Ten-fold Cross-validation Research Articles

Fusion Learning from Non-contrast CT Scans for the Detection of Hemorrhagic Transformation in Stroke Patients.

Hemorrhagic transformation (HT) is a potentially catastrophic complication after acute ischemic stroke. Prevention of HT risk is crucial because it worsens prognosis and increases mortality. This study aimed at developing and validating a computer-aided diagnosis system using pretreatment non-contrast computed tomography (CT) scans for HT prediction in stroke patients undergoing revascularization. This retrospective study included all acute ischemic stroke patients with non-contrast CT before reperfusion therapy who also underwent follow-up MRI from January 2018 to December 2022. Among the 188 evaluated patients, any degree of HT at follow-up imaging was observed in 103 patients. HT diagnosis via MRI was defined as the reference standard for neuroradiologists. Using a database of 2076 serial non-contrast CT images of the brain, pretrained deep learning architectures such as convolutional neural networks and vision transformers (ViTs) were used for feature extraction. The performance of the predictive HT risk model was evaluated via tenfold cross-validation in machine learning classifiers. The accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) were evaluated. Using an individual deep learning architecture, DenseNet201 features achieved the highest accuracy of 87% and an AUC of 0.8863 in the classifier of the subspace ensemble k-nearest neighbor. By combining the DenseNet201 and ViT features, the accuracy and AUC can be improved to 88% and 0.8987, respectively, which are significantly better than those of using ViT alone. Detecting HT in stroke patients is a meaningful but challenging issue. On the basis of the model approach, HT diagnosis would be more automatic, efficient, and consistent, which would be helpful in clinic use.

Image analysis-based identification of high risk ER-positive, HER2-negative breast cancers

BackgroundBreast cancer subtypes Luminal A and Luminal B are classified by the expression of PAM50 genes and may benefit from different treatment strategies. Machine learning models based on H&E images may contain features associated with subtype, allowing early identification of tumors with higher risk of recurrence.MethodsH&E images (n = 630 ER+/HER2-breast cancers) were pixel-level segmented into epithelium and stroma. Convolutional neural network and multiple instance learning were used to extract image features from original and segmented images. Patient-level classification models were trained to discriminate Luminal A versus B image features in tenfold cross-validation, with or without grade adjustment. The best-performing visual classifier was incorporated into envisioned diagnostic protocols as an alternative to genomic testing (PAM50). The protocols were then compared in time-to-recurrence models.ResultsAmong ER+/HER2-tumors, the image-based protocol differentiated recurrence times with a hazard ratio (HR) of 2.81 (95% CI: 1.73–4.56), which was similar to the HR for PAM50 (2.66, 95% CI: 1.65–4.28). Grade adjustment did not improve subtype prediction accuracy, but did help balance sensitivity and specificity. Among high grade participants, sensitivity and specificity (0.734 and 0.474, respectively) became more similar (0.732 and 0.624, respectively) in grade-adjusted models. The original and epithelium-specific images had similar performance and highest accuracy, followed by stroma or binarized images showing only the epithelial-stromal interface.ConclusionsGiven low rates of genomic testing uptake nationally, image-based methods may help identify ER+/HER2-patients who could benefit from testing.

Comparative analysis of impact of classification algorithms on security and performance bug reports

Abstract Identification and classification of bugs, e.g., security and performance are a preemptive and fundamental practice which contributes to the development of secure and efficient software. Software Quality Assurance (SQA) needs to classify bugs into relevant categories, e.g., security and performance bugs since one type of bug may have a higher preference over another, thus facilitating software evolution and maintenance. In addition to classification, it would be ideal for the SQA manager to prioritize security and performance bugs based on the level of perseverance, severity, or impact to assign relevant developers whose expertise is aligned with the identification of such bugs, thus facilitating triaging. The aim of this research is to compare and analyze the prediction accuracy of machine learning algorithms, i.e., Artificial neural network (ANN), Support vector machine (SVM), Naïve Bayes (NB), Decision tree (DT), Logistic regression (LR), and K-nearest neighbor (KNN) to identify security and performance bugs from the bug repository. We first label the existing dataset from the Bugzilla repository with the help of a software security expert to train the algorithms. Our research type is explanatory, and our research method is controlled experimentation, in which the independent variable is prediction accuracy and the dependent variables are ANN, SVM, NB, DT, LR, and KNN. First, we applied preprocessing, Term Frequency-Inverse Document Frequency feature extraction methods, and then applied classification algorithms. The results were measured through accuracy, precision, recall, and F-measure and then the results were compared and validated through the ten-fold cross-validation technique. Comparative analysis reveals that two algorithms (SVM and LR) perform better in terms of precision (0.99) for performance bugs and three algorithms (SVM, ANN, and LR) perform better in terms of F1 score for security bugs as compared to other classification algorithms which are essentially due to the linear dataset and extensive number of features in the dataset.

