Separate Data Sets Research Articles

Gene‐expression programming‐based models for comprehensive evaluation of bond strength and critical slip of various types of FRP bars in concrete

AbstractThis study aims to present a novel approach to assess the bond behavior of various types of FRP bars in concrete, based on an extensive dataset. The novelty of this research lies in evaluating both the bond strength and the critical slip, which forms the basis for creating a bond model between the bars and concrete. Furthermore, the model is constructed using advanced machine learning techniques, specifically gene‐expression programming (GEP), and its reliability is ensured by relying on a more extensive and diverse experimental dataset compared to previous studies. To train and validate the GEP model, data were collected from experiments involving 793 test specimens, categorized into five groups based on surface treatments: smooth (Sm), sand‐coated (SC), helically wrapped (HW), grooved (Gr), and a hybrid surface (Hbr) achieved by combining SC and HW. The model also considers a wide range of other parameters, including compressive strength of concrete and diameter, tensile strength, elastic modulus, and embedment length of FRP bars. The verification results using a separate testing dataset demonstrate that the proposed models accurately predict the bond strength and critical slip of FRP bars in concrete, achieving R2 values of 0.794 and 0.88, respectively. Comprehensive parametric analyses are performed to clarify the influence of key parameters on the bond and slip behavior of FRP bars in concrete. The analysis results confirm the critical role of the surface type and elastic modulus in enhancing bond strength; particularly, among the different surface treatment methods, the HW treatment exhibits the highest bond strength, followed by the HBr, Gr, and SC surface types.

Enhance registration precision of transmission breast images utilizing improved Levenberg-Marquardt optimization algorithm with normalized cross-correlation.

Transmission imaging may become a possible advance for breast cancer screening with non-invasive, cost-effective, and radiation-free approaches for early detection. Frame accumulation can successfully eliminate the issue of low SNR, low grayscale and poor quality in transmission image. However, frame accumulation accuracy can be diminished because of inherent human body instability during image acquisition and the light absorption characteristics of breast tissue, resulting in distorted and misplaced image sequences. Therefore, improved Levenberg-Marquardt optimization algorithm with normalized cross-correlation is used as an innovative approach to rectify image sequences before frame accumulation processing. Two separate sets of data, showing breast images with and without markers, were collected using a halogen bulb and a mobile phone camera to validate the suggested method. The approach includes coarse registration utilizing normalized cross-correlation for initial value estimation, followed by fine registration using Levenberg-Marquardt algorithm. The results demonstrate a notable improvement in both accuracy of registration and frame accumulation quality. Specifically, the registration speed showed a remarkable increase, being 8.7 times faster, especially prominent in images that included markers. These images displayed normalized cross-correlation values reaching up to 0.99. The research emphasizes the future potential of the suggested method in overcoming the image quality challenges associated with breast transmission imaging, providing a significant milestone toward more accurate and efficient early breast cancer screening methods. Moreover, transmission imaging systems for the breast have been developed to verify the safety and effectiveness of the implemented technology.

Development and Validation of a Simplified Financial Toxicity Screening Tool for Use in Clinical Practice.

Cancer-related financial toxicity occurs frequently and is a key driver of inequities in access to care and disparities in treatment outcomes. Current practices to screen for financial toxicity are inconsistent because of the lack of a validated and clinically integrated screening tool. This analysis aimed to create and assess an abbreviated version of the validated Comprehensive Score for Financial Toxicity (COST) tool, a measure of financial toxicity used for research purposes, which could easily be added into often-lengthy clinical screening workflows. At an urban comprehensive cancer center with suburban satellite locations, a financial toxicity screening quality improvement project was conducted from June 2022 to August 2023 as part of routine clinical care: 57,526 longitudinal COST surveys were completed by 38,249 patients with cancer. An iterative algorithm selected the items with highest correlation with the total score. Using a separate validation data set, positive and negative predictive values (PPV and NPV, respectively) of the abbreviated tool (two-item) were assessed against the full COST score, with varying risk thresholds. Inclusion of two COST questions (Q3: "I worry about the financial problems I will have in the future as a result of my illness or treatment"; Q6: "I am satisfied with my current financial situation") yielded a score that had a correlation of 0.922 with the full instrument score. For the two-item scale, PPV ranged from 74% to 91%, and NPV ranged from 91% to 98% when compared with the full COST tool. This analysis of a large data set finds that a simplified COST tool has high predictive value when compared with the full validated measure. An abbreviated COST measure of two questions is suitable for implementation into clinical screening workflows.

