Training Set Research Articles

Construction and evaluation of a risk model for adverse outcomes of necrotizing enterocolitis based on LASSO-Cox regression

ObjectiveThis study aimed to develop a nomogram to predict adverse outcomes in neonates with necrotizing enterocolitis (NEC).MethodsIn this retrospective study on neonates with NEC, data on perinatal characteristics, clinical features, laboratory findings, and x-ray examinations were collected for the included patients. A risk model and its nomogram were developed using the least absolute shrinkage and selection operator (LASSO) Cox regression analyses.ResultsA total of 182 cases of NEC were included and divided into a training set (148 cases) and a temporal validation set (34 cases). Eight features, including weight [p = 0.471, HR = 0.99 (95% CI: 0.98–1.00)], history of congenital heart disease [p < 0.001, HR = 3.13 (95% CI:1.75–5.61)], blood transfusion before onset [p = 0.757, HR = 0.85 (95%CI:0.29–2.45)], antibiotic exposure before onset [p = 0.003, HR = 5.52 (95% CI:1.81–16.83)], C-reactive protein (CRP) at onset [p = 0.757, HR = 1.01 (95%CI:1.00–1.02)], plasma sodium at onset [p < 0.001, HR = 4.73 (95%CI:2.61–8.59)], dynamic abdominal x-ray score change [p = 0.001, HR = 4.90 (95%CI:2.69–8.93)], and antibiotic treatment regimen [p = 0.250, HR = 1.83 (0.65–5.15)], were ultimately selected for model building. The C-index for the predictive model was 0.850 (95% CI: 0.804–0.897) for the training set and 0.7880.788 (95% CI: 0.656–0.921) for the validation set. The area under the ROC curve (AUC) at 8-, 10-, and 12-days were 0.889 (95% CI: 0.822–0.956), 0.891 (95% CI: 0.829–0.953), and 0.893 (95% CI:0.832–0.954) in the training group, and 0.812 (95% CI: 0.633–0.991), 0.846 (95% CI: 0.695–0.998), and 0.798 (95%CI: 0.623–0.973) in the validation group, respectively. Calibration curves showed good concordance between the predicted and observed outcomes, and DCA demonstrated adequate clinical benefit.ConclusionsThe LASSO-Cox model effectively identifies NEC neonates at high risk of adverse outcomes across all time points. Notably, at earlier time points (such as the 8-day mark), the model also demonstrates strong predictive performance, facilitating the early prediction of adverse outcomes in infants with NEC. This early prediction can contribute to timely clinical decision-making and ultimately improve patient prognosis.

Deep learning models for the prediction of acute postoperative pain in PACU for video‐assisted thoracoscopic surgery

BackgroundPostoperative pain is a prevalent symptom experienced by patients undergoing surgical procedures. This study aims to develop deep learning algorithms for predicting acute postoperative pain using both essential patient details and real-time vital sign data during surgery.MethodsThrough a retrospective observational approach, we utilized Graph Attention Networks (GAT) and graph Transformer Networks (GTN) deep learning algorithms to construct the DoseFormer model while incorporating an attention mechanism. This model employed patient information and intraoperative vital signs obtained during Video-assisted thoracoscopic surgery (VATS) surgery to anticipate postoperative pain. By categorizing the static and dynamic data, the DoseFormer model performed binary classification to predict the likelihood of postoperative acute pain.ResultsA total of 1758 patients were initially included, with 1552 patients after data cleaning. These patients were then divided into training set (n = 931) and testing set (n = 621). In the testing set, the DoseFormer model exhibited significantly higher AUROC (0.98) compared to classical machine learning algorithms. Furthermore, the DoseFormer model displayed a significantly higher F1 value (0.85) in comparison to other classical machine learning algorithms. Notably, the attending anesthesiologists' F1 values (attending: 0.49, fellow: 0.43, Resident: 0.16) were significantly lower than those of the DoseFormer model in predicting acute postoperative pain.ConclusionsDeep learning model can predict postoperative acute pain events based on patients' basic information and intraoperative vital signs.

