Precision And Recall Research Articles

IGATTLDA: Integrative graph attention and transformer-based model for predicting lncRNA-Disease associations.

Long non-coding RNAs (lncRNAs) have emerged as significant contributors to the regulation of various biological processes, and their dysregulation has been linked to a variety of human disorders. Accurate prediction of potential correlations between lncRNAs and diseases is crucial for advancing disease diagnostics and treatment procedures. The authors introduced a novel computational method, iGATTLDA, for the prediction of lncRNA-disease associations. The model utilised lncRNA and disease similarity matrices, with known associations represented in an adjacency matrix. A heterogeneous network was constructed, dissecting lncRNAs and diseases as nodes and their associations as edges. The Graph Attention Network (GAT) is employed to process initial features and corresponding adjacency information. GAT identified significant neighbouring nodes in the network, capturing intricate relationships between lncRNAs and diseases, and generating new feature representations. Subsequently, the transformer captures global dependencies and interactions across the entire sequence of features produced by the GAT. Consequently, iGATTLDA successfully captures complex relationships and interactions that conventional approaches may overlook. In evaluating iGATTLDA, it attained an area under the receiver operating characteristic (ROC) curve (AUC) of 0.95 and an area under the precision recall curve (AUPRC) of 0.96 with a two-layer multilayer perceptron (MLP) classifier. These results were notably higher compared to the majority of previously proposed models, further substantiating the model's efficiency in predicting potential lncRNA-disease associations by incorporating both local and global interactions. The implementation details can be obtained from https://github.com/momanyibiffon/iGATTLDA.

A clinical prediction model for distant metastases of pediatric neuroblastoma: an analysis based on the SEER database.

Patients with distant metastases from neuroblastoma (NB) usually have a poorer prognosis, and early diagnosis is essential to prevent distant metastases. The aim was to develop a machine-learning model for predicting the risk of distant metastasis in patients with neuroblastoma to aid clinical diagnosis and treatment decisions. We built a predictive model using data from the Surveillance, Epidemiology, and End Results (SEER) database from 2010 to 2018 on 1,542 patients with neuroblastoma. Seven machine-learning methods were employed to forecast the likelihood of neuroblastoma distant metastases. Univariate and multivariate logistic regression analyses were used to identify independent risk factors for building machine learning models. Secondly, the subject operating characteristic area under the curve (AUC), Precision-Recall (PR) curves, decision curve analysis (DCA), and calibration curves were used to assess model performance. To further explain the optimal model, the Shapley summation interpretation method (SHAP) was applied. Ultimately, the best model was used to create an online calculator that estimates the likelihood of neuroblastoma distant metastases. The study included 1,542 patients with neuroblastoma, multifactorial logistic regression analysis showed that age, histology, tumor size, tumor grade, primary site, surgery, chemotherapy, and radiotherapy were independent risk factors for distant metastasis of neuroblastoma (P < 0.05). Logistic regression (LR) was found to be the optimal algorithm among the seven constructed, with the highest AUC values of 0.835 and 0.850 in the training and validation sets, respectively. Finally, we used the logistic regression model to build a network calculator for distant metastasis of neuroblastoma. The study developed and validated a machine learning model based on clinical and pathological information for predicting the risk of distant metastasis in patients with neuroblastoma, which may help physicians make clinical decisions.

Morphological Prediction of CPAP Associated Acute Respiratory Instability.

Multiple mechanisms are involved in the pathogenesis of obstructive sleep apnea (OSA). Elevated loop gain is a key target for precision OSA care and may be associated with treatment intolerance when the upper airway is the sole therapeutic target. Morphological or computational estimation of LG is not yet widely available or fully validated - there is a need for improved phenotyping/endotyping of apnea to advance its therapy and prognosis. This study proposes a new algorithm to assess self-similarity as a signature of elevated loop gain using respiratory effort signals and presents its use to predict the probability of acute failure (high residual event counts) of continuous positive airway pressure (CPAP) therapy. Effort signals from 2145 split-night polysomnography studies from the Massachusetts General Hospital were analyzed for SS and used to predict acute CPAP therapy effectiveness. Logistic regression models were trained and evaluated using 5-fold cross-validation. Receiver operating characteristic (ROC) and precision-recall (PR) curves with AUC values of 0.82 and 0.84, respectively, were obtained. Self-similarity combined with the central apnea index (CAI) and hypoxic burden outperformed CAI alone. Even in those with a low CAI by conventional scoring criteria or only mild desaturation, SS was related to poor therapy outcomes. The proposed algorithm for assessing SS as a measure of expressed high loop gain is accurate, non-invasive, and has the potential to improve phenotyping/endotyping of apnea, leading to more precise sleep apnea treatment strategies.