Predicting macular hole surgery outcomes: Integrating preoperative OCT features with supervised machine learning statistical models.

To evaluate various supervised machine learning (ML) statistical models to predict anatomical outcomes after macular hole (MH) surgery using preoperative optical coherence tomography (OCT) features. This retrospective study analyzed OCT data from idiopathic MH eyes at baseline and at 1-month post-surgery. The dataset was split 80:20 between training and testing. XLSTAT® statistical software (Lumivero, USA) was used to train different ML models on 10°CT parameters: prefoveal posterior cortical vitreous status, epiretinal membrane, intraretinal cysts, foveal retinal pigment epithelium hyperreflectivity, MH basal diameter, MH area (MHA), hole-forming factor, MH index, tractional hole index, and diameter hole index. The most effective statistical model was identified and was further assessed for accuracy, sensitivity, and specificity on a separate testing dataset. Six ML statistical models were trained on 33,260°CT data points from 3326°CT images of 308 operated MH (300 patients) eyes. Following training and internal validation, the random forest (RF) model achieved the highest accuracy (0.92), precision (0.94), recall (0.97), and F-score (0.96), and lowest misclassification rate. RF model identified the MHA index as the best predictor of post-surgical anatomical success. Following external testing, the RF model confirmed the highest accuracy and lowest misclassification rate (8.8%). ML-based statistical models can be used to predict MH status after surgery. The RF model was the most accurate ML model, and the MHA index was the best predictor of postoperative hole closure after surgery based on preoperative OCT parameters. These predictions may help with future surgical planning for MH patients.

Identification of molecular and cellular infection response biomarkers associated with anthrax infection through comparative analysis of gene expression data.

Bacillus anthracis, a gram-positive bacillus capable of forming spores, causes anthrax in mammals, including humans, and is recognized as a potential biological weapon agent. The diagnosis of anthrax is challenging due to variable symptoms resulting from exposure and infection severity. Despite the availability of a licensed vaccines, their limited long-term efficacy underscores the inadequacy of current human anthrax vaccines, highlighting the urgent need for next-generation alternatives. Our study aimed to identify molecular biomarkers and essential biological pathways for the early detection and accurate diagnosis of human anthrax infection. Using a comparative analysis of Bacillus anthracis gene expression data from the Gene Expression Omnibus (GEO) database, this cost-effective approach enables the identification of shared differentially expressed genes (DEGs) across separate microarray datasets without additional hybridization. Three microarray datasets (GSE34407, GSE14390, and GSE12131) of B. anthracis-infected human cell lines were analyzed via the GEO2R tool to identify shared DEGs. We identified 241 common DEGs (70 upregulated and 171 downregulated) from cell lines treated similarly to lethal toxins. Additionally, 10 common DEGs (5 upregulated and 5 downregulated) were identified across different treatments (lethal toxins and spores) and cell lines. Network meta-analysis identified JUN and GATAD2A as the top hub genes for overexpression, and NEDD4L and GULP1 for underexpression. Furthermore, prognostic analysis and SNP detection of the two identified upregulated hub genes were carried out in conjunction with machine learning classification models, with SVM yielding the best classification accuracy of 87.5%. Our comparative analysis of Bacillus anthracis infection revealed striking similarities in gene expression 241 profiles across diverse datasets, despite variations in treatments and cell lines. These findings underscore how anthrax infection activates shared genes across different cell types, emphasizing this approach in the discovery of novel gene markers. These markers offer insights into pathogenesis and may lead to more effective therapeutic strategies. By identifying these genetic indicators, we can advance the development of precise immunotherapies, potentially enhancing vaccine efficacy and treatment outcomes.

