Model Development Dataset Research Articles

Deep learning-assisted colonoscopy images for prediction of mismatch repair deficiency in colorectal cancer.

Deficient mismatch repair or microsatellite instability is a major predictive biomarker for the efficacy of immune checkpoint inhibitors of colorectal cancer. However, routine testing has not been uniformly implemented due to cost and resource constraints. We developed and validated a deep learning-based classifiers to detect mismatch repair-deficient status from routine colonoscopy images. We obtained the colonoscopy images from the imaging database at Endoscopic Center of the Sixth Affiliated Hospital, Sun Yat-sen University. Colonoscopy images from a prospective trial (Neoadjuvant PD-1 blockade by toripalimab with or without celecoxib in mismatch repair-deficient or microsatellite instability-high locally advanced colorectal cancer) were used to test the model. A total of 5226 eligible images from 892 tumors from the consecutive patients were utilized to develop and validate the deep learning model. 2105 colorectal cancer images from 306 tumors were randomly selected to form model development dataset with a class-balanced approach. 3121 images of 488 proficient mismatch repair tumors and 98 deficient mismatch repair tumors were used to form the independent dataset. The model achieved an AUROC of 0.948 (95% CI 0.919-0.977) on the test dataset. On the independent validation dataset, the AUROC was 0.807 (0.760-0.854), and the NPV in was 94.2% (95% CI 0.918-0.967). On the prospective trial dataset, the model identified 29 tumors among the 33 deficient mismatch repair tumors (87.88%). The model achieved a high NPV in detecting deficient mismatch repair colorectal cancers. This model might serve as an automatic screening tool.

Limited Sampling Strategy for Predicting the Area under Plasma Concentration–Time Curve of Nadolol in Healthy Subjects

AbstractNadolol is a hydrophilic β‐adrenoceptor blocker with a relatively long half‐life and negligible metabolism. It is a substrate of P‐glycoprotein and organic anion transporting polypeptide 1A2, and may serve as an in vivo probe drug for the assessment of drug–drug and food–drug interactions mediated by these transporters. In the present study, we aimed to develop limited sampling strategy (LSS) models for predicting the area under the plasma concentration–time curve (AUC0‐∞) of nadolol. Plasma concentration data (Ct) in healthy volunteers reported in four previous studies were randomly divided into a training dataset for model development (n = 15) and a test dataset for model validation (n = 16). By multiple linear regression analysis, we confirmed that four out of the eight models using two time points and all models using three time points met the acceptable criteria. In particular, the three time point models using (C3, C6, and C24) and (C4, C8, and C24) showed better predictive performances with r2 values of 0.983 and 0.980, respectively. In drug interaction studies of nadolol with itraconazole, rifampicin, grapefruit juice, and green tea extract, both LSS models accurately predicted the AUC0‐∞ with percent mean absolute error ≤11% and percent root mean square error ≤12%. In addition, using digitized pharmacokinetic data of nadolol, both LSS models were further validated by predicting the AUC0‐∞ in different doses. The results suggest that the LSS models using three time points allow a reliable prediction of AUC0‐∞ of nadolol in healthy individuals.

Delta-Radiomics Approach Using Contrast-Enhanced and Noncontrast-Enhanced Computed Tomography Images for Predicting Distant Metastasis in Patients With Borderline Resectable Pancreatic Carcinoma

PurposeTo predict distant metastasis (DM) in patients with borderline resectable pancreatic carcinoma (BRPC) using delta-radiomics features calculated from contrast-enhanced computed tomography (CECT) and non-CECT images. MethodsAmong 250 patients who underwent radiotherapy at our institution between February 2013 and December 2021, 67 patients were deemed eligible. A total of 11 clinical features and 3,906 radiomics features were incorporated. Radiomics features were extracted from CECT and non-CECT images and the differences between these features were calculated, resulting in delta-radiomics features. The patients were randomly divided into the training (70%) and test (30%) datasets for model development and validation. Predictive models were developed with clinical features (clinical model), radiomics features (radiomics model), and a combination of the abovementioned features (hybrid model) using Fine–Gray regression (FG) and random survival forest (RSF). Optimal hyperparameters were determined using stratified five-fold cross-validation. Subsequently, the developed models were applied to the remaining test datasets, and the patients were divided into the high- or low-risk groups based on their risk scores. Prognostic power was assessed using the concordance index (C-index) with 95% confidence intervals obtained through 2,000 bootstrapping iterations. Statistical significance between the two groups was assessed using Gray's test. ResultsAt a median follow-up period of 23.8 months, 47 (70.1%) patients developed DM. The C-indices of the FG-based clinical, radiomics, and hybrid models were 0.548, 0.603, and 0.623, respectively, in the test dataset, whereas those of the RSF-based models were 0.598, 0.680, and 0.727, respectively. The RSF-based model, including delta-radiomics features, significantly divided the cumulative incidence curves into two groups (P<0.05). The feature map of the gray-level size-zone matrix showed that the difference in feature values between CECT and non-CECT images correlated with the incidence of DM. ConclusionsDelta-radiomics features obtained from CECT and non-CECT images using RSF successfully predict the incidence of DM in patients with BRPC.