A high-available segmentation algorithm for corn leaves and leaf spot disease based on feature fusion

Detection and identification of corn diseases are crucial for disease control, and the segmentation of corn disease leaf image is a key step to achieve this goal. However, the images of diseased leaves in real fields are usually very complex, with characteristics of irregular shapes, blurred boundaries and unsharp background, which poses great challenges to disease prevention. To address this issue, our team constructed a dataset of diseased leaves with 857 images. Additionally, this paper proposes a high-availability segmentation algorithm for corn leaves with leaf spot disease, called SEF-UNet, which uses Res-UNet as the backbone network. The algorithm references SElayer and ELA (Efficient Local Attention). Simultaneously,we implement a feature fusion network that focuses on the output of each layer. Experimental results indicate that the Mean Intersection over Union (mIOU),Mean Pixel Accuracy (mPA), Mean Precision (mPrecision), and Mean Recall (mRecall),metrics of SEF-UNet network reach 92.62%, 95.74%, 96.63% and 95.64%.We compared our proposed method with UNet, Res-UNet, PspNet, DeepLabv3+, DANet, CCNet, Segformer-b3, and SEF-UNet under the same experimental conditions. The results demonstrate that our method significantly improves the accuracy of diseased leaf image segmentation. It provides a reference method for disease monitoring, as well as a technical basis for assessing disease severity.

YOLOv8-Based Drone Detection: Performance Analysis and Optimization

The extensive utilization of drones has led to numerous scenarios that encompass both advantageous and perilous outcomes. By using deep learning techniques, this study aimed to reduce the dangerous effects of drone use through early detection of drones. The purpose of this study is the evaluation of deep learning approaches such as pre-trained YOLOv8 drone detection for security issues. This study focuses on the YOLOv8 model to achieve optimal performance in object detection tasks using a publicly available dataset collected by Mehdi Özel for a UAV competition that is sourced from GitHub. These images are labeled using Roboflow, and the model is trained on Google Colab. YOLOv8, known for its advanced architecture, was selected due to its suitability for real-time detection applications and its ability to process complex visual data. Hyperparameter tuning and data augmentation techniques were applied to maximize the performance of the model. Basic hyperparameters such as learning rate, batch size, and optimization settings were optimized through iterative experiments to provide the best performance. In addition to hyperparameter tuning, various data augmentation strategies were used to increase the robustness and generalization ability of the model. Techniques such as rotation, scaling, flipping, and color adjustments were applied to the dataset to simulate different conditions and variations. Among the augmentation techniques applied to the specific dataset in this study, rotation was found to deliver the highest performance. Blurring and cropping methods were observed to follow closely behind. The combination of optimized hyperparameters and strategic data augmentation allowed YOLOv8 to achieve high detection accuracy and reliable performance on the publicly available dataset. This method demonstrates the effectiveness of YOLOv8 in real-world scenarios, while also highlighting the importance of hyperparameter tuning and data augmentation in increasing model capabilities. To enhance model performance, dataset augmentation techniques including rotation and blurring are implemented. Following these steps, a significant precision value of 0.946, a notable recall value of 0.9605, and a considerable precision–recall curve value of 0.978 are achieved, surpassing many popular models such as Mask CNN, CNN, and YOLOv5.

Machine learning approaches to identify the link between heavy metal exposure and ischemic stroke using the US NHANES data from 2003 to 2018.

