Test Dataset Research Articles

Application of HP-LSTM Models for Groundwater Level Prediction in Karst Regions: A Case Study in Qingzhen City

Groundwater serves as an indispensable global resource, essential for agriculture, industry, and the urban water supply. Predicting the groundwater level in karst regions presents notable challenges due to the intricate geological structures and fluctuating climatic conditions. This study examines Qingzhen City, China, introducing an innovative hybrid model, the Hodrick–Prescott (HP) filter–Long Short-Term Memory (LSTM) network (HP-LSTM), which integrates the HP filter with the LSTM network to enhance the precision of groundwater level forecasting. By attenuating short-term noise, the HP-LSTM model improves the long-term trend prediction accuracy. Findings reveal that the HP-LSTM model significantly outperformed the conventional LSTM, attaining R2 values of 0.99, 0.96, and 0.98 on the training, validation, and test datasets, respectively, in contrast to LSTM values of 0.92, 0.76, and 0.95. The HP-LSTM model achieved an RMSE of 0.0276 and a MAPE of 2.92% on the test set, significantly outperforming the LSTM model (RMSE: 0.1149; MAPE: 9.14%) in capturing long-term patterns and reducing short-term fluctuations. While the LSTM model is effective at modeling short-term dynamics, it is more prone to noise, resulting in greater prediction errors. Overall, the HP-LSTM model demonstrates superior robustness for long-term groundwater level prediction, whereas the LSTM model may be better suited for scenarios requiring rapid adaptation to short-term variations. Selecting an appropriate model tailored to specific predictive needs can thus optimize groundwater management strategies.

Detection of weeds in vegetables using image classification neural networks and image processing

Weed management presents a major challenge to vegetable growth. Accurate identification of weeds is essential for automated weeding. However, the wide variety of weed types and their complex distribution creates difficulties in rapid and accurate weed detection. In this study, instead of directly applying deep learning to identify weeds, we first created grid cells on the input images. Image classification neural networks were utilized to identify the grid cells containing vegetables and exclude them from further analysis. Finally, image processing technology was employed to segment the non-vegetable grid images based on their color features. The background grid cells, which contained no green pixels, were identified, while the remaining cells were labeled as weed cells. EfficientNet, GoogLeNet, and ResNet models achieved overall accuracies of over 0.956 in identifying vegetables in the testing dataset, demonstrating exceptional identification performance. Among these models, the ResNet model exhibited the highest computational efficiency, with a classification time of 12.76 ms per image and a corresponding frame rate of 80.31 fps, satisfying the requirement for real-time weed detection. Effectively identifying vegetables and differentiating weeds from soil significantly reduces the complexity of weed detection and improves its accuracy.

Alzheimer's disease image classification based on enhanced residual attention network.

With the increasing number of patients with Alzheimer's Disease (AD), the demand for early diagnosis and intervention is becoming increasingly urgent. The traditional detection methods for Alzheimer's disease mainly rely on clinical symptoms, biomarkers, and imaging examinations. However, these methods have limitations in the early detection of Alzheimer's disease, such as strong subjectivity in diagnostic criteria, high detection costs, and high misdiagnosis rates. To address these issues, this study proposes a deep learning model to detect Alzheimer's disease; it is called Enhanced Residual Attention Network (ERAN) that can classify medical images. By combining residual learning, attention mechanism, and soft thresholding, the feature representation ability and classification accuracy of the model have been improved. The accuracy of the model in detecting Alzheimer's disease has reached 99.36%, with a loss rate of only 0.0264. The experimental results indicate that the Enhanced Residual Attention Network has achieved excellent performance on the Alzheimer's disease test dataset, providing strong support for the early diagnosis and treatment of Alzheimer's disease.