Taxonomy-based prompt engineering to generate synthetic drug-related patient portal messages

Objective:The objectives of this study were to: (1) create a corpus of synthetic drug-related patient portal messages to address the current lack of publicly available datasets for model development, (2) assess differences in language used and linguistics among the synthetic patient portal messages, and (3) assess the accuracy of patient-reported drug side effects for different racial groups. Methods:We leveraged a taxonomy for patient- and clinician-generated content to guide prompt engineering for synthetic drug-related patient portal messages. We generated two groups of messages: the first group (200 messages) used a subset of the taxonomy relevant to a broad range of drug-related messages and the second group (250 messages) used a subset of the taxonomy relevant to a narrow range of messages focused on side effects. Prompts also include one of five racial groups. Next, we assessed linguistic characteristics among message parts (subject, beginning, body, ending) across different prompt specifications (urgency, patient portal taxa, race). We also assessed the accuracy and frequency of patient-reported side effects across different racial groups and compared to real world data. Results:The study generated 450 synthetic patient portal messages, and we assessed linguistic patterns, accuracy of drug-side effect pairs, frequency of pairs compared to real world data. Linguistic analysis with LIWC revealed variations in language usage and politeness and analysis of positive predictive values identified differences in symptoms reported based on urgency levels and racial groups in the prompt. We also found similar levels of drug-side effect pair occurrence to the SIDER database. Conclusion:This study demonstrates the potential of synthetic patient portal messages as a valuable resource for healthcare research. After creating a corpus of synthetic drug-related patient portal messages, we identified significant language differences and evaluated the accuracy and frequency of drug-side effect pairs across various prompts.

Prediction of femoral head collapse in osteonecrosis using deep learning segmentation and radiomics texture analysis of MRI

BackgroundFemoral head collapse is a critical pathological change and is regarded as turning point in disease progression in osteonecrosis of the femoral head (ONFH). In this study, we aim to build an automatic femoral head collapse prediction pipeline for ONFH based on magnetic resonance imaging (MRI) radiomics.MethodsIn the segmentation model development dataset, T1-weighted MRI of 222 hips from two hospitals were retrospectively collected and randomly split into training (n = 190) and test (n = 32) sets. In the prognosis prediction model development dataset, 206 hips were also retrospectively collected from two hospitals and divided into training set (n = 155) and external test set (n = 51) according to data source. A deep learning model for automatic lesion segmentation was trained with nnU-Net, from which three-dimensional regions of interest were segmented and a total of 107 radiomics features were extracted. After intra-class correlation coefficients screening, feature correlation coefficient screening and Least Absolute Shrinkage and Selection Operator regression feature selection, a machine learning model for ONFH prognosis prediction was trained with Logistic Regression (LR) and Light Gradient Boosting Machine (LightGBM) algorithm.ResultsThe segmentation model achieved an average dice similarity coefficient of 0.848 and an average 95% Hausdorff distance of 3.794 in the test set, compared to the manual segmentation results. After feature selection, nine radiomics features were included in the prognosis prediction model. External test showed that the LightGBM model exhibited acceptable predictive performance. The area under the curve (AUC) of the prediction model was 0.851 (95% CI: 0.7268–0.9752), with an accuracy of 0.765, sensitivity of 0.833, and specificity of 0.727. Decision curve analysis showed that the LightGBM model exhibited favorable clinical utility.ConclusionThis study presents an automated pipeline for predicting femoral head collapse in ONFH with acceptable performance. Further research is necessary to determine the clinical applicability of this radiomics-based approach and to assess its potential to assist in treatment decision-making for ONFH.