An integrated cloud system based serverless android app for generalised tractor drawbar pull prediction model using machine learning

Knowing tractor drawbar pull is crucial to ensure the tractor can handle the required workload efficiently and safely, preventing soil damage and optimising field productivity. The present study proposes a novel approach for tractor drawbar pull prediction by utilising the tractor's geometric parameters and forward speed to develop a cloud-infused, server-less, machine learning-based real-time generalised tractor drawbar pull prediction model for any tractor between the 6-58 kW power range. The drawbar pull prediction models from ANN and six ML algorithms were developed, and the data analysis with hyperparameter tuning concluded that the Extreme Gradient Boosting (XGB) ML model outperformed the other ML models. A reasonable accuracy with R2 = 0.93 and MAPE = 6.77% was achieved using the XGB ML model for a separate validation dataset, which was not used for training. Furthermore, a cloud-based serverless Android App integrated with the XGB ML-based drawbar pull prediction model was developed for real-time tractor drawbar pull prediction and monitoring during tillage operations. The field validation demonstrated the XGB ML model's generalisation ability and effectiveness, with R2 = 0.90 and maximum MAPE of 9.86%. It can be used to simulate and optimize tractor performance, guiding manufacturers in selecting geometric parameters for tractor design.

Improving the Perception of Objects Under Daylight Foggy Conditions in the Surrounding Environment

Autonomous driving (AD) technology has seen significant advancements in recent years; however, challenges remain, particularly in achieving reliable performance under adverse weather conditions such as heavy fog. In response, we propose a multi-class fog density classification approach to enhance the AD system performance. By categorizing fog density into multiple levels (25%, 50%, 75%, and 100%) and generating separate datasets for each class using the CARLA simulator, we improve the perception accuracy for each specific fog density level and analyze the effects of varying fog intensities. This targeted approach offers benefits such as improved object detection, specialized training for each fog class, and increased generalizability. Our results demonstrate enhanced perception of various objects, including cars, buses, trucks, vans, pedestrians, and traffic lights, across all fog densities. This multi-class fog density method is a promising advancement toward achieving reliable AD performance in challenging weather, improving both the precision and recall of object detection algorithms under diverse fog conditions.

Exploratory Analysis of Skin Cancer Dermatoscopic Image Datasets and Classification Methods

Skin cancer is a critical global health issue, where early detection significantly improves treatment outcomes. In this review paper, titled Exploratory Analysis of Skin Cancer Dermatoscopic Image Datasets and Classification Methods, we systematically explore the intersection of artificial intelligence (AI) and skin cancer diagnosis. Our approach began with a detailed literature survey of 40 research studies, providing insights into advancements and challenges in AI-based classification methods. This survey emphasizes the role of machine learning algorithms, particularly convolutional neural networks (CNNs), and datasets such as HAM10000, ISIC 2019, and ISIC 2024 in improving diagnostic performance. Next, we conducted exploratory data analysis (EDA) on the HAM10000, ISIC 2019, and ISIC 2024 datasets, uncovering critical patterns related to lesion distribution, anatomical sites, and demographic factors. These analyses highlight biases and imbalances in the datasets, which are crucial to address for robust model training. Finally, we discuss the creation and evaluation of a machine learning model trained on a separate dataset. Initial experiments revealed challenges such as overfitting and class imbalance. Through advanced data augmentation techniques and the integration of an Augmentor pipeline, we mitigated these issues, achieving improved accuracy and generalizability. This paper provides a comprehensive framework for integrating literature insights, dataset analysis, and iterative model improvement to develop effective AI-based solutions for skin cancer detection. It underscores the importance of addressing dataset biases, adopting diverse datasets, and refining methodologies to advance AI applications in dermatology.