Effects of 8 days intake of hydrogen-rich water on muscular endurance performance and fatigue recovery during resistance training

BackgroundExercise-induced oxidative stress and inflammation can impair muscular function in humans. The antioxidant and anti-inflammatory properties of molecular hydrogen (H2) highlight its potential to be as an effective nutritional supplement to support muscular function performance in healthy adults. However, the effects of H2 supplementation on muscular endurance performance in trained individuals have not been well characterized. This study aimed to assess the effects of intermittent hydrogen-rich water (HRW) supplementation before, during, and after resistance training on muscular endurance performance, neuromuscular status, and subjective perceptual responses after a 48-h recovery period.MethodsThis randomized, double-blinded, placebo-controlled cross-over study included 18 trained men aged 19.7 ± 0.9 years. Participants in this study were instructed to consume 1,920 mL of HRW or pure water (Placebo) daily for 7 days. Additionally, participants were required to supplement with HRW or pure water five times during the training day (1,260 mL total). This included drinking 210 mL 30 min and 1 min before training, 210 mL between training sets, 210 mL immediately after training, and 420 mL 30 min into the recovery period. Participants performed half-squat exercises with the load set at 70% of one repetition maximum for six sets (half-squat exercise performed to repetitions failure each set). We measured the power output and number of repetitions in the free barbell half-squat used to assess muscular endurance performance in participants. The countermovement jump (CMJ) height, total quality recovery scale (TQRS), and muscle soreness visual analog scale (VAS) scores were measured to assess fatigue recovery status after training, as well as at 24 and 48 h of recovery.ResultsThe total power output (HRW: 50,866.7 ± 6,359.9W, Placebo: 46,431.0 ± 9,376.5W, p = 0.032) and the total number of repetitions (HRW:78.2 ± 9.5 repetitions, Placebo: 70.3 ± 9.5 repetitions, p = 0.019) in the H2 supplemented group were significantly higher than in the placebo group. However, there was no statistically significant difference (p< 0.05) between the H2 and placebo groups in CMJ, TQRS, and VAS.ConclusionEight days of intermittent HRW intake could significantly improve muscular endurance performance in trained individuals, making it a promising strategy for athletes or fitness enthusiasts looking to boost muscular endurance during resistance training or competitions. However, it should be noted that HRW intake alone may not be adequate to accelerate recovery from muscle soreness or fatigue following high-intensity training.

Enhancing stereotactic ablative boost radiotherapy dose prediction for bulky lung cancer: A multi-scale dilated network approach with scale-balanced structure loss.

Partial stereotactic ablative boost radiotherapy (P-SABR) effectively treats bulky lung cancer; however, the planning process for P-SABR requires repeated dose calculations. To improve planning efficiency, we proposed a novel deep learning method that utilizes limited data to accurately predict the three-dimensional (3D) dose distribution of the P-SABR plan for bulky lung cancer. We utilized data on 74 patients diagnosed with bulky lung cancer who received P-SABR treatment. The patient dataset was randomly divided into a training set (51 plans) with augmentation, validation set (7 plans), and testing set (16 plans). We devised a 3D multi-scale dilated network (MD-Net) and integrated a scale-balanced structure loss into the loss function. A comparative analysis with a classical network and other advanced networks with multi-scale analysis capabilities and other loss functions was conducted based on the dose distributions in terms of the axial view, average dose scores (ADSs), and average absolute differences of dosimetric indices (AADDIs). Finally, we analyzed the predicted dosimetric indices against the ground-truth values and compared the predicted dose-volume histogram (DVH) with the ground-truth DVH. Our proposed dose prediction method for P-SABR plans for bulky lung cancer demonstrated strong performance, exhibiting a significant improvement in predicting multiple indicators of regions of interest (ROIs), particularly the gross target volume (GTV). Our network demonstrated increased accuracy in most dosimetric indices and dose scores in different ROIs. The proposed loss function significantly enhanced the predictive performance of the dosimetric indices. The predicted dosimetric indices and DVHs were equivalent to the ground-truth values. Our study presents an effective model based on limited datasets, and it exhibits high accuracy in the dose prediction of P-SABR plans for bulky lung cancer. This method has potential as an automated tool for P-SABR planning and can help optimize treatments and improve planning efficiency.

CAM: a novel aid system to analyse the coloration quality of thick blood smears using image processing and machine learning techniques.