A smartphone based AI model to detect left ventricular systolic dysfunction on 12-lead ECG

Abstract Background Despite echocardiography being essential for cardiac assessment, a point-of-care tool for rapid left ventricular ejection fraction (LVEF) evaluation is lacking in current clinical practice, highlighting the need for a widescale, easily deployable method to assess LV function. Purpose To develop and validate two artificial intelligence (AI) models that accurately detect reduced LVEF on a single 12-lead ECG using a smartphone, facilitating early identification of patients at risk for heart failure who may benefit from more detailed echocardiographic evaluation. Methods A large collection of ECGs and transthoracic echocardiograms (TTEs) was sourced from our hospital data vault from 2011 to 2021. ECGs were paired to TTEs within a 24h time window and the resulting pairs were randomly assigned to the model development dataset (50%) and validation dataset (50%). Two distinct AI-ECG models were developed: one to detect LVEF≤40% and another for LVEF&amp;lt;50%, derived from the ESC definition of heart failure. Both AI-ECG models were paired with smartphone based ECG digitization technology. Key performance metrics include area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1-score. Results A total of 1,205,370 ECGs (198,274 patients) and 291,433 TTEs (105,849 patients) were collected and paired, resulting in 109,809 ECG-TTE pairs (56,236 patients). The validation dataset consisted of 25,510 distinct TTE-ECG pairs (25,510 patients). Prevalence of LVEF≤40% and LVEF&amp;lt;50% was 5.4% and 7.9% respectively. The LVEF≤40% model demonstrated an AUC of 0.963 (95% CI: 0.959-0.966), sensitivity 0.924 (95% CI: 0.91-0.937), specificity 0.887 (95% CI: 0.883-0.891), and F1-score of 0.474 (95% CI: 0.457-0.490). PPV and NPV were 0.318 (95% CI: 0.304-0.333) and 0.995 (95% CI: 0.994-0.996) respectively. Performance of the LVEF &amp;lt;50% model shows an AUC of 0.952 (95% CI: 0.947-0.956), with a slightly lower sensitivity of 0.899 (95% CI: 0.886-0.912), specificity of 0.875 (95% CI: 0.871-0.879), PPV of 0.382 (95% CI: 0.368-0.395), NPV of 0.99 (95% CI: 0.989-0.992), and F1-score of 0.536 (0.521-0.55). Conclusion The AI models exhibit the ability to identify reduced LVEF from standard 12-lead ECGs using a smartphone. Results suggest potential clinical utility as a preliminary screening tool to select patients for confirmatory echocardiography. This AI application could potentially offer a timely and cost-effective method for identifying patients at risk for future development of heart failure.

Detection, counting, and maturity assessment of blueberries in canopy images using YOLOv8 and YOLOv9

Blueberry growers currently rely on manual or subjective approaches to harvest decision-making and yield estimation. Canopy image-based, pre-harvest blueberry detection provides a promising means for automated harvest maturity assessment, fruit counting, and yield estimation to optimize crop management and reduce labor costs. Data-driven deep-learning-based object detectors, especially the YOLO (You Only Look Once) series models, have emerged as a powerful tool for a diversity of computer vision tasks in agriculture. However, the detection of such small, densely populated objects as blueberries in unstructured scenes can pose great challenges, especially given the lack of dedicated datasets for model development. This study, therefore, presents the first publicly available dataset of blueberry canopy images with 17,955 ripe and unripe blueberries, which were captured in diverse orchard conditions, and performance evaluation of YOLOv8l (large) and YOLOv9-c (compact) with comparable complexity for blueberry detection and whereby fruit counting and ripe fruit percentage estimation. At the input image resolution of 2560 × 2560 pixels, both models performed similarly in terms of detection accuracy with around 91 % mAP@50, except that YOLOv9-c was far more time-consuming to train. Trained with the input of higher resolution images of 3520 × 3520 pixels, YOLOv8l achieved an overall mAP@50 of nearly 93 %, and an RMSEs (root-mean-square errors) of 10.4 in fruit counting and 3.62 % in estimating the percentage of ripe fruit of each image. Both the blueberry dataset11https://zenodo.org/records/14002517 and software programs22https://github.com/vicdxxx/BlueberryDetectionAndCounting of this study have been or will be made publicly available, which are expected to be useful for promoting greater community efforts for further algorithm development.