Predicting Tumor Progression in Patients with Cervical Cancer Using Computer Tomography Radiomic Features

The objective of this study was to evaluate the effectiveness of utilizing radiomic features from radiation planning computed tomography (CT) scans in predicting tumor progression among patients with cervical cancers. A retrospective analysis was conducted on individuals who underwent radiotherapy for cervical cancer between 2015 and 2020, utilizing an institutional database. Radiomic features, encompassing first-order statistical, morphological, Gray-Level Co-Occurrence Matrix (GLCM), Gray-Level Run Length Matrix (GLRLM), and Gray-Level Dependence Matrix (GLDM) features, were extracted from the primary cervical tumor on the CT scans. The study encompassed 112 CT scans from patients with varying stages of cervical cancer ((FIGO Staging of Cervical Cancer 2018): 24% at stage I, 47% at stage II, 21% at stage III, and 10% at stage IV). Of these, 31% (n = 35/112) exhibited tumor progression. Univariate feature analysis identified three morphological features that displayed statistically significant differences (p < 0.05) between patients with and without progression. Combining these features enabled a classification model to be developed with a mean sensitivity, specificity, accuracy, and AUC of 76.1% (CI 1.5%), 70.4% (CI 4.1%), 73.6% (CI 2.1%), and 0.794 (CI 0.029), respectively, employing nested ten-fold cross-validation. This research highlights the potential of CT radiomic models in predicting post-radiotherapy tumor progression, offering a promising approach for tailoring personalized treatment decisions in cervical cancer.

Detection of basal cell carcinoma by machine learning-assisted ex vivo confocal laser scanning microscopy.

Ex vivo confocal laser scanning microscopy (EVCM) is an emerging imaging modality that enables near real-time histology of whole tissue samples. However, the adoption of EVCM into clinical routine is partly limited because the recognition of modality-specific diagnostic features requires specialized training. Therefore, we aimed to build a machine learning algorithm for the detection of basal cell carcinoma (BCC) in images acquired using EVCM and, in turn, facilitate the examiner's decision-making process. In this proof-of-concept study, histologically confirmed BCC fresh tissue samples were used to generate 50 EVCM images to train and assess a convolutional neural network architecture (MobileNet-V1) via tenfold cross-validation. Overall sensitivity and specificity of the model for detecting BCC and tumor-free regions on EVCM images compared to expert evaluation were 0.88 and 0.85, respectively. We constructed receiver operator characteristic and precision-recall curves from the aggregated tenfold cross-validation to assess the model's performance; the area under the curve was 0.94 and 0.87, respectively. Subsequently, the performance of one of the selected machine learning models was assessed with 19 new EVCM images of tumor-containing (n = 10) and 9 tumor-free (n = 9) skin tissue. A sensitivity of 0.83 and a specificity of 0.92 were achieved for the BCC group. The specificity for the tumor-free control group was 0.98. The deep learning model developed in our study holds the potential to assist the diagnostic decision-making process and diminish the training time of novices by depicting tumor-positive regions in EVCM images.

Flexural capacity prediction of reinforced UHPC beams using an interpretable machine learning model