Battling malaria's morbidity and mortality rates demands innovative methods related to malaria diagnosis. Thick blood smears (TBS) are the gold standard for diagnosing malaria, but their coloration quality is dependent on supplies and adherence to standard protocols. Machine learning has been proposed to automate diagnosis, but the impact of smear coloration on parasite detection has not yet been fully explored. To develop Coloration Analysis in Malaria (CAM), an image database containing 600 images was created. The database was randomly divided into training (70%), validation (15%), and test (15%) sets. Nineteen feature vectors were studied based on variances, correlation coefficients, and histograms (specific variables from histograms, full histograms, and principal components from the histograms). The Machine Learning Matlab Toolbox was used to select the best candidate feature vectors and machine learning classifiers. The candidate classifiers were then tuned for validation and tested to ultimately select the best one. This work introduces CAM, a machine learning system designed for automatic TBS image quality analysis. The results demonstrated that the cubic SVM classifier outperformed others in classifying coloration quality in TBS, achieving a true negative rate of 95% and a true positive rate of 97%. An image-based approach was developed to automatically evaluate the coloration quality of TBS. This finding highlights the potential of image-based analysis to assess TBS coloration quality. CAM is intended to function as a supportive tool for analyzing the coloration quality of thick blood smears.

Follicular lymphoma comprises germinal center-like and memory-like molecular subtypes with prognostic significance

A robust prognostic and biological classification for newly diagnosed follicular lymphoma (FL) using molecular profiling remains challenging. FL tumors from patients treated in the RELEVANCE trial with rituximab-chemotherapy (R-chemo) or rituximab-lenalidomide (R2) were analyzed using RNA-sequencing, DNA-sequencing, immunohistochemistry (IHC) and/or fluorescence in situ hybridization. Unsupervised gene clustering identified two gene expression signatures (GS) enriched with normal memory (MEM) B-cells and germinal center (GC) B-cells signals, respectively. These two GS were combined into a 20-genes predictor (FL20) to classify patients into MEM-like (n=160) or GC-like (n=164) subtypes, which also displayed different mutational profiles. In the R-chemo arm, MEM-like patients had significantly shorter progression free survival (PFS) than GC-like patients (HR=2.13; p=0.0023), and this prognostic correlation remained significant in a multivariable model including FLIPI (p=0.005). In the R2 arm, both subtypes had comparable PFS, demonstrating a R2 benefit over R-chemo for MEM-like patients (HR=0.54; p=0.011). The prognostic value of FL20 was validated in an independent FL cohort with R-chemo treatment (GSE119214 (n=137)). An IHC algorithm (FLCM) using FOXP1, LMO2, CD22 and MUM1 antibodies was developed with significant prognostic correlation with FL20 in a training set of RELEVANCE (n=264) patients, which was then validated in a different set of patients (n=116). These data indicate that FL tumors can be classified into MEM-like and GC-like subtypes that are biologically distinct and clinically different in risk profile. The FLCM assay can be used in routine clinical practice to identify MEM-like FL patients who might benefit from therapies other than R-chemo, such as the R2 combination. ClinicalTrials.gov identifier: RELEVANCE: NCT01476787 and NCT01650701 INTRODUCTION.

Data mining of PubChem bioassay records reveals diverse OXPHOS inhibitory chemotypes as potential therapeutic agents against ovarian cancer

Focused screening on target-prioritized compound sets can be an efficient alternative to high throughput screening (HTS). For most biomolecular targets, compound prioritization models depend on prior screening data or a target structure. For phenotypic or multi-protein pathway targets, it may not be clear which public assay records provide relevant data. The question also arises as to whether data collected from disparate assays might be usefully consolidated. Here, we report on the development and application of a data mining pipeline to examine these issues. To illustrate, we focus on identifying inhibitors of oxidative phosphorylation, a druggable metabolic process in epithelial ovarian tumors. The pipeline compiled 8415 available OXPHOS-related bioassays in the PubChem data repository involving 312,093 unique compound records. Application of PubChem assay activity annotations, PAINS (Pan Assay Interference Compounds), and Lipinski-like bioavailability filters yields 1852 putative OXPHOS-active compounds that fall into 464 clusters. These chemotypes are diverse but have relatively high hydrophobicity and molecular weight but lower complexity and drug-likeness. These chemotypes show a high abundance of bicyclic ring systems and oxygen containing functional groups including ketones, allylic oxides (alpha/beta unsaturated carbonyls), hydroxyl groups, and ethers. In contrast, amide and primary amine functional groups have a notably lower than random prevalence. UMAP representation of the chemical space shows strong divergence in the regions occupied by OXPHOS-inactive and -active compounds. Of the six compounds selected for biological testing, 4 showed statistically significant inhibition of electron transport in bioenergetics assays. Two of these four compounds, lacidipine and esbiothrin, increased in intracellular oxygen radicals (a major hallmark of most OXPHOS inhibitors) and decreased the viability of two ovarian cancer cell lines, ID8 and OVCAR5. Finally, data from the pipeline were used to train random forest and support vector classifiers that effectively prioritized OXPHOS inhibitory compounds within a held-out test set (ROCAUC 0.962 and 0.927, respectively) and on another set containing 44 documented OXPHOS inhibitors outside of the training set (ROCAUC 0.900 and 0.823). This prototype pipeline is extensible and could be adapted for focus screening on other phenotypic targets for which sufficient public data are available.Scientific contributionHere, we describe and apply an assay data mining pipeline to compile, process, filter, and mine public bioassay data. We believe the procedure may be more broadly applied to guide compound selection in early-stage hit finding on novel multi-protein mechanistic or phenotypic targets. To demonstrate the utility of our approach, we apply a data mining strategy on a large set of public assay data to find drug-like molecules that inhibit oxidative phosphorylation (OXPHOS) as candidates for ovarian cancer therapies.Graphical