Development of the machine learning model that is highly validated and easily applicable to predict radiographic knee osteoarthritis progression.

Many models using the aid of artificial intelligence have been recently proposed to predict the progression of knee osteoarthritis. However, previous models have not been properly validated with an external data set or have reported poor predictive performances. Therefore, the purpose of this study was to design a machine learning model for knee osteoarthritis progression, focusing on high validation quality and clinical applicability. A retrospective analysis was conducted on prospectively collected data, using the Osteoarthritis Initiative data set (5966 knees) for model development and the Multicenter Osteoarthritis Study data set (3392 knees) for validation. The analysis aimed to predict Kellgren-Lawrence grade (KLG) progression over 4-5 years in knees with initial KLG of 0, 1, or 2. Possible predictors included demographics, comorbidities, history of meniscectomy, gait speed, Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) scores, and radiological findings. The Random Forest algorithm was employed for the predictive model development. Baseline KLG, contralateral knee osteoarthritis, lateral joint space narrowing (JSN) grade, BMI, medial JSN grade, and total WOMAC score were six features selected for the model in descending order of importance. Odds ratios of baseline KLG, contralateral knee osteoarthritis, and lateral JSN grade were 1.76, 2.59, and 4.74, respectively (all p < 0.001). The area-under-the-curve of the ROC curve in the validation set was 0.76 with an accuracy of 0.68 and an F1-score of 0.56. The progression of knee osteoarthritis in 4 ~ 5 years could be well-predicted using easily available variables. This simple and validated model may aid surgeons in knee osteoarthritis patient management.

Thai Baht and Chinese Yuan Exchange Rate Forecasting Models: ARIMA and SMA-ARIMA Comparison

This study evaluates the effectiveness of classical Autoregressive Integrated Moving Average (ARIMA) models and kth Simple Moving Average - ARIMA (kth SMA-ARIMA) models in forecasting the exchange rate between the Thai Baht (THB) and the Chinese Yuan (CNY). The analysis uses a dataset of historical monthly exchange rates from January 2011 to November 2022, covering 143 months. The dataset is divided into two segments: the initial 127 months are used as the training dataset for model development, while the subsequent 16 months serve as the testing dataset to evaluate forecast accuracy. The Akaike Information Criterion (AIC) is the decision criterion for model selection during the development phase. The forecasting models' effectiveness is subsequently assessed on the testing dataset using two statistical measures: the Mean Absolute Percentage Error (MAPE) and the Root Mean Square Error (RMSE). The findings indicate that the classical ARIMA (0,1,1) model outperforms the kth SMA-ARIMA models in this study, exhibiting the lowest RMSE and MAPE of 0.1702 and 2.6644, respectively. Additionally, a focused comparison of the kth SMA-ARIMA models for k = 2, 3, and 4 reveals that the 2nd SMA-ARIMA (0,1,2) model demonstrates superior performance compared to the 3rd and 4th SMA-ARIMA models. This superiority is reflected in their respective RMSE values of 0.3202, 0.5146, and 0.6339, and corresponding MAPE values of 5.3533, 8.7531, and 10.4949. These results provide valuable insights for decision-makers in the financial sector, enhancing investment strategy formulation based on anticipated currency movements.

Integrated Multiphase-Flow Modeling Technique Yields Accurate Downhole Pressure Predictions