There is limited understanding of the link between exposure to heavy metals and ischemic stroke (IS). This research aimed to develop efficient and interpretable machine learning (ML) models to associate the relationship between exposure to heavy metals and IS. The data of this research were obtained from the National Health and Nutrition Examination Survey (US NHANES, 2003-2018) database. Seven ML models were used to identify IS caused by exposure to heavy metals. To assess the strength of the models, we employed 10-fold cross-validation, the area under the curve (AUC), F1 scores, Brier scores, Matthews correlation coefficient (MCC), precision-recall (PR) curves, and decision curve analysis (DCA) curves. Following these tests, the best-performing model was selected. Finally, the DALEX package was used for feature explanation and decision-making visualization. A total of 15,575 participants were involved in this study. The best-performing ML models, which included logistic regression (LR) (AUC: 0.796) and XGBoost (AUC: 0.789), were selected. The DALEX package revealed that age, total mercury in blood, poverty-to-income ratio (PIR), and cadmium were the most significant contributors to IS in the logistic regression and XGBoost models. The logistic regression and XGBoost models showed high efficiency, accuracy, and robustness in identifying associations between heavy metal exposure and IS in NHANES 2003-2018 participants.

YOLOv8-Coal: a coal-rock image recognition method based on improved YOLOv8.

To address issues such as misdetection and omission due to low light, image defocus, and worker occlusion in coal-rock image recognition, a new method called YOLOv8-Coal, based on YOLOv8, is introduced to enhance recognition accuracy and processing speed. The Deformable Convolution Network version 3 enhances object feature extraction by adjusting sampling positions with offsets and aligning them closely with the object's shape. The Polarized Self-Attention module in the feature fusion network emphasizes crucial features and suppresses unnecessary information to minimize irrelevant factors. Additionally, the lightweight C2fGhost module combines the strengths of GhostNet and the C2f module, further decreasing model parameters and computational load. The empirical findings indicate that YOLOv8-Coal has achieved substantial enhancements in all metrics on the coal rock image dataset. More precisely, the values for AP50, AP50:95, and AR50:95 were improved to 77.7%, 62.8%, and 75.0% respectively. In addition, optimal localization recall precision (oLRP) were decreased to 45.6%. In addition, the model parameters were decreased to 2.59M and the FLOPs were reduced to 6.9G. Finally, the size of the model weight file is a mere 5.2 MB. The enhanced algorithm's advantage is further demonstrated when compared to other commonly used algorithms.

Object and Pedestrian Detection on Road in Foggy Weather Conditions by Hyperparameterized YOLOv8 Model

Connected cooperative and automated (CAM) vehicles and self-driving cars need to achieve robust and accurate environment understanding. With this aim, they are usually equipped with sensors and adopt multiple sensing strategies, also fused among them to exploit their complementary properties. In recent years, artificial intelligence such as machine learning- and deep learning-based approaches have been applied for object and pedestrian detection and prediction reliability quantification. This paper proposes a procedure based on the YOLOv8 (You Only Look Once) method to discover objects on the roads such as cars, traffic lights, pedestrians and street signs in foggy weather conditions. In particular, YOLOv8 is a recent release of YOLO, a popular neural network model used for object detection and image classification. The obtained model is applied to a dataset including about 4000 foggy road images and the object detection accuracy is improved by changing hyperparameters such as epochs, batch size and augmentation methods. To achieve good accuracy and few errors in detecting objects in the images, the hyperparameters are optimized by four different methods, and different metrics are considered, namely accuracy factor, precision, recall, precision–recall and loss.

Subject-level spinal osteoporotic fracture prediction combining deep learning vertebral outputs and limited demographic data.

Osteoporotic vertebral fractures are common and morbid. Automated opportunistic screening for incidental vertebral fractures from radiographs, the highest volume imaging modality, could improve osteoporosis detection and management. We consider how to form patient-level fracture predictions and summarization to guide management, using our previously developed vertebral fracture classifier on segmented radiographs from a prospective cohort study of US men (MrOS). We compare the performance of logistic regression (LR) and generalized additive models (GAM) with combinations of individual vertebral scores and basic demographic covariates. Subject-level LR and GAM models were created retrospectively using all fracture predictions or summary variables such as order statistics, adjacent vertebral interactions, and demographic covariates (age, race/ethnicity). The classifier outputs for 8663 vertebrae from 1176 thoracic and lumbar radiographs in 669 subjects were divided by subject to perform stratified fivefold cross-validation. Models were assessed using multiple metrics, including receiver operating characteristic (ROC) and precision-recall (PR) curves. The best model (AUC-ROC = 0.968) was a GAM using the top three maximum vertebral fracture scores and age. Using top-ranked scores only, rather than all vertebral scores, improved performance for both model classes. Adding age, but not ethnicity, to the GAMs improved performance slightly. Maximal vertebral fracture scores resulted in the highest-performing models. While combining multiple vertebral body predictions risks decreasing specificity, our results demonstrate that subject-level models maintain good predictive performance. Thresholding strategies can be used to control sensitivity and specificity as clinically appropriate.