Analyzing risk factors and constructing a predictive model for superficial esophageal carcinoma with submucosal infiltration exceeding 200 micrometers

ObjectiveSubmucosal infiltration of less than 200 μm is considered an indication for endoscopic surgery in cases of superficial esophageal cancer and precancerous lesions. This study aims to identify the risk factors associated with submucosal infiltration exceeding 200 micrometers in early esophageal cancer and precancerous lesions, as well as to establish and validate an accompanying predictive model.MethodsRisk factors were identified through least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression. Various machine learning (ML) classification models were tested to develop and evaluate the most effective predictive model, with Shapley Additive Explanations (SHAP) employed for model visualization.ResultsPredictive factors for early esophageal invasion into the submucosa included endoscopic ultrasonography or magnifying endoscopy> SM1(P<0.001,OR = 3.972,95%CI 2.161–7.478), esophageal wall thickening(P<0.001,OR = 12.924,95%CI,5.299–33.96), intake of pickled foods(P=0.04,OR = 1.837,95%CI,1.03–3.307), platelet-lymphocyte ratio(P<0.001,OR = 0.284,95%CI,0.137–0.556), tumor size(P<0.027,OR = 2.369,95%CI,1.128–5.267), the percentage of circumferential mucosal defect(P<0.001,OR = 5.286,95%CI,2.671–10.723), and preoperative pathological type(P<0.001,OR = 4.079,95%CI,2.254–7.476). The logistic regression model constructed from the identified risk factors was found to be the optimal model, demonstrating high efficacy with an area under the curve (AUC) of 0.922 in the training set, 0.899 in the validation set, and 0.850 in the test set.ConclusionA logistic regression model complemented by SHAP visualizations effectively identifies early esophageal cancer reaching 200 micrometers into the submucosa.

Chemometric Discrimination of the Geographical Origin of Rheum tanguticum by Stable Isotope Analysis

Rheum tanguticum is one of the primary rhubarb species used for food and medicinal purposes, and it has recently been gaining more attention and recognition. This research represents the first attempt to use stable isotopes and elemental analysis via IRMS to identify the geographical origin of Rheum tanguticum. A grand total of 190 rhubarb samples were gathered from 38 locations spread throughout the provinces of Gansu, Sichuan, and Qinghai in China. The carbon content showed a decreasing trend in the order of Qinghai, followed by Sichuan, and then Gansu. Nitrogen content was notably higher, with Qinghai and Sichuan displaying similar levels, while Gansu had the lowest nitrogen levels. Significant differences were noted in the δ13C (−28.9 to −26.5‰), δ15N (2.6 to 5.6‰), δ2H (−120.0 to −89.3‰), and δ18O (16.0‰ to 18.8‰) isotopes among the various rhubarb cultivation areas. A significant negative correlation was found between %C and both longitude and humidity. Additionally, δ13C and δ15N isotopes were negatively correlated with longitude, and δ15N showed a negative correlation with humidity as well. δ2H and δ18O isotopes exhibited a strong positive correlation with latitude, while significant negative correlations were observed between δ2H and δ18O isotopes and temperature, precipitation, and humidity. The LDA, PLS-DA, and k-NN models all exhibited strong classification performance in both the training and validation sets, achieving accuracy rates between 82.1% and 91.7%. The combination of stable isotopes, elemental analysis, and chemometrics provides a reliable and efficient discriminant model for accurately determining the geographical origin of R. tanguticum in different regions. In the future, the approach will aid in identifying the geographical origin and efficacy of rhubarb in other studies.