_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 214855, “Integrated Multiphase-Flow Modeling for Downhole Pressure Predictions,” by Abdullah Alkhezzi and Yilin Fan, SPE, Colorado School of Mines. The paper has not been peer reviewed. _ This work presents an integrated multiphase flow model for downhole pressure predictions. The aim of the model is to produce more-accurate downhole pressure predictions under wide flowing conditions while maintaining a simple form. As a component of the integrated model, an improved two-fluid model for segregated flow is proposed. Results of the two-fluid segregated model are compared with five state-of-the-art existing models, while results of the integrated model are compared with three models. Introduction Throughout the life of the well, knowing bottomhole flowing pressure (Pwf) and the pressure profile of the wellbore are of great significance. Unfortunately, deploying downhole pressure gauges to obtain Pwfreadings is often not economical or practical. The common practice is to apply hydraulic models to predict Pwfgiven surface measurements. The prediction of such behavior is simple when dealing with single-phase fluid flow. Unfortunately, this condition is rare in the petroleum industry. The existence of multiple phases introduces multiple complexities hindering the accuracy of downhole pressure predictions. To account for such variations, fluid-property models must be integrated into the calculation procedure. The complexity, coupled with field-data scarcity, results in the deficiency of work that evaluates point models on actual wells. In this work, the authors evaluated the performance of a few widely used multiphase-flow point-based models on actual field data using a marching algorithm and developed a simplified, yet more precise, integrated model. Integrated Model Development Data Set Description. In this study, data points were collected for two main purposes. First, experimental data sets were collected to model and improve the segregated flow model. Second, field data were collected to test the integrated multiphase model. To improve the segregated flow model, 1,478 experimental data points were obtained from various sources in the literature. To evaluate the integrated multiphase flow model, 313 data points from two main sources of data were used (literature and actual field data). In total, four data sets were obtained from the literature and one was obtained from Civitas Resources. Integrated Model Description. The multiphase-flow point model incorporated in the authors’ integrated modeling consists of three main components: critical gas velocity estimation for the onset of liquid loading and hydraulic multiphase flow models before and after the onset of liquid loading. The proposed model characterizes the flow based on whether liquid loading occurs. The onset of liquid loading corresponds with the transition of the flow pattern from segregated to intermittent flow. At points where the superficial gas velocity is estimated to be lower than the critical gas velocity and the flow is considered intermittent or bubbly, the drift-flux model is to be used, which works well in flow patterns where liquid volume is high. Another strength of the model is its ability to capture countercurrent flow. The drift-flux homogeneous-like approach makes the process simple and continuous and simultaneously captures some physics through slippage. For points where the superficial gas velocity is greater than the critical gas velocity, two-fluid modeling is to be used.

Development of Clinically Validated Artificial Intelligence Model for Detecting ST-segment Elevation Myocardial Infarction

ObjectiveAlthough the importance of primary percutaneous coronary intervention (PCI) has been emphasized for ST-segment elevation myocardial infarction (STEMI), the appropriateness of the cardiac catheterization laboratory (CCL) activation remains suboptimal. This study aimed to develop a precise artificial intelligence (AI) model for diagnosis of STEMI and accurate CCL activation. MethodsWe used electrocardiography (ECG) waveform data from a prospective PCI registry in Korea in this study. Two independent board-certified cardiologists established a criterion standard (STEMI or Not-STEMI) for each ECG based on corresponding coronary angiography data. We developed a deep ensemble model by combining five convolutional neural networks. In addition, we performed clinical validation based on a symptom-based ECG dataset, comparisons with clinical physicians, and external validation. ResultsWe used 18,697 ECGs for the model development dataset and 1,745 (9.3%) were STEMI. The AI model achieved an accuracy of 92.1%, sensitivity of 95.4% and specificity of 91.8 %. The performances of the AI model were well-balanced and outstanding in the clinical validation, comparison with clinical physicians, and the external validation. ConclusionsThe deep ensemble AI model showed a well-balanced and outstanding performance. As visualized with the Grad-CAM, the AI model has a reasonable explainability. Further studies with prospective validation regarding clinical benefit in a real-world setting should be warranted.

Deep learning with mixup augmentation for improved pore detection during additive manufacturing

In additive manufacturing (AM), process defects such as keyhole pores are difficult to anticipate, affecting the quality and integrity of the AM-produced materials. Hence, considerable efforts have aimed to predict these process defects by training machine learning (ML) models using passive measurements such as acoustic emissions. This work considered a dataset in which keyhole pores of a laser powder bed fusion (LPBF) experiment were identified using X-ray radiography and then registered both in space and time to acoustic measurements recorded during the LPBF experiment. Due to AM’s intrinsic process controls, where a pore-forming event is relatively rare, the acoustic datasets collected during monitoring include more non-pores than pores. In other words, the dataset for ML model development is imbalanced. Moreover, this imbalanced and sparse data phenomenon remains ubiquitous across many AM monitoring schemes since training data is nontrivial to collect. Hence, we propose a machine learning approach to improve this dataset imbalance and enhance the prediction accuracy of pore-labeled data. Specifically, we investigate how data augmentation helps predict pores and non-pores better. This imbalance is improved using recent advances in data augmentation called Mixup, a weak-supervised learning method. Convolutional neural networks (CNNs) are trained on original and augmented datasets, and an appreciable increase in performance is reported when testing on five different experimental trials. When ML models are trained on original and augmented datasets, they achieve an accuracy of 95% and 99% on test datasets, respectively. We also provide information on how dataset size affects model performance. Lastly, we investigate the optimal Mixup parameters for augmentation in the context of CNN performance.