Extracting seizure control metrics from clinic notes of patients with epilepsy: A natural language processing approach

ObjectivesMonitoring seizure control metrics is key to clinical care of patients with epilepsy. Manually abstracting these metrics from unstructured text in electronic health records (EHR) is laborious. We aimed to abstract the date of last seizure and seizure frequency from clinical notes of patients with epilepsy using natural language processing (NLP). MethodsWe extracted seizure control metrics from notes of patients seen in epilepsy clinics from two hospitals in Boston. Extraction was performed with the pretrained model RoBERTa_for_seizureFrequency_QA, for both date of last seizure and seizure frequency, combined with regular expressions. We designed the algorithm to categorize the timing of last seizure (“today”, “1–6 days ago”, “1–4 weeks ago”, “more than 1–3 months ago”, “more than 3–6 months ago”, “more than 6–12 months ago”, “more than 1–2 years ago”, “more than 2 years ago”) and seizure frequency (“innumerable”, “multiple”, “daily”, “weekly”, “monthly”, “once per year”, “less than once per year”). Our ground truth consisted of structured questionnaires filled out by physicians. Model performance was measured using the areas under the receiving operating characteristic curve (AUROC) and precision recall curve (AUPRC) for categorical labels, and median absolute error (MAE) for ordinal labels, with 95 % confidence intervals (CI) estimated via bootstrapping. ResultsOur cohort included 1773 adult patients with a total of 5658 visits with reported seizure control metrics, seen in epilepsy clinics between December 2018 and May 2022. The cohort average age was 42 years old, the majority were female (57 %), White (81 %) and non-Hispanic (85 %). The models achieved an MAE (95 % CI) for date of last seizure of 4 (4.00–4.86) weeks, and for seizure frequency of 0.02 (0.02–0.02) seizures per day. ConclusionsOur NLP approach demonstrates that the extraction of seizure control metrics from EHR is feasible allowing for large-scale EHR research.

Artificial intelligence interpretation of touch print smear cytology of testicular specimen from patients with azoospermia.

Identification of mature sperm at microdissection testicular sperm extraction (mTESE) is a crucial step of sperm retrieval to help patients with non-obstructive azoospermia (NOA) proceed to intracytoplasmic sperm injection. Touch print smear (TPS) cytology allows immediate interpretation and prompt sperm identification intraoperatively. In this study, we leverage machine learning (ML) to facilitate TPS reading and conquer the learning curve for new operators. One hundred seventy-six microscopic TPS images from the testicular specimen of patients with azoospermia at Taipei Veterans General Hospital were retrospectively collected, including categories of Sertoli cell, primary spermatocytes, round spermatids, elongated spermatids, immature sperm, and mature sperm. Among them, 118 images were assigned as the training set and 29 images as the validation set. RetinaNet (Lin et al. in IEEE Trans Pattern Anal Mach Intell. 42:318-327, 2020), a one-stage detection framework, was adopted for cell detection. The performance was evaluated at the cell level with average precision (AP) and recall, and the precision-recall (PR) curve was displayed among an independent testing set that contains 29 images that aim to assess the model. The training set consisted of 4772 annotated cells, including 1782 Sertoli cells, 314 primary spermatocytes, 443 round spermatids, 279 elongated spermatids, 504 immature sperm, and 1450 mature sperm. This study demonstrated the performance of each category and the overall AP and recall on the validation set, which were 80.47% and 96.69%. The overall AP and recall were 79.48% and 93.63% on the testing set, while increased to 85.29% and 93.80% once the post-meiotic cells were merged into one category. This study proposed an innovative approach that leveraged ML methods to facilitate the diagnosis of spermatogenesis at mTESE for patients with NOA. With the assistance of ML techniques, surgeons could determine the stages of spermatogenesis and provide timely histopathological diagnosis for infertile males.