Quantification of Argan Oil (Argania spinosa L.) Adulterated with Avocado, Flaxseed, Walnut, and Pumpkin Oils Using Fourier-Transform Infrared Spectroscopy and Advanced Chemometric and Machine Learning Techniques

The increasing trend in the popularity of argan oil (AGO), a multi-beneficial health and cosmetic product, can leave it prone to adulteration. The overall goal of this study was to utilize an attenuated total reflectance Fourier-transform infrared spectroscopic and chemometric methods, including partial least squares (PLS), principal component regression (PCR), and artificial neural network (ANN) for the authentication of AGO in the presence of other oil adulterants, avocado oil (AVO), pumpkin seed oil (PSO), flaxseed oil (FSO), and walnut seed oil (WNO). All three chemometrics methods were able to effectively quantify the FSO adulterant concentration across all statistical models, with the most optimal results in the ANN model as applied in the testing set data (RMSEP = 1.454 %v/v, R2 = 0.821). Comparable results were also obtained for PLS (RMSEP = 1.727 %v/v, R2 = 0.807) and PCR (RMSEP = 1.731 %v/v, R2 = 0.846).

Remaining useful life prediction of lithium-ion batteries using a novel particle flow filter framework with grey model

Remaining useful life (RUL) prediction is a crucial aspect of the prognostics health management of lithium-ion batteries (LIBs). Owing to the influence of resampling technology, particle degradation is often observed in the particle filter-based RUL prediction of LIBs, resulting in a low prediction accuracy and large uncertainty. In this paper, a novel particle flow filter with the grey model method (GM-PFF) is proposed to forecast the RUL and state of health of batteries. First, the least squares method is employed to obtain the initial values for double exponential empirical model parameters. Subsequently, the grey model is used to predict the current cycle capacity of LIBs as an observation value for the particle flow filter, solving the inaccurate estimation problem of the state of particle flow filter observation values, and the particle flow filter method is employed to update model parameters. Finally, a test dataset is divided into early, middle, and late stages to predict the RUL of LIBs and obtain the probability distributions. On the CALCE and NASA PCoE LIB dataset, GM-PFF reduces RMSE by 1% compared to PFF, exhibiting a higher prediction accuracy and effectively addressing the particle degradation problem.

Deep Learning Predicts Non-Normal Transmission Distributions in High-Field Asymmetric Waveform Ion Mobility (FAIMS) Directly from Peptide Sequence.

Peptide ion mobility adds an extra dimension of separation to mass spectrometry-based proteomics. The ability to accurately predict peptide ion mobility would be useful to expedite assay development and to discriminate true answers in a database search. There are methods to accurately predict peptide ion mobility through drift tube devices, but methods to predict mobility through high-field asymmetric waveform ion mobility (FAIMS) are underexplored. Here, we successfully model peptide ions' FAIMS mobility using a multi-label classification scheme to account for non-normal transmission distributions. We trained two models from over 100,000 human peptide precursors: a random forest and a long-term short-term memory (LSTM) neural network. Both models had different strengths, and the ensemble average of model predictions produced a higher F2 score than either model alone. Finally, we explored cases where the models make mistakes and demonstrate the predictive performance of F2 = 0.66 (AUROC = 0.928) on a new test data set of nearly 40,000 E. coli peptide ions. The deep learning model is easily accessible via https://faims.xods.org.

A platform combining automatic segmentation and automatic measurement of the maxillary sinus and adjacent structures.

To develop a platform including a deep convolutional neural network (DCNN) for automatic segmentation of the maxillary sinus (MS) and adjacent structures, and automatic algorithms for measuring 3-dimensional (3D) clinical parameters. 175 CBCTs containing 242 MS were used as the training, validating and testing datasets at the ratio of 7:1:2. The datasets contained healthy MS and MS with mild (2-4mm), moderate (4-10mm) and severe (10- mm) mucosal thickening. A DCNN algorithm adopting 2.5D structure was trained for automatic segmentation. Automatic measuring algorithms were further developed to evaluate the clinical reliability of the DCNN. The median Dice Similarity Coefficient (DSC) for the air cavity, mucosa, teeth and maxillary bone segmentation were 0.990, 0.850, 0.961 and 0.953, respectively. The Intra-class Correlation Coefficien (ICC) of all automatic measuring algorithms exceeded 0.975. The 95% confidence interval (95%CI) of all volumetric metric bias were within ± 0.5 cm3, of all 2D metric bias were within ± 1mm. The DCNN also produced satisfying outcome for notably incomplete MS and edentulous alveolar crest. The DCNN provided clinically reliable results. The automatic measuring algorithms could reveal 3D information embedded in CBCT 2D planes on the basis of automatic segmentation. This platform helps dentists to conduct instant 3D reconstruction and automatic measuring of 3D clinical parameters of MS and adjacent structures.