Addressing the issues of inconsistency and limited accuracy in the current prediction models for the flexural capacity of reinforced ultrahigh-performance concrete (UHPC) beams, this paper proposes a machine-learning prediction model for the flexural capacity of reinforced UHPC beams based on interpretable Extreme Gradient Boosting (XGBoost) model. Firstly, a database containing 221 sets of experimental UHPC beam data is established, and the database quality is evaluated by Pearson correlation coefficient and Mahalanobis distance. Then, ten-fold cross-validation and the whale optimization algorithm (WOA) are employed to find the optimal hyperparameter combination for XGBoost. Different indices are used to evaluate the prediction accuracy of the XGBoost model with four typical machine learning models and other computational methods. The Shapley additive explanations (SHAP) method is used to provide interpretations of the prediction results. The results indicate that the XGBoost model outperforms the four other machine learning models and related standards. Compared with the existing formulas, the proposed prediction model demonstrates higher accuracy in predicting the flexural capacity of UHPC beams. The SHAP method effectively explains the prediction results of the XGBoost model, providing insights into accurately assessing the factors influencing the flexural capacity of UHPC beams. SHAP analysis reveals that the most significant factors affecting the flexural capacity of UHPC beams are the cross-sectional height and reinforcement ratio, while the least significant factors are the aspect ratio of steel fibers and volume fraction of steel fibers.

Enhancing mortality prediction in patients with spontaneous intracerebral hemorrhage: Radiomics and supervised machine learning on non-contrast computed tomography

PurposeThis study aims to develop a Radiomics-based Supervised Machine-Learning model to predict mortality in patients with spontaneous intracerebral hemorrhage (sICH). MethodsRetrospective analysis of a prospectively collected clinical registry of patients with sICH consecutively admitted at a single academic comprehensive stroke center between January-2016 and April-2018. We conducted an in-depth analysis of 105 radiomic features extracted from 105 patients. Following the identification and handling of missing values, radiomics values were scaled to 0-1 to train different classifiers. The sample was split into 80-20% training-test and validation cohort in a stratified fashion. Random Forest(RF), K-Nearest Neighbor(KNN), and Support Vector Machine(SVM) classifiers were evaluated, along with several feature selection methods and hyperparameter optimization strategies, to classify the binary outcome of mortality or survival during hospital admission. A tenfold stratified cross-validation method was used to train the models, and average metrics were calculated. ResultsRF, KNN, and SVM, with the "DropOut+SelectKBest" feature selection strategy and no hyperparameter optimization, demonstrated the best performances with the least number of radiomic features and the most simplified models, achieving a sensitivity range between 0.90 and 0.95 and AUC range from 0.97 to 1 on the validation dataset. Regarding the confusion matrix, the SVM model did not predict any false negative test (negative predicted value 1). ConclusionRadiomics-based Supervised Machine Learning models can predictmortality during admission in patients with sICH. SVM with the "DropOut+SelectKBest"feature selection strategy and no hyperparameter optimization was the best simplifiedmodel to detect mortality during admission in patients with sICH.

A clinical-radiomics nomogram based on automated segmentation of chest CT to discriminate PRISm and COPD patients

PurposeIt is vital to develop noninvasive approaches with high accuracy to discriminate the preserved ratio impaired spirometry (PRISm) group from the chronic obstructive pulmonary disease (COPD) groups. Radiomics has emerged as an image analysis technique. This study aims to develop and confirm the new radiomics-based noninvasive approach to discriminate these two groups. MethodsTotally 1066 subjects from 4 centers were included in this retrospective research, and classified into training, internal validation or external validation sets. The chest computed tomography (CT) images were segmented by the fully automated deep learning segmentation algorithm (Unet231) for radiomics feature extraction. We established the radiomics signature (Rad-score) using the least absolute shrinkage and selection operator algorithm, then conducted ten-fold cross-validation using the training set. Last, we constructed a radiomics signature by incorporating independent risk factors using the multivariate logistic regression model. Model performance was evaluated by receiver operating characteristic (ROC) curve, calibration curve, and decision curve analyses (DCA). ResultsThe Rad-score, including 15 radiomic features in whole-lung region, which was suitable for diffuse lung diseases, was demonstrated to be effective for discriminating between PRISm and COPD. Its diagnostic accuracy was improved through integrating Rad-score with a clinical model, and the area under the ROC (AUC) were 0.82(95 %CI 0.79–0.86), 0.77(95 %CI 0.72–0.83) and 0.841(95 %CI 0.78–0.91) for training, internal validation and external validation sets, respectively. As revealed by analysis, radiomics nomogram showed good fit and superior clinical utility. ConclusionsThe present work constructed the new radiomics-based nomogram and verified its reliability for discriminating between PRISm and COPD.