Falling Detection of Toddlers Based on Improved YOLOv8 Models

If toddlers are not promptly checked and rescued after falling from relatively high locations at homes, they are at risk of severe health complications. We present a toddler target extraction method and real-time falling alarm. The procedure is executed in two stages: In stage I, a GELAN-integrated YOLOv8 model is used to extract the body features. Based on this, a head capture technique is developed to obtain the head features. In stage II, the “safe zone” is calculated through Generalized Hough Transform (GHT). The spatial location is compared to the preceding stage’s two centers of mass points, K for the toddler’s body and H for the head. Position status detection is performed on the extracted data. We gathered 230 RGB-captured daily videos of toddlers aged 13 to 30 months playing and experiencing upside-down falls. We split 500 video clips (×30 FPS) from 200 videos into 8:2 training and validation sets. A test set of 100 clips (×30 FPS) was cut from another 30 videos. The experimental results suggested that the framework has higher precision and recall in detection, as well as improved mean average precision and F1 scores compared to YOLOv3, v5, v6, and v8. It meets the standard FPS requirement for surveillance cameras and has an accuracy of 96.33 percent.

Intratumoral and peritumoral radiomics model for the preoperative prediction of cribriform component in invasive lung adenocarcinoma: a multicenter study.

This study aimed to investigate the predictive value of intratumoral and peritumoral radiomics model for the cribriform component (CC) of invasive lung adenocarcinoma (LUAD). The 144 patients with invasive LUAD from our center were randomly divided into training set (n = 100) and internal validation set (n = 44) in a ratio of 7:3, and 75 patients from center 2 were regarded as the external validation set. Clinical risk factors were examined using univariate and multivariate logistic regression to construct the clinical model. We extracted radiomics features from gross tumor volume (GTV), gross and peritumoral volume (GPTV), and peritumoral volume (PTV), respectively. Radiomics models were constructed with selected features. A combined model based on the optimal Radscore and clinically independent predictors was constructed, and its predictive performance was assessed by receiver operating characteristic curve (ROC), calibration curve, and decision curve analysis (DCA). The area under curves (AUCs) of the GTV model were 0.882 (95% CI 0.817-0.948), 0.794 (95% CI 0.656-0.932), and 0.766 (95% CI 0.657-0.875) in the training, internal validation, and external validation sets, and the PTV model had AUCs of 0.812 (95% CI 0.725-0.899), 0.749 (95% CI 0.597-0.902), and 0.670 (95% CI 0.543-0.798) in the training, internal validation, and external validation sets, respectively. However, the GPTV radiomics model showed better predictive performance compared with the GTV and PTV radiomics models, with the AUCs of 0.950 (95% CI 0.911-0.989), 0.844 (95% CI 0.728-0.959), and 0.815 (95% CI 0.713-0.917) in the training, internal validation and external validation sets, respectively. In the clinical model, tumor shape, lobulation sign and maximal diameter were the independent predictors of CC in invasive LUAD. The combined model including independent clinical predictors and GPTV-Radscore show the considerable instructive to clinical practice, with the AUCs of 0.954(95% CI 0.918-0.990), 0.861(95% CI 0.752-0.970), and 0.794(95% CI 0.690-0.898) in training, internal validation, and external validation sets, respectively. DCA showed that the combined model had good clinical value and correction effect. Radiomics model is a very powerful tool for predicting CC growth pattern in invasive LUAD and can help clinicians make the strategies of treatment and surveillance in patients with invasive LUAD.

Interpretable machine learning model for predicting the prognosis of antibody positive autoimmune encephalitis patients

ObjectiveThe objective was to utilize nine machine learning (ML) methods to predict the prognosis of antibody positive autoimmune encephalitis (AE) patients. MethodsThe encephalitis data from the Global Burden of Disease (GBD) study is analyzed to reflect the disease burden of encephalitis. This study included 187 patients with AE. 121 patients as training set and 67 patients as validation set. Decision trees (DT), random forest (RF), extreme gradient boosting (XGBoost), k-nearest neighbor (KNN), support vector machine (SVM), naive bayes (NB), neural network (NN), light gradient boosting machine (LGBM), and logistic regression (LR) are ML methods used to construct predictive models. The constructed models were validated for discrimination, calibration and clinical applicability using validation set data. Shapley additive explanation (SHAP) analysis was used to explain the model. ResultsThe number of encephalitis worldwide deaths, incidence and prevalence is increasing every year from 2010 to 2021. The training set included 121 patients with AE. Univariate analysis and LASSO screening identified six variables. The results of constructing models using 9 ML methods showed RF had the highest accuracy (0.860), followed by XGBoost (0.826), with F1 scores of 0.844 and 0.807, respectively. Validation set data showed good discrimination, calibration and clinical applicability of the model. The SHAP values of infection, CSF monocyte percentage, and prealbumin were 0.906, 0.790, and 0.644, respectively. LimitationsAs a rare disease, the sample size of this study is relatively small. ConclusionThe model constructed using RF and XGBoost has good performance, good discrimination, calibration, clinical applicability, and interpretability.