Deep learning for forensic age estimation using orthopantomograms in children, adolescents, and young adults.

Forensic age estimation from orthopantomograms (OPGs) can be performed more quickly and accurately using convolutional neural networks (CNNs), making them an ideal extension to standard forensic age estimation methods. This study evaluates improvements in forensic age prediction for children, adolescents, and young adults by training a custom CNN from a previous study, using a larger, diverse dataset with a focus on dental growth features. 21,814 OPGs from 13,766 individuals aged 1 to under 25 years were utilized. The custom CNN underwent 1000 epochs of training and validation using 16,000 and 4000 OPGs, respectively. The best model was chosen by the least mean absolute error (MAE) and evaluated with an additional test dataset of 1814 independent OPGs. Furthermore, the CNN was applied to OPGs from 15 available forensic age estimations conducted by experts certified by the Study Group on Forensic Age Diagnostics (AGFAD), and the results were compared. A MAE of 0.93 ± 0.81 years and a mean-signed error (MSE) of -0.06 ± 1.23 years were achieved in the test dataset. 63% of predictions were accurate within 1 year, and 95% within 2.5 years. Results of the CNN were comparable to those obtained by experts, effectively highlighting discrepancies in the reported ages of individuals. Using a large and diverse dataset along with custom deep learning techniques, forensic age estimation can be significantly improved, often providing predictions accurate to within 1 year. This approach offers a reliable, robust, and objective complement to standard forensic age estimation methods. Question The potential of custom convolutional neural networks for forensic age estimation, along with a large, diverse dataset, warrants further investigation, offering valuable support to experts. Findings For 1814 test-orthopantomograms, 63% of predictions were accurate within 1 year and 95% within 2.5 years, similar to expert estimates in 15 forensic cases. Clinical relevance Many individuals' fates depend on accurate age estimation. Forensic age estimation can benefit from applying CNN-based methods to further enhance reliability and accuracy.

Data-driven optimisation of process parameters for reducing developed surface area ratio in laser powder bed fusion

Abstract Optimising process parameters is a key category of approaches to improve the surface quality of laser powder bed fusion parts. So far, many optimisation methods have been presented, which provide effective ideas and approaches for improving surface quality. However, these methods all focus on the improvement of $$R_a$$ R a , $$S_a$$ S a , $$R_{sk}$$ R sk , or $$R_{\Delta q}$$ R Δ q . These parameters are sufficient for most applications. They are however not suitable for applications where functional performance is linked with surface area. To quantify the quality of surfaces for these applications, the developed surface area ratio ( $$S_{dr}$$ S dr ) is a more appropriate parameter as it can be used to quantitatively express the exposed functional surface area. In this paper, a data-driven method for optimising five process parameters, namely layer thickness, laser power, hatch spacing, point distance, and exposure time, to reduce the developed surface area ratio of laser powder bed fusion parts is proposed. Firstly, experiments are designed, and actual build and measurement experiments are conducted to acquire a fixed amount of data. A Bayesian ridge regression model for predicting a developed surface area ratio from the five process parameters is then trained and tested and compared with several other machine learning models using the acquired data. After that, optimisation of the five process parameters to reduce developed surface area ratio is carried out using the genetic algorithm, in which the objective function values (developed surface area ratios) are predicted using the established Bayesian ridge regression model. Finally, an additional actual build and measurement experiment is conducted to validate the optimisation. The testing results show that the Bayesian ridge regression model can obtain an average R $$^2$$ 2 score of 0.77 and an average mean absolute error of 8.48 on the testing dataset. The validation results suggest that the developed surface area ratios generated by the optimisation are relatively small, and on average, they are 52.11% smaller than the developed surface area ratio under the process parameters recommended by the used laser powder bed fusion system.