Development and external validation of a head and neck cancer risk prediction model.

Head and neck cancer (HNC) incidence is on the rise, often diagnosed at late stage and associated with poor prognoses. Risk prediction tools have a potential role in prevention and early detection. The IARC-ARCAGE European case-control study was used as the model development dataset. A clinical HNC risk prediction model using behavioral and demographic predictors was developed via multivariable logistic regression analyses. The model was then externally validated in the UK Biobank cohort. Model performance was tested using discrimination and calibration metrics. 1926 HNC cases and 2043 controls were used for the development of the model. The development dataset model including sociodemographic, smoking, and alcohol variables had moderate discrimination, with an area under curve (AUC) value of 0.75 (95% CI, 0.74-0.77); the calibration slope (0.75) and tests were suggestive of good calibration. 384 616 UK Biobank participants (with 1177 HNC cases) were available for external validation of the model. Upon external validation, the model had an AUC of 0.62 (95% CI, 0.61-0.64). We developed and externally validated a HNC risk prediction model using the ARCAGE and UK Biobank studies, respectively. This model had moderate performance in the development population and acceptable performance in the validation dataset. Demographics and risk behaviors are strong predictors of HNC, and this model may be a helpful tool in primary dental care settings to promote prevention and determine recall intervals for dental examination. Future addition of HPV serology or genetic factors could further enhance individual risk prediction.

DREAMER: a computational framework to evaluate readiness of datasets for machine learning

BackgroundMachine learning (ML) has emerged as the predominant computational paradigm for analyzing large-scale datasets across diverse domains. The assessment of dataset quality stands as a pivotal precursor to the successful deployment of ML models. In this study, we introduce DREAMER (Data REAdiness for MachinE learning Research), an algorithmic framework leveraging supervised and unsupervised machine learning techniques to autonomously evaluate the suitability of tabular datasets for ML model development. DREAMER is openly accessible as a tool on GitHub and Docker, facilitating its adoption and further refinement within the research community..ResultsThe proposed model in this study was applied to three distinct tabular datasets, resulting in notable enhancements in their quality with respect to readiness for ML tasks, as assessed through established data quality metrics. Our findings demonstrate the efficacy of the framework in substantially augmenting the original dataset quality, achieved through the elimination of extraneous features and rows. This refinement yielded improved accuracy across both supervised and unsupervised learning methodologies.ConclusionOur software presents an automated framework for data readiness, aimed at enhancing the integrity of raw datasets to facilitate robust utilization within ML pipelines. Through our proposed framework, we streamline the original dataset, resulting in enhanced accuracy and efficiency within the associated ML algorithms.

Hybrid machine learning model based predictions for properties of poly(2-hydroxyethyl methacrylate)-poly(vinyl alcohol) composite cryogels embedded with bacterial cellulose