IMPLICATIONS OF COMPUTE AND INTERPRETABILITY FOR THE DIFFERENT SDN SECURITY MODELS

The main focus of this project is to design and implement an IDS for SDNs that are currently causing great interest in the networking society. Making the control plane and the data plane modularity, SDNs provide a higher level of flexibility and programming capabilities to manage the network but concurrently budding novel security risks. Several machine learning algorithms are in use such as the Logistic Regression, Decision Tree, Random Forest, etc. , which are used to identify malice within the network. As for the dataset sdn_intrusion. csv, preprocessing steps included were cleaning where missing values were managed, categorical data was encoded as well as handling for outliers existed. This paper applied feature selection by analysing correlation between the input features and the target variable with the aim of enhancing the model accuracy. The models were assessed with cross validation and several metrics like accuracy, Precision, Recall, F1 score, AUC and the results are quantitatively and qualitatively presented in the form of confusion matrices, ROC and Precision Recall curves. As the findings indicate, Extended Trees (Extra Trees Classifier) and Histograms Gradient Boosting models are efficient in identifying intrusions within SDN architectures. In conclusion, the study highlights the necessity of the proper choice of the machine learning algorithm and preliminary data processing for the effective IDS construction in SDNs, which enhances the readiness level of the networks against cyber threats. KEYWORDS: Software Defined Networks, Intrusion Detection System, Machine Learning, Network Security, Logistic Regression, Extra Trees Classifier, Random Forest, LightGBM, Precision-Recall Curve, ROC Curve.

VCF observer: a user-friendly software tool for preliminary VCF file analysis and comparison

BackgroundAdvancements over the past decade in DNA sequencing technology and computing power have created the potential to revolutionize medicine. There has been a marked increase in genetic data available, allowing for the advancement of areas such as personalized medicine. A crucial type of data in this context is genetic variant data which is stored in variant call format (VCF) files. However, the rapid growth in genomics has presented challenges in analyzing and comparing VCF files.ResultsIn response to the limitations of existing tools, this paper introduces a novel web application that provides a user-friendly solution for VCF file analyses and comparisons. The software tool enables researchers and clinicians to perform high-level analysis with ease and enhances productivity. The application’s interface allows users to conveniently upload, analyze, and visualize their VCF files using simple drag-and-drop and point-and-click operations. Essential visualizations such as Venn diagrams, clustergrams, and precision–recall plots are provided to users. A key feature of the application is its support for metadata-based file grouping, accomplished through flexible data matrix uploads, streamlining organization and analysis of user-defined categories. Additionally, the application facilitates standardized benchmarking of VCF files by integrating user-provided ground truth regions and variant lists.ConclusionsBy providing a user-friendly interface and supporting essential visualizations, this software enhances the accessibility of VCF file analysis and assists researchers and clinicians in their scientific inquiries.

Focal cortical dysplasia (type II) detection with multi-modal MRI and a deep-learning framework

Focal cortical dysplasia type II (FCD-II) is a prominent cortical development malformation associated with drug-resistant epileptic seizures that leads to lifelong cognitive impairment. Efficient MRI, followed by its analysis (e.g., cortical abnormality distinction, precise localization assistance, etc.) plays a crucial role in the diagnosis and supervision (e.g., presurgery planning and postoperative care) of FCD-II. Involving machine learning techniques particularly, deep-learning (DL) approaches, could enable more effective analysis techniques. We performed a comprehensive study by choosing six different well-known DL models, three image planes (axial, coronal, and sagittal) of two MRI modalities (T1w and FLAIR), demographic characteristics (age and sex) and clinical characteristics (brain hemisphere and lobes) to identify a suitable DL model for analysing FCD-II. The outcomes show that the DenseNet201 model is more suitable because of its superior classification accuracy, high-precision, F1-score, and large area under the receiver operating characteristic (ROC) curve and precision–recall (PR) curve.

Enhancing Missense Variant Pathogenicity Prediction with MissenseNet: Integrating Structural Insights and ShuffleNet-Based Deep Learning Techniques.