Exploring Feature Relationships in Brain Stroke Data Using Polynomial Feature Transformation and Linear Regression Modeling

A Cerebral vascular accident, commonly known as a stroke, is a pathological condition that impacts the brain due to the rupture of capillaries. It occurs when there is a disturbance in the typical blood circulation and essential physiological processes of the brain. Stroke prediction plays a crucial role in early diagnosis and intervention, potentially improving patient outcomes. This paper proposes a machine learning model that leverages polynomial feature transformation and linear regression modeling for stroke prediction. The model addresses the challenge of capturing non-linear relationships between features and the target variable while maintaining interpretability. The proposed approach involves preprocessing data by separating categorical and numerical features, applying one-hot encoding to categorical features, and generating polynomial features up to the second degree for numerical features. This tailored preprocessing is facilitated by a Column Transformer. For model development, a machine learning pipeline is constructed, splitting the data into training and testing sets. Despite utilizing polynomial features, linear regression is employed as the final model, allowing for the capture of both linear and non-linear relationships while maintaining interpretability. This work contributes to stroke prediction by offering a balanced approach that considers model complexity and interpretability, showcasing the potential of linear regression with polynomial features for accurate predictions and insights into feature-target relationships. The proposed model exhibited superior performance compared to other existing models, achieving a remarkable testing accuracy of 99.2%.

Deep learning-based characterization of pathological subtypes in lung invasive adenocarcinoma utilizing 18F-deoxyglucose positron emission tomography imaging

SummaryObjectiveTo evaluate the diagnostic efficacy of a deep learning (DL) model based on PET/CT images for distinguishing and predicting various pathological subtypes of invasive lung adenocarcinoma.MethodsA total of 250 patients diagnosed with invasive lung cancer were included in this retrospective study. The pathological subtypes of the cancer were recorded. PET/CT images were analyzed, including measurements and recordings of the short and long diameters on the maximum cross-sectional plane of the CT image, the density of the lesion, and the associated imaging signs. The SUVmax, SUVmean, and the lesion's long and short diameters on the PET image were also measured. A manual diagnostic model was constructed to analyze its diagnostic performance across different pathological subtypes. The acquired images were first denoised, followed by data augmentation to expand the dataset. The U-Net network architecture was then employed for feature extraction and network segmentation. The classification network was based on the ResNet residual network to address the issue of gradient vanishing in deep networks. Batch normalization was applied to ensure the feature matrix followed a distribution with a mean of 0 and a variance of 1. The images were divided into training, validation, and test sets in a ratio of 6:2:2 to train the model. The deep learning model was then constructed to analyze its diagnostic performance across different pathological subtypes.ResultsStatistically significant differences (P < 0.05) were observed among the four different subtypes in PET/CT imaging performance. The AUC and diagnostic accuracy of the manual diagnostic model for different pathological subtypes were as follows: APA: 0.647, 0.664; SPA: 0.737, 0.772; PPA: 0.698, 0.780; LPA: 0.849, 0.904. Chi-square tests indicated significant statistical differences among these subtypes (P < 0.05). The AUC and diagnostic accuracy of the deep learning model for the different pathological subtypes were as follows: APA: 0.854, 0.864; SPA: 0.930, 0.936; PPA: 0.878, 0.888; LPA: 0.900, 0.920. Chi-square tests also indicated significant statistical differences among these subtypes (P < 0.05). The Delong test showed that the diagnostic performance of the deep learning model was superior to that of the manual diagnostic model (P < 0.05).ConclusionsThe deep learning model based on PET/CT images exhibits high diagnostic efficacy in distinguishing and diagnosing various pathological subtypes of invasive lung adenocarcinoma, demonstrating the significant potential of deep learning techniques in accurately identifying and predicting disease subgroups.