Supermacroporous composite cryogels with enhanced adjustable functionality have received extensive interest in bioseparation, tissue engineering, and drug delivery. However, the variations in their components significantly impactfinal properties. This study presents a two-step hybrid machine learning approach for predicting the properties of innovative poly(2-hydroxyethyl methacrylate)-poly(vinyl alcohol) composite cryogels embedded with bacterial cellulose (pHEMA-PVA-BC) based on their compositions. By considering the ratios of HEMA (1.0–22.0 wt%), PVA (0.2–4.0 wt%), poly(ethylene glycol) diacrylate (1.0–4.5 wt%), BC (0.1–1.5 wt%), and water (68.0–96.0 wt%) as investigational variables, overlay sampling uniform design (OSUD) was employed to construct a high-quality dataset for model development. The random forest (RF) model was used to classify the preparation conditions. Then four models of artificial neural network, RF, gradient boosted regression trees (GBRT), and XGBoost were developed to predict the basic properties of the composite cryogels. The results showed that the RF model achieved an accurate three-class classification of preparation conditions. Among the four models, the GBRT model exhibited the best predictive performance of the basic properties, with the mean absolute percentage error of 16.04 %, 0.85 %, and 2.44 % for permeability, effective porosity, and height of theoretical plate (1.0 cm/min), respectively. Characterization results of the representative pHEMA-PVA-BC composite cryogel showed an effective porosity of 81.01 %, a permeability of 1.20 × 10−12 m2, and a range of height of theoretical plate between 0.40–0.49 cm at flow velocities of 0.5–3.0 cm/min. These indicate that the pHEMA-PVA-BC cryogel was an excellent material with supermacropores, low flow resistance and high mass transfer efficiency. Furthermore, the model output demonstrates that the alteration of the proportions of PVA (0.2–3.5 wt%) and BC (0.1–1.5 wt%) components in composite cryogels resulted in significant changes in the material basic properties. This work represents an attempt to efficiently design and prepare target composite cryogels using machine learning and providing valuable insights for the efficient development of polymers.

Machine learning ground motion models for a critical site with long-term earthquake records

Ground motion models derived using data from widespread stations and earthquakes induce significant aleatory variability when applied to predict ground motions for a single site. This issue can be even worse when the sites of interest show complex site response characteristics and do not have enough similar sites in the ground motion model development dataset. This work provides a machine-learning-based ground motion modeling framework for sites with numerous earthquake recordings. The Onahama Port Array, a liquefiable site in Japan that has recorded more than 1200 earthquakes in the last 25 years, was selected as a case study. The gradient boosting method was employed to predict response spectra with periods from 0.01 to 10 s for earthquakes with Japan Meteorological Agency (JMA) magnitude ranging from 2.4 to 9.0. Different predictor combinations were tested, and two gradient boosting models were recommended: the basic model that requires earthquake magnitude and epicentral distance as inputs, and the optimal model that requires three additional inputs, focal depth, azimuth, and rainfall. The basic and optimal gradient boosting models have the root mean square error of 0.004 and 0.001 g, average r2 of 5-fold cross validation of 0.915 and 0.983, respectively. The results shed light on seismic hazard assessment for critical sites with long-term earthquake recordings.

UAV and Satellite Synergies for Mapping Grassland Aboveground Biomass in Hulunbuir Meadow Steppe.

Aboveground biomass (AGB) is an important indicator of the grassland ecosystem. It can be used to evaluate the grassland productivity and carbon stock. Satellite remote sensing technology is useful for monitoring the dynamic changes in AGB across a wide range of grasslands. However, due to the scale mismatch between satellite observations and ground surveys, significant uncertainties and biases exist in mapping grassland AGB from satellite data. This is also a common problem in low- and medium-resolution satellite remote sensing modeling that has not been effectively solved. The rapid development of uncrewed aerial vehicle (UAV) technology offers a way to solve this problem. In this study, we developed a method with UAV and satellite synergies for estimating grassland AGB that filled the gap between satellite observation and ground surveys and successfully mapped the grassland AGB in the Hulunbuir meadow steppe in the northeast of Inner Mongolia, China. First, based on the UAV hyperspectral data and ground survey data, the UAV-based AGB was estimated using a combination of typical vegetation indices (VIs) and the leaf area index (LAI), a structural parameter. Then, the UAV-based AGB was aggregated as a satellite-scale sample set and used to model satellite-based AGB estimation. At the same time, spatial information was incorporated into the LAI inversion process to minimize the scale bias between UAV and satellite data. Finally, the grassland AGB of the entire experimental area was mapped and analyzed. The results show the following: (1) random forest (RF) had the best performance compared with simple regression (SR), partial least squares regression (PLSR) and back-propagation neural network (BPNN) for UAV-based AGB estimation, with an R2 of 0.80 and an RMSE of 76.03 g/m2. (2) Grassland AGB estimation through introducing LAI achieved higher accuracy. For UAV-based AGB estimation, the R2 was improved by an average of 10% and the RMSE was reduced by an average of 9%. For satellite-based AGB estimation, the R2 was increased from 0.70 to 0.75 and the RMSE was decreased from 78.24 g/m2 to 72.36 g/m2. (3) Based on sample aggregated UAV-based AGB and an LAI map, the accuracy of satellite-based AGB estimation was significantly improved. The R2 was increased from 0.57 to 0.75, and the RMSE was decreased from 99.38 g/m2 to 72.36 g/m2. This suggests that UAVs can bridge the gap between satellite observations and field measurements by providing a sufficient training dataset for model development and AGB estimation from satellite data.