Combining MRI radiomics and clinical features for early identification of drug-resistant epilepsy in people with newly diagnosed epilepsy

ObjectiveTo identify newly diagnosed patients with drug-resistant epilepsy (DRE) based on radiomics and clinical features. MethodsA radiomics approach was used to combine clinical features with magnetic resonance imaging (MRI) features extracted by the ResNet-18 deep learning model to predict DRE. Three machine learning classifiers were built, and k-fold cross-validation was used to assess the classifier outcomes, and other evaluation metrics of accuracy, sensitivity, specificity, F1 score, and area under the curve (AUC) were used to evaluate the performance of these models. ResultsOne hundred and thirty-four newly diagnosed epilepsy patients with 13 available clinical features and 1394 MRI features extracted by the ResNet-18 model were included in our study. Then three machine learning classifiers were built based on5 clinical features and 8 MRI features, including Support Vector Machine (SVM), Gradient-Boosted Decision Tree (GBDT) and Random Forest. After internally validation, the GBDT model performed the best, with an average accuracy of 0.85 [95% confidence interval (CI) 0.77–0.91], sensitivity of 0.97 [95% CI 0.85–1.00], specificity of 0.96 [95% CI 0.83–1.00], F1 score of 0.81 [95% CI 0.77–0.89], AUC of 0.95 [95% CI 0.82–0.99], and ten-fold cross validation avg score of 0.96 [95% CI 0.89–0.99] in test set. SignificanceThis study offers a novel approach for early diagnosis of DRE. Radiomics can provide potential diagnostic and predictive information to support personalized treatment decisions.

Investigation and application of data balancing and combined discriminant model in rock burst severity prediction

In the development of intelligent rock burst prediction models, issues such as incomplete data coverage and data imbalance are frequently encountered. These issues may lead to risks of overfitting in predictive models, poor generalization capabilities, and increased bias, which in turn may result in misjudgments and unpredictable losses. To accurately predict rock burst disasters and mitigate or eliminate related threats, this paper proposes a composite prediction model that integrates Density-Based Nonlinear Resampling (DBNR)-Tomek Link data balancing algorithms with Bayesian Optimization (BO)-Multilayer Perceptron (MLP)-Random Forest (RF). Initially, this study collected and organized a total of 301 recorded rock burst disaster field observation data, covering various tectonic plates, engineering types, rock origins, rock textures, and rock burst types. Subsequently, from a data analysis perspective, we employed the PCA-SSA-K-means unsupervised clustering algorithm to delve into the underlying information contained within the data, thereby validating the rationality of categorizing rock bursts into four grades. Then, using the L2 norm to optimize the dimensionality of the indicators and supplementing with indicator importance ranking and hypothesis testing, we selected the maximum tangential stress of the surrounding rock, the ratio of the maximum tangential stress of the surrounding rock to the uniaxial compressive strength of the rock (stress coefficient), and the elastic energy index as the criteria for rock burst intensity grading. Following that, the DBNR-Tomek Link sampling method was applied to balance the sample data, optimizing the data sample ratio and ultimately expanding the sample size to 396, improving the proportion of data samples from 2:3:4:1 to 1:1:2:1, thereby enhancing the model’s generalization performance. Ultimately, a BO-MLP-RF composite prediction model was constructed based on Bayesian Optimization (BO), Multilayer Perceptron (MLP), and Random Forest (RF) algorithms, with the Bayesian Optimization method ensuring that the model fits the training data well and generalizes to the test data. The results of tenfold cross-validation demonstrated that the model’s accuracy is consistently around 92.5%, combining the training results of rock burst models with imbalanced datasets, proving that the MLP model, adept at modeling nonlinear data, and the RF model, skilled in modeling large-scale data, serve as basic classifiers. This demonstrates that the application of data balancing and combined discriminative model schemes have enhanced the model’s predictive performance and stability. The model is capable of providing high-accuracy, high-efficiency early warning monitoring services for rock burst phenomena in rock engineering, thereby ensuring engineering safety.