The classification of missense variant pathogenicity continues to pose significant challenges in human genetics, necessitating precise predictions of functional impacts for effective disease diagnosis and personalized treatment strategies. Traditional methods, often compromised by suboptimal feature selection and limited generalizability, are outpaced by the enhanced classification model, MissenseNet (Missense Classification Network). This model, advancing beyond standard predictive features, incorporates structural insights from AlphaFold2 protein predictions, thus optimizing structural data utilization. MissenseNet, built on the ShuffleNet architecture, incorporates an encoder-decoder framework and a Squeeze-and-Excitation (SE) module designed to adaptively adjust channel weights and enhance feature fusion and interaction. The model's efficacy in classifying pathogenicity has been validated through superior accuracy compared to conventional methods and by achieving the highest areas under the Receiver Operating Characteristic (ROC) and Precision-Recall (PR) curves (Area Under the Curve and Area Under the Precision-Recall Curve) in an independent test set, thus underscoring its superiority.

ConvNext Mixer‐Based Encoder Decoder Method for Nuclei Segmentation in Histopathology Images

ABSTRACTHistopathology, vital in diagnosing medical conditions, especially in cancer research, relies on analyzing histopathology images (HIs). Nuclei segmentation, a key task, involves precisely identifying cell nuclei boundaries. Manual segmentation by pathologists is time‐consuming, prompting the need for robust automated methods. Challenges in segmentation arise from HI complexities, necessitating advanced techniques. Recent advancements in deep learning, particularly Convolutional Neural Networks (CNNs), have transformed nuclei segmentation. This study emphasizes feature extraction, introducing the ConvNext Mixer‐based Encoder‐Decoder (CNM‐ED) model. Unlike traditional CNN based models, the proposed CNM‐ED model enables the extraction of spatial and long context features to address the inherent complexities of histopathology images. This method leverages a multi‐path strategy using a traditional CNN architecture as well as different paths focused on obtaining customized long context features using the ConvNext Mixer block structure that combines ConvMixer and ConvNext blocks. The fusion of these diverse features in the final segmentation output enables improved accuracy and performance, surpassing existing state‐of‐the‐art segmentation models. Moreover, our multi‐level feature extraction strategy is more effective than models using self‐attention mechanisms such as SwinUnet and TransUnet, which have been frequently used in recent years. Experimental studies were conducted using five different datasets (TNBC, MoNuSeg, CoNSeP, CPM17, and CryoNuSeg) to analyze the performance of the proposed CNM‐ED model. Comparisons were made with various CNN based models in the literature using evaluation metrics such as accuracy, AJI, macro F1 score, macro intersection over union, macro precision, and macro recall. It was observed that the proposed CNM‐ED model achieved highly successful results across all metrics. Through comparisons with state‐art‐of models from the literature, the proposed CNM‐ED model stands out as a promising advancement in nuclei segmentation, addressing the intricacies of histopathological images. The model demonstrates enhanced diagnostic capabilities and holds the potential for significant progress in medical research.

Predicting the involvement of polyQ- and polyA in protein-protein interactions by their amino acid context

Homorepeats, specifically polyglutamine (polyQ) and polyalanine (polyA), are often implicated in protein-protein interactions (PPIs). So far, a method to predict the participation of homorepeats in protein interactions is lacking. We propose a machine learning approach to identify PPI-involved polyQ and polyA regions within the human proteome based on known interacting regions. Using the dataset of human homorepeats, we identified 157 polyQ and 745 polyA regions potentially involved in PPIs. Machine learning models, trained on amino acid context and homorepeat length, demonstrated high precision (0.90–0.98) but variable recall (0.42–0.85). Random forest outperformed other models (AUC polyQ = 0.686, AUC polyA = 0.732) using the positions surrounding the homorepeat −10 to +10. Integrating paralog information marginally improved predictions but was excluded for model simplicity. Further optimization revealed that for polyQ, using amino acid surrounding positions from −6 to +6 increased AUC to 0.715. For polyA, no improvement was found. Incorporating coiled coil overlap information enhanced polyA predictions (AUC = 0.745) but not polyQ. Finally, we applied these models to predict PPI involvement across all polyQ and polyA regions, identifying potential interactions. Case studies illustrated the method's predictive capacity, highlighting known interacting regions with high scores and elucidating potential false negatives.

