveness of our methods in improving model accuracy across various client counts and noise conditions.

Micromachining performance of additively manufactured titanium alloys in synergistic application of ultrasonic elliptical machining and textured tool methods

Micromachining performance of additively manufactured titanium alloys in synergistic application of ultrasonic elliptical machining and textured tool methods

Altered effective connectivity in Parkinson's disease patients with rapid eye movement sleep behavior disorder: a resting-state functional magnetic resonance imaging study and support vector machine analysis.

Rapid eye movement sleep behavior disorder (RBD) is associated with pathological α-synuclein deposition and may have different damage directions due to α-synuclein spreading orientations. Recent functional imaging studies of Parkinson's disease (PD) with RBD have identified abnormalities in connectivity, but effective connectivity (EC) for this altered orientation is understudied. Here, we aimed to explore altered intrinsic functional connectivity (FC) and EC in PD patients with probable RBD (pRBD). This was a cross-sectional study. A total of 31 PD patients with pRBD (PD-pRBD), 35 PD without pRBD (PD-npRBD), and 32 healthy controls (HCs) underwent resting-state functional magnetic resonance imaging (RS-fMRI) scans. The voxel-wise degree centrality (DC) calculation was first performed to investigate the inherent connectivity of the PD-pRBD patients. Subsequently, we applied Granger causality analysis (GCA) to probe the causal effects of anomalous brain regions. Finally, the support vector machine (SVM) method was executed to evaluate the DC values in identifying PD-pRBD. PD-pRBD patients exhibited reduced z-DC values in the right precentral gyrus relative to PD-npRBD (voxel-level P<0.001, cluster-level P<0.05), as well as decreased z-DC values in the right postcentral gyrus and the superior parietal lobule compared to HCs. Then, our GCA revealed that decreased EC was located predominantly from the right precentral gyrus to the right caudate nucleus in the PD-pRBD group. Additionally, the SVM results revealed that the z-DC values of the right precentral gyrus could discriminate PD-pRBD from the PD-npRBD group [area under the curve (AUC) =0.905]. The altered z-DC in the right precentral gyrus and the anomaly causal effects from the precentral motor cortex to the ipsilateral striatum represented by the caudate nucleus might play vital roles in the pathogenesis of PD-pRBD. It was speculated that the attenuation of FC from the precentral motor cortex to the subcortical striatum might be associated with nocturnal muscle dyskinesia and behavioral abnormalities in PD-pRBD patients. This disruption pattern may be a prospective imaging marker in the characterization of PD with pRBD.

Investigation & analysis of Electrical Discharge Machining on Titanium Diboride Ceramic material with various Electrode materials and dielectric fluids

Abstract Titanium diboride is a highly used material in aerospace industry due of its remarkable longevity. Its strength poses a problem to traditional machining methods. We want to improve the EDM process parameters in TiB2 material by comparing two different dielectric fluids and four different electrode materials. The input parameters were adjusted to find the optimal combination. By studying the fluctuations of output parameters such as MRR, TWR, SR and overcut when processing titanium diboride, researchers can gain valuable insights. This is achieved through experimental studies and the results are then verified using a theoretical regression model. Experimental results confirmed that discharge current, voltage and Ton had a remarkable influence on MRR, TWR and surface roughness in TiB2 material. It has been observed that tool wear rate when using kerosene-based dielectric fluids is higher than that of deionized water-based dielectric fluids. Tungsten copper electrodes cause slightly less tool wear than tungsten electrodes, and produce higher MRR overall, when compared to other three electrode materials. Using a dielectricand a copper electrode can help you obtain a smoother surface. When deionized water is used as adielectric fluid with tungsten electrodes, the crossover value drops dramatically. Brass electrodesproduce ahigher overcut value compared to the other three electrode materials. The microstructure of the EDM hole was distinguished by SEM analysis, and the material composition of TIB2 was determined by EDS analysis&amp;#xD;

Prediction of the Stability of Protein Substructures Using AI/ML Techniques.

This chapter explores the innovative application of machine learning techniques to understand and predict the stability of protein substructures. Accurately identifying stable substructures within proteins necessitates incorporating the local context, crucial for elucidating the roles of supersecondary structures. This approach emphasizes the importance of contextual information in understanding the stability and functionality of protein regions, thereby providing a more comprehensive view of protein mechanics and interactions. The chapter focuses on our findings regarding the DnaK Hsp70 chaperone protein, utilizing it as a case study. This research highlights how context-dependent physico-chemical features derived from protein sequences can accurately classify residues into stable and unstable substructures by leveraging logistic regression, random forest, and support vector machine methods. The findings represent a pivotal step towards the rational design of proteins with tailored properties, offering new insights into protein engineering and the fundamental principles underpinning protein supersecondary structures.