From Labquakes to Megathrusts: Scaling Deep Learning Based Pickers Over 15 Orders of Magnitude

AbstractThe application of machine learning techniques in seismology has greatly advanced seismological analysis, especially for earthquake detection and seismic phase picking. However, machine learning approaches still face challenges in generalizing to data sets that differ from their original training setting. Previous studies focused on retraining or transfer‐learning models for these scenarios, but require high‐quality labeled data sets. This paper demonstrates a new approach for augmenting already trained models without the need for additional training data. We propose four strategies—rescaling, model aggregation, shifting, and filtering—to enhance the performance of pre‐trained models on out‐of‐distribution data sets. We further devise various methodologies to ensemble the individual predictions from these strategies to obtain a final unified prediction result featuring prediction robustness and detection sensitivity. We develop an open‐source Python module quakephase that implements these methods and can flexibly process input continuous seismic data of any sampling rate. With quakephase and pre‐trained ML models from SeisBench, we perform systematic benchmark tests on data recorded by different types of instruments, ranging from acoustic emission sensors to distributed acoustic sensing, and collected at different scales, spanning from laboratory acoustic emission events to major tectonic earthquakes. Our tests highlight that rescaling is essential for dealing with small‐magnitude seismic events recorded at high sampling rates as well as larger magnitude events having long coda and remote events with long wave trains. Our results demonstrate that the proposed methods are effective in augmenting pre‐trained models for out‐of‐distribution data sets, especially in scenarios with limited labeled data for transfer learning.

Groundnut (ARACHIS HYPOGAEA L.) seed defect classification using ensemble deep learning techniques

Groundnut is an oil seed cash crop, that can be consumed directly by humans and animals. It is also used as an intelligence in the industrial production of butter and other products. However, Groundnut seeds can become infected with fungi, viruses, pests, physical damage, and high heat, which can negatively impact crop yield, quality, and economic value, and pose health hazards. To address these challenges, computer vision technology detects and classifies defects. This study introduced an ensemble deep learning defected classification model that combines VGG16 and InceptionV3 using seed images. To enhance model performance, collected images are preprocessed using novel techniques tailored to different image types. Preprocessing techniques are chosen based on image-quality evaluation metrics. Watershed segmentation is applied followed by detecting the region of interest in the image using YOLOv3. The image dataset is augmented and balanced using a Generative Adversarial Network (GAN). The model development involves a combination of classical and deep-based features, comparing features extracted with (HOG and GLCM) to those extracted with InceptionV3 and VGG16. The ensemble model achieves an accuracy of 96.25 % with a split ratio of 10 % for testing, 10 % for validation, and 80 % for training set. This research benefits farmers looking to improve yield, consumers of groundnut products, and researchers studying seed defects. The work provides valid insights for future researchers regarding dataset creation, augmentation, methods, and feature selection for modeling. However, the model experienced misclassifications due to the similar appearance of sample images in different classes. Future researchers should address these challenges and consider factors such as oil content and defect grading.

Bayesian Optimization Using Simulation-Based Multiple Information Sources over Combinatorial Structures

Bayesian optimization due to its flexibility and sample efficiency has become a standard approach for simulation optimization. To reduce this problem, one can resort to cheaper surrogates of the objective function. Examples are ubiquitous, from protein engineering or material science to tuning machine learning algorithms, where one could use a subset of the full training set or even a smaller related dataset. Cheap information sources in the optimization scheme have been studied in the literature as the multi-fidelity optimization problem. Of course, cheaper sources may hold some promise toward tractability, but cheaper models offer an incomplete model inducing unknown bias and epistemic uncertainty. In this manuscript, we are concerned with the discrete case, where fx,wi is the value of the performance measure associated with the environmental condition wi and p(wi) represents the relevance of the condition wi (i.e., the probability of occurrence or the fraction of time this condition occurs). The main contribution of this paper is the proposal of a Gaussian-based framework, called augmented Gaussian process (AGP), based on sparsification, originally proposed for continuous functions and its generalization in this paper to stochastic optimization using different risk profiles for combinatorial optimization. The AGP leverages sample and cost-efficient Bayesian optimization (BO) of multiple information sources and supports a new acquisition function to select the new source–location pair considering the cost of the source and the (location-dependent) model discrepancy. An extensive set of computational results supports risk-aware optimization based on CVaR (conditional value-at-risk). Computational experiments confirm the actual performance of the MISO-AGP method and the hyperparameter optimization on benchmark functions and real-world problems.