Quartile Regression and Ensemble Models for Extreme Events of Multi-Time Step-Ahead Monthly Reservoir Inflow Forecasting

Amidst changing climatic conditions, accurately predicting reservoir inflows in an extreme event is challenging and inevitable for reservoir management. This study proposed an innovative strategy under such circumstances through rigorous experimentation and investigations using 18 years of monthly data collected from the Huai Nam Sai reservoir in the southern region of Thailand. The study employed a two-step approach: (1) isolating extreme and normal events using quantile regression (QR) at the 75th, 80th, and 90th quantiles and (2) comparing the forecasting performance of individual machine learning models and their combinations, including Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), and Multiple Linear Regression (MLR). Forecasting accuracy was assessed at four lead times—3, 6, 9, and 12 months—using ten-fold cross-validation, resulting in 16 model configurations for each forecast period. The results show that combining quantile regression (QR) to distinguish between extreme and normal events with hybrid models significantly improves the accuracy of monthly reservoir inflow forecasting, except for the 9-month lead time, where the XG model continues to deliver the best performance. The top-performing models, based on normalized scores for 3-, 6-, 9-, and 12-month-ahead forecasts, are XG-MLR-75, RF-XG-80, XG-75, and XG-RF-75, respectively. Another crucial finding of this research is the uneven decline in prediction accuracy as lead time increases. Notably, the model performed best at t + 9, followed by t + 3, t + 12, and t + 6, respectively. This pattern is influenced by model characteristics, error propagation, temporal variability, data dynamics, and seasonal effects. Improving the accuracy and efficiency of hybrid model forecasting can greatly enhance hydrological operational planning and management.

Intratumoral and peritumoral MRI-based radiomics for predicting extrapelvic peritoneal metastasis in epithelial ovarian cancer

ObjectivesTo investigate the potential of intratumoral and peritumoral radiomics derived from T2-weighted MRI to preoperatively predict extrapelvic peritoneal metastasis (EPM) in patients with epithelial ovarian cancer (EOC).MethodsIn this retrospective study, 488 patients from four centers were enrolled and divided into training (n = 245), internal test (n = 105), and external test (n = 138) sets. Intratumoral and peritumoral models were constructed based on radiomics features extracted from the corresponding regions. A combined intratumoral and peritumoral model was developed via a feature-level fusion. An ensemble model was created by integrating this combined model with specific independent clinical predictors. The robustness and generalizability of these models were assessed using tenfold cross-validation and both internal and external testing. Model performance was evaluated by the area under the receiver operating characteristic curve (AUC). The Shapley Additive Explanation method was employed for model interpretation.ResultsThe ensemble model showed superior performance across the tenfold cross-validation, with the highest mean AUC of 0.844 ± 0.063. On the internal test set, the peritumoral and ensemble models significantly outperformed the intratumoral model (AUC = 0.786 and 0.832 vs. 0.652, p = 0.007 and p < 0.001, respectively). On the external test set, the AUC of the ensemble model significantly exceeded those of the intratumoral and peritumoral models (0.843 vs. 0.750 and 0.789, p = 0.008 and 0.047, respectively).ConclusionPeritumoral radiomics provide more informative insights about EPM than intratumoral radiomics. The ensemble model based on MRI has the potential to preoperatively predict EPM in EOC patients.Critical relevance statementIntegrating both intratumoral and peritumoral radiomics information based on MRI with clinical characteristics is a promising noninvasive method to predict EPM to guide preoperative clinical decision-making for EOC patients.Key PointsPeritumoral radiomics can provide valuable information about extrapelvic peritoneal metastasis in epithelial ovarian cancer.The ensemble model demonstrated satisfactory performance in predicting extrapelvic peritoneal metastasis.Combining intratumoral and peritumoral MRI radiomics contributes to clinical decision-making in epithelial ovarian cancer.Graphical

Rain detection for rain-contaminated ground-based microwave radiometer data using physics-informed machine learning method

Because the radiation signal is strongly influenced by emission and scattering from rain, microwave radiometer data suffer from rain contamination. The traditional method of using rain gauges to detect rain for microwave radiometers has limitations. For example, it can only detect rain that reaches the ground and is ineffective for raindrops suspended in the atmosphere that can still contaminate remote sensing data. This article presents a rain detection method for microwave radiometer measurements, based on Gradient Boosted Decision Trees (GBDT). First, the characteristic that the increase in microwave radiometer brightness temperature when raindrops are present in the atmosphere, along with the seasonal dependency of rainfall patterns, is combined with meteorological variables to form feature vectors. Then, the GBDT is employed to classify data into rain-free and rain-contaminated categories. Microwave radiometer (MWR) measurements and simultaneous Micro Rain Radar (MRR) target classification collected from the Swiss Plateau in 2008 are utilized to train the model, which is subsequently tested using two testing schemes: ten-fold cross-validation technique and time series test sets. Compared with the detection accuracy of the integrated liquid water (ILW) threshold method (73.6% and 68.3%) in both testing schemes, our GBDT-based method achieved superior accuracy, recording approximately 100% and 98.4%, respectively. The proposed method exhibits strong generalization capabilities, allowing it to directly detect rain contamination in time series data and effectively overcome the time dependence of rainfall occurrence. In addition, compared with the ILW threshold method, the GBDT-based method considers various rainfall patterns contained in various seasons. Features selected for this method enable its direct application to other tropospheric microwave radiometer systems.