An exploration into the diagnostic capabilities of microRNAs for myocardial infarction using machine learning

BackgroundMicroRNAs (miRNAs) have shown potential as diagnostic biomarkers for myocardial infarction (MI) due to their early dysregulation and stability in circulation after MI. Moreover, they play a crucial role in regulating adaptive and maladaptive responses in cardiovascular diseases, making them attractive targets for potential biomarkers. However, their potential as novel biomarkers for diagnosing cardiovascular diseases requires systematic evaluation.MethodsThis study aimed to identify a miRNA biomarker panel for early-stage MI detection using bioinformatics and machine learning (ML) methods. miRNA expression data were obtained for early-stage MI patients and healthy controls from the Gene Expression Omnibus. Separate datasets were allocated for training and independent testing. Differential expression analysis was performed to identify dysregulated miRNAs in the training set. The least absolute shrinkage and selection operator (LASSO) was applied for feature selection to prioritize relevant miRNAs associated with MI. The selected miRNAs were used to develop ML models including support vector machine, Gradient Boosted, XGBoost, and a hard voting ensemble (HVE).ResultsDifferential expression analysis discovered 99 dysregulated miRNAs in the training set. LASSO feature selection prioritized 21 miRNAs. Ten miRNAs were identified in both the LASSO subset and independent test set. The HVE model trained with the selected miRNAs achieved an accuracy of 0.86 and AUC of 0.83 on the independent test set.ConclusionsAn integrated framework for robust miRNA selection from omics data shows promise for developing accurate diagnostic models for early-stage MI detection. The HVE model demonstrated good performance despite differences between training and test datasets.

Pseudo-spectral angle mapping for pixel and cell classification in highly multiplexed immunofluorescence images.

The rapid development of highly multiplexed microscopy has enabled the study of cells embedded within their native tissue. The rich spatial data provided by these techniques have yielded exciting insights into the spatial features of human disease. However, computational methods for analyzing these high-content images are still emerging; there is a need for more robust and generalizable tools for evaluating the cellular constituents and stroma captured by high-plex imaging. To address this need, we have adapted spectral angle mapping-an algorithm developed for hyperspectral image analysis-to compress the channel dimension of high-plex immunofluorescence (IF) images. Here, we present pseudo-spectral angle mapping (pSAM), a robust and flexible method for determining the most likely class of each pixel in a high-plex image. The class maps calculated through pSAM yield pixel classifications which can be combined with instance segmentation algorithms to classify individual cells. In a dataset of colon biopsies imaged with a 13-plex staining panel, 16 pSAM class maps were computed to generate pixel classifications. Instance segmentations of cells with Cellpose2.0 ( -score of ) were combined with these class maps to provide cell class predictions for 13 cell classes. In addition, in a separate unseen dataset of kidney biopsies imaged with a 44-plex staining panel, pSAM plus Cellpose2.0 ( -score of ) detected a diverse set of 38 classes of structural and immune cells. In summary, pSAM is a powerful and generalizable tool for evaluating high-plex IF image data and classifying cells in these high-dimensional images.

Enhancing Transformer Protection: A Machine Learning Framework for Early Fault Detection

The reliable operation of power transformers is essential for grid stability, yet existing fault detection methods often suffer from inaccuracies and high false alarm rates. This study introduces a machine learning framework leveraging voltage signals for early fault detection. Simulating diverse fault conditions—including single line-to-ground, line-to-line, turn-to-ground, and turn-to-turn faults—on a laboratory-scale three-phase transformer, we evaluated decision trees, support vector machines, and logistic regression models on a dataset of 6000 samples. Decision trees emerged as the most effective, achieving 99.90% accuracy during 5-fold cross-validation and 95% accuracy on a separate test set of 400 unseen samples. Notably, the framework achieved a low false alarm rate of 0.47% on a separate 6000-sample healthy condition dataset. These results highlight the proposed method’s potential to provide a cost-effective, robust, and scalable solution for enhancing transformer fault detection and advancing grid reliability. This demonstrates the efficacy of voltage-based machine learning for transformer diagnostics, offering a practical and resource-efficient alternative to traditional methods.

Self-supervised learning improves robustness of deep learning lung tumor segmentation models to CT imaging differences.