MRI-based radiomics for stratifying recurrence risk of early-onset rectal cancer: a multicenter study

Early-onset rectal cancer (EORC) is characterized by a unique disease process with different clinicopathological features compared with late-onset rectal cancer (LORC). Research on the risk of recurrence in EORC patients, however, is limited. We aim to develop a predictive model to accurately predict EORC recurrence risk. Rectal cancer patients who underwent radical surgery and T2-weighted imaging and diffusion-weighted imaging magnetic resonance imaging (MRI) were retrospectively enrolled from three medical institutions from November 2012 to November 2018. Differences in clinicopathological characteristics between EORC and LORC were compared. Five prediction models for disease-free survival were constructed based on clinicopathological variables and five radiomic features from pretreatment MRI of the EORC. A fixed cut-off value calculated in the training set was used to stratify EORC patients into high-risk and low-risk groups of post-operative recurrence. Model performance was evaluated by concordance index (C-index) and receiver operating characteristic curve. A total of 264 EORC patients (median age, 43 years, 163 males) and 778 LORC patients (median age, 62 years, 520 males) were enrolled. Pretreatment positive carcinoembryonic antigen [hazard ratio (HR)= 2.84, P= 0.006], pathological positive lymph node status (pN positive) [HR= 2.86, P= 0.011] and MRI-based radiomics score [HR= 2.72, P < 0.001] are independent risk factors for disease-free survival in EORC patients. The EORC-ClinPathRadiom model, constructed by integrating the clinicopathological characteristics and MRI-based radiomics features of EORC, showed C-index of 0.82, 0.82, and 0.81 in the training, internal, and external test sets, respectively. This model effectively stratified EORC patients into high risk and low risk of recurrence (HRs for the training, internal, and external test sets were 8.96, 6.81, and 7.46, respectively). The EORC-ClinPathRadiom model can effectively predict and stratify the risk of post-operative recurrence in EORC patients.

MRI T2WI-based radiomics combined with KRAS gene mutation constructed models for predicting liver metastasis in rectal cancer

BackgroundThe study aimed to identify the optimal model for predicting rectal cancer liver metastasis (RCLM). This involved constructing various prediction models to aid clinicians in early diagnosis and precise decision-making.MethodsA retrospective analysis was conducted on 193 patients diagnosed with rectal adenocarcinoma were randomly divided into training set (n = 136) and validation set (n = 57) at a ratio of 7:3. The predictive performance of three models was internally validated by 10-fold cross-validation in the training set. Delineation of the tumor region of interest (ROI) was performed, followed by the extraction of radiomics features from the ROI. The least absolute shrinkage and selection operator (LASSO) regression algorithm and multivariate Cox analysis were employed to reduce the dimensionality of radiomics features and identify significant features. Logistic regression was employed to construct three prediction models: clinical, radiomics, and combined models (radiomics + clinical). The predictive performance of each model was assessed and compared.ResultsKRAS mutation emerged as an independent predictor of liver metastasis, yielding an odds ratio (OR) of 8.296 (95%CI: 3.471–19.830; p < 0.001). 5 radiomics features will be used to construct radiomics model. The combined model was built by integrating radiomics model with clinical model. In both the training set (AUC:0.842, 95%CI: 0.778–0.907) and the validation set (AUC: 0.805; 95%CI: 0.692–0.918), the AUCs for the combined model surpassed those of the radiomics and clinical models.ConclusionsOur study reveals that KRAS mutation stands as an independent predictor of RCLM. The radiomics features based on MR play a crucial role in the evaluation of RCLM. The combined model exhibits superior performance in the prediction of liver metastasis.Clinical Trial NumberNot applicable.

Deep-learning based discrimination of pathologic complete response using MRI in HER2-positive and triple-negative breast cancer

Distinguishing between pathologic complete response and residual cancer after neoadjuvant chemotherapy (NAC) is crucial for treatment decisions, but the current imaging methods face challenges. To address this, we developed deep-learning models using post-NAC dynamic contrast-enhanced MRI and clinical data. A total of 852 women with human epidermal growth factor receptor 2 (HER2)-positive or triple-negative breast cancer were randomly divided into a training set (n = 724) and a validation set (n = 128). A 3D convolutional neural network model was trained on the training set and validated independently. The main models were developed using cropped MRI images, but models using uncropped whole images were also explored. The delayed-phase model demonstrated superior performance compared to the early-phase model (area under the receiver operating characteristic curve [AUC] = 0.74 vs. 0.69, P = 0.013) and the combined model integrating multiple dynamic phases and clinical data (AUC = 0.74 vs. 0.70, P = 0.022). Deep-learning models using uncropped whole images exhibited inferior performance, with AUCs ranging from 0.45 to 0.54. Further refinement and external validation are necessary for enhanced accuracy.