Advancing pharmacogenomics research: automated extraction of insights from PubMed using SpaCy NLP framework.

This paper presents a methodology for automatically extracting insights from PubMed articles using a Natural Language Processing (NLP) framework. Our approach, leveraging advanced NLP techniques and Named Entity Recognition (NER), is crucial for advancing pharmacogenomics and other scientific fields that benefit from streamlined access to literature through automated services like RESTful APIs.Building a new NLP model presents several challenges. First, it is essential to have a thorough understanding of the field in order to define relevant entities. Second, the construction of a diverse and consistent set of examples is crucial. Finally, the effective utilization of pre-established models is of paramount importance, as demonstrated in this work.Our model, validated via ten-fold cross-validation, achieved over 70% recall and precision for all entities in the training set. We provide a reproducible pipeline for the scientific community and propose a structured approach for qualitative analysis and clustering of results. This methodology refines literature reviews, optimizes knowledge extraction, and supports broader application across diverse research domains. An online platform could further extend these benefits to researchers, educators, and practitioners.

Optimizing Machine Learning Models with Data-level Approximate Computing: The Role of Diverse Sampling, Precision Scaling, Quantization and Feature Selection Strategies

Efficiency, low-power consumption, and real-time processing in embedded machine learning implementations are critical, particularly for models deployed in environments with large-scale data processing and resource-constrained environments. This paper investigates the application of approximate computing techniques as a viable solution to reduce computational complexity and optimize machine learning models, focusing on two widely used supervised machine learning models: k-nearest neighbors (KNN) and support vector machines (SVM). Although many studies compare machine learning classification techniques, the combined use of optimization strategies remains underexplored. Specifically, the combined utilization of feature selection, sampling, quantization, precision scaling, and relaxation methods for the purpose of optimizing and acquiring training and validation data is underexplored, particularly within the context of medical diagnosis datasets. In this paper, we propose a framework that uses data-level approximate computing techniques, including by diverse sampling strategies, precision scaling, quantization, and feature selection methods, to evaluate the impact of these techniques on the computational efficiency and accuracy of KNN and SVM models. Experimental results demonstrate that with careful application of approximate computing strategies, especially in critical applications such as medical diagnosis, it is possible to achieve considerable gains in efficiency while maintaining acceptable levels of accuracy. The combined application of these methods by selecting 3 features and quantizing the data values to 8 levels, then applying random sampling with 30% reductions and scaling the precision at 5 bits resulted in reductions of 87.5% in computation, 76.9% in memory usage, and 17% in delay, without any degradation in accuracy, as validated by tenfold cross-validation, Training Data validation, and full dataset validation. This study confirms the potential of approximate computing to optimize machine learning workflows, making it particularly suitable for applications with limited computational resources. The source code is publicly available online https://github.com/AyadMDalloo/DatalvlAxC.

Evaluation of machine learning and deep learning models for daily air quality index prediction in Delhi city, India.