Self-supervised learning (SSL) is an approach to extract useful feature representations from unlabeled data, and enable fine-tuning on downstream tasks with limited labeled examples. Self-pretraining is a SSL approach that uses curated downstream task dataset for both pretraining and fine-tuning. Availability of large, diverse, and uncurated public medical image sets presents the opportunity to potentially create foundation models by applying SSL in the "wild" that are robust to imaging variations. However, the benefit of wild- versus self-pretraining has not been studied for medical imageanalysis. Compare robustness of wild versus self-pretrained models created using convolutional neural network (CNN) and transformer (vision transformer [ViT] and hierarchical shifted window [Swin]) models for non-small cell lung cancer (NSCLC) segmentation from 3D computed tomography (CT) scans. CNN, ViT, and Swin models were wild-pretrained using unlabeled 10,412 3D CTs sourced from the cancer imaging archive and internal datasets. Self-pretraining was applied to same networks using a curated public downstream task dataset (n = 377) of patients with NSCLC. Pretext tasks introduced in self-distilled masked image transformer were used for both pretraining approaches. All models were fine-tuned to segment NSCLC (n = 377 training dataset) and tested on two separate datasets containing early (public n = 156) and advanced stage (internal n = 196) NSCLC. Models were evaluated in terms of: (a) accuracy, (b) robustness to image differences from contrast, slice thickness, and reconstruction kernels, and (c) impact of pretext tasks for pretraining. Feature reuse was evaluated using centered kernel alignment. Wild-pretrained Swin models resulted in higher feature reuse at earlier level layers and increased feature differentiation close to output. Wild-pretrained Swin outperformed self-pretrained models for analyzed imaging acquisitions. Neither ViT nor CNN showed a clear benefit of wild-pretraining compared to self-pretraining. Masked image prediction pretext task that forces networks to learn the local structure resulted in higher accuracy compared to contrastive task that models global image information. Wild-pretrained Swin networks were more robust to analyzed CT imaging differences for lung tumor segmentation than self-pretrained methods. ViT and CNN models did not show a clear benefit for wild-pretraining overself-pretraining.

Oral Health, Inflammation, and Cardiometabolic Factors in the VA Million Veteran Program.

Associations between cardiometabolic comorbidities and self-reported oral health (OH) are often underexplored in large biobank datasets. While these associations are unaffected by dental care access, they could be mediated by immune responses and inflammation. This study assessed the associations between cardiometabolic comorbidities and self-reported OH, periodontitis, and tooth loss using the International Classification of Diseases (ICD) codes in participants from the U.S. Veterans Affairs Million Veteran Program (MVP), adjusting for immune and inflammatory covariates. Data from 154,167 MVP participants were extracted from January 2011 to September 2021, including lifetime cardiometabolic comorbidities, self-reported OH, ICD-coded periodontitis and tooth loss, and laboratory measurements. Multivariate logistic regression analysis was used to calculate the odds ratios of cardiometabolic comorbidities for self-reported OH, periodontitis, and tooth loss, adjusting for demographic, socioeconomic, cardiovascular, and inflammatory (neutrophil and lymphocyte cell counts) risk factors. A separate dataset was used for additional sensitivity analyses, adjusting for serum levels of C-reactive protein and albumin. Complete data were analyzed for 154,167 participants (19%). Most participants (92%) were male and from European ancestry (94%). The mean age was 65.5 y (SD 11.4 y). Ten percent of participants had excellent self-reported OH. Fourteen percent had any periodontitis, and 17% had any tooth loss. Significant associations were found between tooth loss and congestive heart failure (odds ratio [OR], 1.74, P < 0.001) and peripheral vascular diseases (OR, 1.82, P < 0.001). There were also significant associations between congestive heart failure and self-reported OH (excellent versus "poor/fair/good/very good"), with increasing odds as self-reported OH declined (P < 0.001 for trend). These associations remained significant even after sensitivity analyses, albeit with slight attenuation. This study of veterans underscores the important cardiometabolic links of self-reported poor OH and tooth loss, akin to those observed with periodontitis, even after adjusting for potential confounders related to demographics, lifestyle, and inflammation. Exploring cardiometabolic associations with self-reported OH, clinically diagnosed periodontitis, and tooth loss using the ICD in the Veterans Affairs Million Veteran Program, we found significant associations. These associations persisted after adjustment for inflammatory confounders. These findings emphasized the benefit of assessing OH as a vital indicator of overall cardiometabolic health in large-scale biobank studies.