Improving Uncertainty-Error Correspondence in Deep Bayesian Medical Image Segmentation

Increased usage of automated tools like deep learning in medical image segmentation has alleviated the bottleneck of manual contouring. This has shifted manual labour to quality assessment (QA) of automated contours which involves detecting errors and correcting them. A potential solution to semi-automated QA is to use deep Bayesian uncertainty to recommend potentially erroneous regions, thus reducing time spent on error detection. Previous work has investigated the correspondence between uncertainty and error, however, no work has been done on improving the “utility” of Bayesian uncertainty maps such that it is only present in inaccurate regions and not in the accurate ones. Our work trains the FlipOut model with the Accuracy-vs-Uncertainty (AvU) loss which promotes uncertainty to be present only in inaccurate regions. We apply this method on datasets of two radiotherapy body sites, c.f. head-and-neck CT and prostate MR scans. Uncertainty heatmaps (i.e. predictive entropy) are evaluated against voxel inaccuracies using Receiver Operating Characteristic (ROC) and Precision-Recall (PR) curves. Numerical results show that when compared to the Bayesian baseline the proposed method successfully suppresses uncertainty for accurate voxels, with similar presence of uncertainty for inaccurate voxels. Code to reproduce experiments is available at &lt;a href='https://github.com/prerakmody/bayesuncertainty-error-correspondence'&gt;https://github.com/prerakmody/bayesuncertainty-error-correspondence&lt;/a&gt;

Deep-learning-based method for the segmentation of ureter and renal pelvis on non-enhanced CT scans

This study aimed to develop a deep-learning (DL) based method for three-dimensional (3D) segmentation of the upper urinary tract (UUT), including ureter and renal pelvis, on non-enhanced computed tomography (NECT) scans. A total of 150 NECT scans with normal appearance of the left UUT were chosen for this study. The dataset was divided into training (n = 130) and validation sets (n = 20). The test set contained 29 randomly chosen cases with computed tomography urography (CTU) and NECT scans, all with normal appearance of the left UUT. An experienced radiologist marked out the left renal pelvis and ureter on each scan. Two types of frameworks (entire and sectional) with three types of DL models (basic UNet, UNet3 + and ViT-UNet) were developed, and evaluated. The sectional framework with basic UNet model achieved the highest mean precision (85.5%) and mean recall (71.9%) on the test set compared to all other tested methods. Compared with CTU scans, this method had higher axial UUT recall than CTU (82.5% vs 69.1%, P < 0.01). This method achieved similar or better visualization of UUT than CTU in many cases, however, in some cases, it exhibited a non-ignorable false-positive rate. The proposed DL method demonstrates promising potential in automated 3D UUT segmentation on NECT scans. The proposed DL models could remarkably improve the efficiency of UUT reconstruction, and have the potential to save many patients from invasive examinations such as CTU. DL models could also serve as a valuable complement to CTU.

Fund transfer fraud detection: Analyzing irregular transactions and customer relationships with self-attention and graph neural networks

This paper presents a method for identifying fraudulent fund transfers using real bank data, analyzing customer information, transactional activities, and customer relationships. The preprocessing step transforms high-dimensional, irregular transaction time series into regular time series, then further compresses them in a latent space using a self-attention-based autoencoder. To address the scarcity of fraudulent data samples and mitigate training issues caused by data imbalance, various deep generative models, including the conditional variational autoencoder and Wasserstein generative adversarial network, are applied to generate additional fraudulent raw data and augment fraud samples in latent space. The reparameterization trick is integrated into the encoder–decoder structure to boost the model’s generative capabilities. Additionally, a Graph Neural Network (GNN) is used to model customer relationships. The proposed approach utilizes end-to-end learning, integrating the autoencoder’s reconstruction loss, KL divergence loss (when reparameterization trick is applied), and classification loss for fraud detection. To optimize computational resources, neighborhood sampling for GNN is combined with mini-batch training for the autoencoder, improving both training efficiency and model reliability. Comprehensive experiments demonstrate the effectiveness of the proposed fusion network, highlighting the importance of each component and preprocessing step. For example, the areas under the precision–recall curves for fraud detection show notable improvements in our model. For suspicious transactions identified by the bank’s rules, other models range from 0.66% to 22.15%, while our model reached 27%. For non-suspicious transactions, other models range from 2.53% to 22.00%, with our model achieving 22.90%. This model has potential for wider applications in anomaly detection, particularly in datasets with irregular time series and complex customer relationships.