The air quality index (AQI), based on criteria for air contaminants, is defined to provide a shared vision of air quality. As air pollution continues to rise in global cities due to urbanization and climate change, air pollution monitoring and forecasting models for effective air quality monitoring that gather and forecast information about air pollution concentration are essential in every city. Air quality predictions have evolved to be more helpful for management. Recently, better performance and ability have developed due to the involvement of machine learning (ML) and artificial intelligence (AI) in forecasting air quality in urban cities in India. This paper focuses on air pollution as a significant ecological problem that directly impacts human health and the distribution of an environmental system in urban areas. Hence, we have developed advanced models for daily AQI forecasting to understand the air effluence level in the upcoming days. In this research, six data-driven models have been developed and implemented for daily AQI forecasting in the study area; it is crucial for understanding the future air pollution levels to plan and control air pollution in the entire city. The developed model is applied to air quality datasets. A comparison of the performance of ML models tested here indicates that the XGBoost algorithm achieves the highest coefficient of determination (R2) and root-mean-square deviation (RMSE) value of 0.99 and lower values value of 4.65 than other models in the testing phase. The results of the artificial neural network (ANN) algorithm are slightly lower than the extreme gradient boosting (XGBoost model); the ANN model results are as R2, mean squared error (MSE), and RMSE values of 0.99, 13.99, and 198.88, respectively. All the models were subjected to a ten-fold cross-validation model. However, the RF cross-validation model outperforms other models; the RF model result shows the R2, RMSE, and MSE values of 0.99, 3.64, and 4.12, respectively. This study also employed two interpretable models, namely feature importance analysis and Shapley additive explanation (SHAP), to evaluate both the global and local methods in a manner that is independent of specific ML models. The feature importance shows that particle matter (PM) 2.5, PM10, carbon monoxide (CO), and nitrogen oxides (NOx) were the most influential variables. The results determined that such novel DL and ML models may improve the accuracy of AQI forecasts and understanding of air pollution, particularly in metropolitan cities.

Predicting intraoperative 5-ALA-induced tumor fluorescence via MRI and deep learning in gliomas with radiographic lower-grade characteristics.

Lower-grade gliomas typically exhibit 5-aminolevulinic acid (5-ALA)-induced fluorescence in only 20-30% of cases, a rate that can be increased by doubling the administered dose of 5-ALA. Fluorescence can depict anaplastic foci, which can be precisely sampled to avoid undergrading. We aimed to analyze whether a deep learning model could predict intraoperative fluorescence based on preoperative magnetic resonance imaging (MRI). We evaluated a cohort of 163 glioma patients categorized intraoperatively as fluorescent (n = 83) or non-fluorescent (n = 80). The preoperative MR images of gliomas lacking high-grade characteristics (e.g., necrosis or irregular ring contrast-enhancement) consisted of T1, T1-post gadolinium, and FLAIR sequences. The preprocessed MRIs were fed into an encoder-decoder convolutional neural network (U-Net), pre-trained for tumor segmentation using those three MRI sequences. We used the outputs of the bottleneck layer of the U-Net in the Variational Autoencoder (VAE) as features for classification. We identified and utilized the most effective features in a Random Forest classifier using the principal component analysis (PCA) and the partial least square discriminant analysis (PLS-DA) algorithms. We evaluated the performance of the classifier using a tenfold cross-validation procedure. Our proposed approach's performance was assessed using mean balanced accuracy, mean sensitivity, and mean specificity. The optimal results were obtained by employing top-performing features selected by PCA, resulting in a mean balanced accuracy of 80% and mean sensitivity and specificity of 84% and 76%, respectively. Our findings highlight the potential of a U-Net model, coupled with a Random Forest classifier, for pre-operative prediction of intraoperative fluorescence. We achieved high accuracy using the features extracted by the U-Net model pre-trained for brain tumor segmentation. While the model can still be improved, it has the potential for evaluating when to administer 5-ALA to gliomas lacking typical high-grade radiographic features.

Construction and SHAP interpretability analysis of a risk prediction model for feeding intolerance in preterm newborns based on machine learning

ObjectiveTo construct a highly accurate and interpretable feeding intolerance (FI) risk prediction model for preterm newborns based on machine learning (ML) to assist medical staff in clinical diagnosis.MethodsIn this study, a sample of 350 hospitalized preterm newborns were retrospectively analysed. First, dual feature selection was conducted to identify important feature variables for model construction. Second, ML models were constructed based on the logistic regression (LR), decision tree (DT), support vector machine (SVM) and eXtreme Gradient Boosting (XGBoost) algorithms, after which random sampling and tenfold cross-validation were separately used to evaluate and compare these models and identify the optimal model. Finally, we apply the SHapley Additive exPlanation (SHAP) interpretable framework to analyse the decision-making principles of the optimal model and expound upon the important factors affecting FI in preterm newborns and their modes of action.ResultsThe accuracy of XGBoost was 87.62%, and the area under the curve (AUC) was 92.2%. After the application of tenfold cross-validation, the accuracy was 83.43%, and the AUC was 89.45%, which was significantly better than those of the other models. Analysis of the XGBoost model with the SHAP interpretable framework showed that a history of resuscitation, use of probiotics, milk opening time, interval between two stools and gestational age were the main factors affecting the occurrence of FI in preterm newborns, yielding importance scores of 0.632, 0.407, 0.313, 0.313, and 0.258, respectively. A history of resuscitation, first milk opening time ≥ 24 h and interval between stools ≥ 3 days were risk factors for FI, while the use of probiotics and gestational age ≥ 34 weeks were protective factors against FI in preterm newborns.ConclusionsIn practice, we should improve perinatal care and obstetrics with the aim of reducing the occurrence of hypoxia and preterm delivery. When feeding, early milk opening, the use of probiotics, the stimulation of defecation and other measures should be implemented with the aim of reducing the occurrence of FI.