It is a "small world": Relations between performance on five spatial tasks and five mathematical tasks in undergraduate students.

One of the most robust relations in cognition is that between spatial and mathematical reasoning. One important question is whether this relation is domain general or if specific relations exist between performance on different types of spatial tasks and performance on different types of mathematical tasks. In this study, we explore unique relations between performance on five spatial tasks and five mathematical tasks. An exploratory factor analysis conducted on Data Set 1 (N = 391) yielded a two-factor model, one spatial factor and one mathematical factor with significant cross-domain factor loadings. The general two-factor model structure was replicated in a confirmatory factor analysis conducted in a separate data set (N = 364) but the strength of the factor loadings differed by task. Multidimensional scaling and network-based analyses conducted on the combined data sets reveal one spatial cluster, with a central node and one more tightly interconnected mathematical cluster. Both clusters were interconnected via the math task assessing geometry and spatial sense. The unique links identified with the network-based analysis are representative of a "small-world network." These results have theoretical implications for our understanding of the spatial-mathematical relation and practical implications for our understanding of the limitations of transfer between spatial training paradigms and mathematical tasks. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Synthetic Data Generation with Modified Artificial Bee Colony Optimization Algorithm and Statistical Modeling

Machine learning is a powerful decision support system used in analyzing and evaluating real-life data. This system aims to create new solutions and improve performance. Therefore, it is related to the field of data science. There are data on the basis of this relationship The effectiveness of drawing meaningful insights from data depends on the quality of the model's training. To improve this performance, the variety of combinations among the data and the total number of data in the dataset should be increased. But in this topic, insufficient data access, legal regulations, ethical rules, confidentiality procedures, privacy, data sharing restrictions and cost parameters are obstacles. Synthetic data generation is a basic step in the field of data science in order to solve all these problems, improve functionality and provide powerful machine-learning inferences. Therefore, a new synthetic data generation approach consisting of 3 basic stages is proposed in this study. In the first stage, synthetic data production similar to the distribution of the original data was carried out with the modified ABC (Artificial Bee Colony) optimization algorithm. In the second stage, the category information of the independent variables was determined by the statistical evaluation analyzed with regression methods among the artificial data produced. In the third stage, the efficiency and applicability of the artificial data produced were evaluated with supervised machine learning classifiers. As a result of the evaluation, it has been proven that the proposed synthetic data generation approach improves the performance of machine learning classifiers in proportion to the increasing number of data. The decision tree algorithm that showed maximum performance produced success rates of 100%, 92.5%, 100%, 85%, and 66% on 5 separate enriched datasets, respectively.

Surgical Insight-guided Deep Learning for Colorectal Lesion Management.

Colonoscopy stands as a pivotal diagnostic tool in identifying gastrointestinal diseases, including potentially malignant tumors. The procedure, however, faces challenges in the precise identification of lesions during visual inspections. The recent strides in AI and machine learning technologies have opened avenues for enhanced medical imaging analysis, including in the field of colonoscopy. In this study, we developed and evaluated a deep learning (DL) model, ColoNet, for detecting lesions in colonoscopic images. We analyzed 1760 images from 306 patients who underwent colorectal surgery between 2009 and 2022, meeting specific inclusion criteria. These images were used to train and validate ColoNet, employing the YOLOv8 architecture and various data augmentation techniques. Deep learning metrics are assessed via YOLO architecture and trained model diagnostic accuracy was assessed via sensitivity, specifity, positive predictive value, and negative predictive value. Our results from the validation dataset revealed a precision of 0.79604, a recall of 0.78086, an mAP50 of 0.83243, and an mAP50-95 of 0.4439. In addition, on a separate real-time dataset of 91 images consisting both healthy and suspect lesions, ColoNet achieved a sensitivity of 70.73%, specificity of 92.00%, positive predictive value (PPV) of 87.88%, and negative predictive value (NPV) of 79.31%. The positive and negative likelihood ratios were 8.84 and 0.32, respectively, with an overall accuracy of 82.42%. In conclusion, our model has demonstrated promising results, indicating its potential as a valuable tool to assist surgeons during colonoscopy procedures. Its ability to detect suspicious lesions with potential malignancy offers a noteworthy advancement in the early diagnosis and management of colorectal cancers. Further multicentric, prospective research and validation are warranted to fully realize its clinical applicability and impact.