Clinical Applications, Challenges, and Recommendations for Artificial Intelligence in Musculoskeletal and Soft-Tissue Ultrasound: AJR Expert Panel Narrative Review.

Artificial intelligence (AI) is increasingly used in clinical practice for musculoskeletal imaging tasks, such as disease diagnosis and image reconstruction. AI applications in musculoskeletal imaging have focused primarily on radiography, CT, and MRI. Although musculoskeletal ultrasound stands to benefit from AI in similar ways, such applications have been relatively underdeveloped. In comparison with other modalities, ultrasound has unique advantages and disadvantages that must be considered in AI algorithm development and clinical translation. Challenges in developing AI for musculoskeletal ultrasound involve both clinical aspects of image acquisition and practical limitations in image processing and annotation. Solutions from other radiology subspecialties (e.g., crowdsourced annotations coordinated by professional societies), along with use cases (most commonly rotator cuff tendon tears and palpable soft-tissue masses), can be applied to musculoskeletal ultrasound to help develop AI. To facilitate creation of high-quality imaging datasets for AI model development, technologists and radiologists should focus on increasing uniformity in musculoskeletal ultrasound performance and increasing annotations of images for specific anatomic regions. This Expert Panel Narrative Review summarizes available evidence regarding AI's potential utility in musculoskeletal ultrasound and challenges facing its development. Recommendations for future AI advancement and clinical translation in musculoskeletal ultrasound are discussed.

Margin quality analysis for wedge resection of lung cancer and construction of a predictive model.

Insufficient pulmonary wedge resection margin is associated with malignant positive margins and high local recurrence risk for lung cancer. This study aimed to identify the risk factors of insufficient or guideline discordant resection margin distance and establish a predictive model to preoperatively estimate the risk of discordant margin for individual patient. Guideline discordant resection margin was defined as ratio of resection margin distance to tumor size less than one. Patients who had pulmonary malignancies and underwent wedge resection between April 2014 and February 2023 were enrolled and stratified by quality of resection margin. Multivariable logistic regression analysis was employed to identify risk factors of guideline discordant margin and a predictive model was developed. Data from March 2023 to January 2024 were collected for internal validation. A total of 530 patients were included. The incidence of guideline discordant wedge resection margin was 37.2%. Longer tumor's max distance to pleura and larger tumor size were variables associated with increased risk and included in the final model. Preoperative localization and right-side surgery were protective variables in the predictive model. A nomogram was built based on the predictive model. The model showed satisfying predictive performance with a concordance index of 0.720 for the predictive model, and 0.761 for internal validation. The goodness-if-fit tests were non-significant for both model development and internal validation data set. The preoperative predictive model and nomogram show good predictive performance to estimate the risk of guideline discordant wedge resection margin. Individualized surgical plans or preoperative nodule localization can be made for high-risk patients.

Transfer learning to leverage larger datasets for improved prediction of protein stability changes

Amino acid mutations that lower a protein's thermodynamic stability are implicated in numerous diseases, and engineered proteins with enhanced stability can be important in research and medicine. Computational methods for predicting how mutations perturb protein stability are, therefore, of great interest. Despite recent advancements in protein design using deep learning, in silico prediction of stability changes has remained challenging, in part due to a lack of large, high-quality training datasets for model development. Here, we describe ThermoMPNN, a deep neural network trained to predict stability changes for protein point mutations given an initial structure. In doing so, we demonstrate the utility of a recently released megascale stability dataset for training a robust stability model. We also employ transfer learning to leverage a second, larger dataset by using learned features extracted from ProteinMPNN, a deep neural network trained to predict a protein's amino acid sequence given its three-dimensional structure. We show that our method achieves state-of-the-art performance on established benchmark datasets using a lightweight model architecture that allows for rapid, scalable predictions. Finally, we make ThermoMPNN readily available as a tool for stability prediction and design.