Computational fluid dynamics and shape analysis enhance aneurysm rupture risk stratification.

Accurately quantifying the rupture risk of unruptured intracranial aneurysms (UIAs) is crucial for guiding treatment decisions and remains an unmet clinical challenge. Computational Flow Dynamics and morphological measurements have been shown to differ between ruptured and unruptured aneurysms. It is not clear if these provide any additional information above routinely available clinical observations or not. Therefore, this study investigates whether incorporating image-derived features into the established PHASES score can improve the classification of aneurysm rupture status. A cross-sectional dataset of 170 patients (78 with ruptured aneurysm) was used. Computational fluid dynamics (CFD) and shape analysis were performed on patients' images to extract additional features. These derived features were combined with PHASES variables to develop five ridge constrained logistic regression models for classifying the aneurysm rupture status. Correlation analysis and principal component analysis were employed for image-derived feature reduction. The dataset was split into training and validation subsets, and a ten-fold cross validation strategy with grid search optimisation and bootstrap resampling was adopted for determining the models' coefficients. Models' performances were evaluated using the area under the receiver operating characteristic curve (AUC). The logistic regression model based solely on PHASES achieved AUC of 0.63. All models incorporating derived features from CFD and shape analysis demonstrated improved performance, reaching an AUC of 0.71. Non-sphericity index (shape variable) and maximum oscillatory shear index (CFD variable) were the strongest predictors of a ruptured status. This study demonstrates the benefits of integrating image-based fluid dynamics and shape analysis with clinical data for improving the classification accuracy of aneurysm rupture status. Further evaluation using longitudinal data is needed to assess the potential for clinical integration.

MRI Radiomics-Based Machine Learning to Predict Lymphovascular Invasion of HER2-Positive Breast Cancer.

This study aims to develop and prospectively validate radiomic models based on MRI to predict lymphovascular invasion (LVI) status in patients with HER2-positive breast cancer. A total of 225 patients with HER2-positive breast cancer who preoperatively underwent breast MRI were selected, forming the training set (n = 99 LVI-positive, n = 126 LVI-negative). A prospective validation cohort included 130 patients with breast cancer from the Affiliated Zhongshan Hospital of Dalian University (n = 57 LVI-positive, n = 73 LVI-negative). A total of 390 radiomic features and eight conventional radiological characteristics were extracted. For the optimum feature selection phase, the LASSO regression model with tenfold cross-validation (CV) was employed to identify features with non-zero coefficients. The conventional radiological (CR) model was determined based on visual morphological (VM) features and the optimal radiomic features correlated with LVI, identified through multivariate logistic analyses. Subsequently, various machine learning (ML) models were developed using algorithms such as support vector machine (SVM), k-nearest neighbor (KNN), gradient boosting machine (GBM), and random forest (RF). The performance of ML and CR models. The results show that the AUC of the CR model in the training and validation sets were 0.81 (95% confidence interval [CI], 0.74-0.86) and 0.82 (95% CI, 0.69-0.89), respectively. The ML model achieved the best performance, with AUCs of 0.96 (95% CI, 0.99-1.00) in the training set and 0.95 (95% CI, 0.89-0.96) in the validation set. There were significant differences between the CR and ML models in predicting LVI status. Our study demonstrated that the machine learning models exhibited superior performance in predicting LVI status based on pretreatment MRI compared to the CR model, which does not necessarily rely on a priori knowledge of visual morphology.