Independent Validation of a HER2-low Focused IHC Scoring System for Enhanced Pathologist Precision and Consistency.

For two decades the ASCO CAP HER2 testing criteria have included 0 and 1+ scores, but this distinction was inconsequential. Now, based on the DESTINY Breast-04 Trial (DB-04) results, for patients with metastatic breast cancer it underpins eligibility for T-DXd treatment. Discerning 0 from 1+ IHC staining is challenging, as HER2 low is not a biologically distinct cancer subset, there are no reference standards or controls and second-tier tests do not apply. Prior reports cast doubt on the reliability of pathologists' IHC scoring, with resulting treatment misalignments. With IRB approval, our group of 9 breast pathologists from 8 Australian laboratories had previously established HER2-low focused scoring conventions, based on the ASCO CAP 2018 HER2 guidelines, and specifying common staining pitfalls. We reported the results of a first set of 60 breast cancers evaluated with these methods 1. After a five-month washout, for the present validation study, we have compiled a second set of 64 HER2 negative invasive breast cancer core biopsies, all assessed with the Ventana 4B5 HER2 assay. We have each scored digitized images of HER2 IHC slides of the cases. Using the majority opinion as the target score, we have calculated our performance metrics. We have compared the results of our performance in set 1 and set 2 to assess the effectiveness of our approach and learning retention. The cases in this validation set included 40 (62.5%) HER2-low, 10 (17.2%) ultralow (UL) and 13 (18.8%) null cancers. Concordance was not achieved in one case. For distinguishing HER2 low from other cancers (UL and null combined) the mean values of our performance metrics were: accuracy 89.58%, sensitivity 90.83%, specificity 87.50%, positive predictive value: 95.63%, negative predictive value 83.59% and Cohen's kappa score 0.81. Comparing these results with our initial study, we have maintained our high level of performance across these parameters. Our mean kappa score is now in the excellent range for concordance. Maintaining high performance across a range of measures in two separate datasets validates the effectiveness of our HER2 low-focused scoring conventions. Having validated our approach, we will use these reference case sets with expert level consensus scores for peer training and updating our national HER2 IHC external quality assurance program. In our ongoing studies, we are also assessing the performance of software algorithms to determine their suitability for pre-screening of HER2 IHC slides.

Deep learning to predict risk of lateral skull base cerebrospinal fluid leak or encephalocele.

Skull base features, including increased foramen ovale (FO) cross-sectional area, are associated with lateral skull base spontaneous cerebrospinal fluid (sCSF) leak and encephalocele. Manual measurement requires skill in interpreting imaging studies and is time consuming. The goal of this study was to develop a fully automated deep learning method for FO segmentation and to determine the predictive value in identifying patients with sCSF leak or encephalocele. A retrospective cohort study at a tertiary care academic hospital of 34 adults with lateral skull base sCSF leak or encephalocele were compared with 815 control patients from 2013-2021. A convolutional neural network (CNN) was constructed for image segmentation of axial computed tomography (CT) studies. Predicted FO segmentations were compared to manual segmentations, and receiver operating characteristic (ROC) curves were constructed. 295 CTs were used for training and validation of the CNN. A separate dataset of 554 control CTs was matched 5:1 on age and sex with the sCSF leak/encephalocele group. The mean Dice score was 0.81. The sCSF leak/encephalocele group had greater mean (SD) FO cross-sectional area compared to the control group, 29.0 (7.7) mm2 versus 24.3 (7.6) mm2 (P = .002, 95% confidence interval 0.02-0.08). The area under the ROC curve was 0.69. CNNs can be used to segment the cross-sectional area of the FO accurately and efficiently. Used together with other predictors, this method could be used as part of a clinical tool to predict the risk of sCSF leak or encephalocele.

Cross-shaped windows transformer with self-supervised pretraining for clinically significant prostate cancer detection in bi-parametric MRI.

Bi-parametric magnetic resonance imaging (bpMRI) has demonstrated promising results in prostate cancer (PCa) detection. Vision transformers have achieved competitive performance compared to convolutional neural network (CNN) in deep learning, but they need abundant annotated data for training. Self-supervised learning can effectively leverage unlabeled data to extract useful semantic representations without annotation and its associated costs. This study proposes a novel self-supervised learning framework and a transformer model to enhance PCa detection using prostate bpMRI. We introduce a novel end-to-end Cross-Shaped windows (CSwin) transformer UNet model, CSwin UNet, to detect clinically significant prostate cancer (csPCa) in prostate bpMRI. We also propose a multitask self-supervised learning framework to leverage unlabeled data and improve network generalizability. Using a large prostate bpMRI dataset (PI-CAI) with 1476 patients, we first pretrain CSwin transformer using multitask self-supervised learning to improve data-efficiency and network generalizability. We then finetune using lesion annotations to perform csPCa detection. We also test the network generalization using a separate bpMRI dataset with 158 patients (Prostate158). Five-fold cross validation shows that self-supervised CSwin UNet achieves 0.888±0.010 aread under receiver operating characterstics curve (AUC) and 0.545±0.060 Average Precision (AP) on PI-CAI dataset, significantly outperforming four comparable models (nnFormer, Swin UNETR, DynUNet, Attention UNet, UNet). On model generalizability, self-supervised CSwin UNet achieves 0.79 AUC and 0.45 AP, still outperforming all other comparable methods and demonstrating good generalization to external data. This study proposes CSwin UNet, a new transformer-based model for end-to-end detection of csPCa, enhanced by self-supervised pretraining to enhance network generalizability. We employ an automatic weighted loss (AWL) to unify pretext tasks, improving representation learning. Evaluated on two multi-institutional public datasets, our method surpasses existing methods in detection metrics and demonstrates good generalization to external data.

RHD-HA-9: A scale for nurses to assess readiness for hospital discharge in older adults following hip arthroplasty-development and psychometric testing.

Designing an instrument to assess discharge readiness post-hip surgery is essential due to trends showing poor patient outcomes, such as pain management issues, mobility challenges, and insufficient home support. A structured assessment tool would help ensure patients are better prepared for recovery, reducing the risk of complications and readmission. To develop and test the psychometric properties of the Readiness for Hospital Discharge Scale (RHD-HA-9) for hip arthroplasty. Items were generated from a comprehensive literature review and individual, face-to-face interviews with experts and patients. A cross-sectional study was conducted across four tertiary governmental hospitals to evaluate the psychometric properties of the scale. Data were collected from a total of 200 older adults who had undergone hip arthroplasty between June 2020 and February 2021. Exploratory Factor Analysis (EFA) was performed on data from 100 older adults to identify the underlying factor structure, followed by Confirmatory Factor Analysis (CFA) on a separate 100-patient dataset to validate the model. The questionnaire's internal consistency, corrected item-total correlations, inter-rater reliability, construct, concurrent, and predictive validity were assessed. The RHD-HA-9 included nine items, categorized into two factors: the physical performance of hip function and barriers to physical activity. EFA and CFA confirmed these factors, explaining 62% of the total variance. Model fit indices were acceptable (CFI = 0.97, TLI = 0.96, SRMR = 0.04), though RMSEA was 0.12. Chi-square was significant (χ2 = 0.056, df = 24, p <0.001). The scale showed excellent internal consistency (Cronbach's α = 0.89) and stability (ICC = 0.94). ROC analysis identified a cutoff of 9.5, with a sensitivity of 90.7%, specificity of 70.6%, and AUC of 0.89. The RHD-HA-9 demonstrated strong psychometric properties for assessing discharge readiness in older adults following hip arthroplasty. It identifies patients who need additional support during their transition home. Nurses can use this tool to accurately assess patient needs and implement effective post-discharge care, thereby enhancing patient outcomes.