Revolutionizing Citrus Health: AI-Based Detection of Greening Disease and Nutrient Deficiencies in Sweet Orange

This paper explores the application of Artificial Intelligence (AI) techniques for detecting plant diseases, focusing on citrus greening disease and various foliar nutrient deficiencies in sweet orange. Agriculture faces numerous challenges from cultivation to harvesting, including significant yield losses due to disease infections and environmental hazards from excessive use of insecticides and fungicides. As the global population grows, the demand for food is surging, and traditional farming methods fall short in meeting this demand, often degrading soil health through intensive pesticide use. AI offers significant advantages over conventional techniques in disease detection. This study employs AI for visual detection of citrus greening disease and foliar nutrient deficiencies, achieving 87% accuracy on a test dataset of infected, nutrient-deficient, and healthy sweet orange leaves. Performance is evaluated using metrics such as Accuracy, Recall, Precision, and F1-Score. Specifically, Convolutional Neural Network (CNN) architectures, including Visual Geometry Group (VGG-16), are utilized for image-based detection and classification, demonstrating the potential of AI in enhancing plant health monitoring.

Automatic Segmentation of the Nasolacrimal Canal: Application of the nnU-Net v2 Model in CBCT Imaging

Background/Objectives: There are various challenges in the segmentation of anatomical structures with artificial intelligence due to the different structural features of the relevant region/tissue. The aim of this study was to detect the nasolacrimal canal (NLC) using the nnU-Net v2 convolutional neural network (CNN) model in cone beam-computed tomography (CBCT) images and to evaluate the successful performance of the model in automatic segmentation. Methods: CBCT images of 100 patients were randomly selected from the data archive. The raw data were transferred to the 3D Slicer imaging software in DICOM format (Version 4.10.2; MIT, Massachusetts, USA). NLC was labeled using the polygonal type of manual method. The dataset was split into training, validation and test sets in a ratio of 8:1:1. nnU-Net v2 architecture was applied to the training and test datasets to predict and generate appropriate algorithm weight factors. The confusion matrix was used to check the accuracy and performance of the model. As a result of the test, the Dice Coefficient (DC), Intersection over Union (IoU), F1-Score and 95% Hausdorff distance (95% HD) metrics were calculated. Results: By testing the model, DC, IoU, F1-Scores and 95% HD metric values were found to be 0.8465, 0.7341, 0.8480 and 0.9460, respectively. According to the data obtained, the receiver-operating characteristic (ROC) curve was drawn and the AUC value under the curve was determined to be 0.96. Conclusions: These results showed that the proposed nnU-Net v2 model achieves NLC segmentation on CBCT images with high precision and accuracy. The automated segmentation of NLC may assist clinicians in determining the surgical technique to be used to remove lesions, especially those affecting the anterior wall of the maxillary sinus.

A Review of Simultaneous Localization and Mapping for the Robotic-Based Nondestructive Evaluation of Infrastructures

The maturity of simultaneous localization and mapping (SLAM) methods has now reached a significant level that motivates in-depth and problem-specific reviews. The focus of this study is to investigate the evolution of vision-based, LiDAR-based, and a combination of these methods and evaluate their performance in enclosed and GPS-denied (EGD) conditions for infrastructure inspection. This paper categorizes and analyzes the SLAM methods in detail, considering the sensor fusion type and chronological order. The paper analyzes the performance of eleven open-source SLAM solutions, containing two visual (VINS-Mono, ORB-SLAM 2), eight LiDAR-based (LIO-SAM, Fast-LIO 2, SC-Fast-LIO 2, LeGO-LOAM, SC-LeGO-LOAM A-LOAM, LINS, F-LOAM) and one combination of the LiDAR and vision-based method (LVI-SAM). The benchmarking section analyzes accuracy and computational resource consumption using our collected dataset and a test dataset. According to the results, LiDAR-based methods performed well under EGD conditions. Contrary to common presumptions, some vision-based methods demonstrate acceptable performance in EGD environments. Additionally, combining vision-based techniques with LiDAR-based methods demonstrates superior performance compared to either vision-based or LiDAR-based methods individually.

Unsupervised defect detection for solar photovoltaic cells based on convolutional autoencoder

ABSTRACT Solar photovoltaic (PV) cells are inevitably subject to defects during the production process, affecting their power generation efficiency and life. Electroluminescence (EL) imaging is the mainstream non-destructive method for PV cell defect detection. Aiming at PV cell EL images, an unsupervised defect detection method was proposed. Specifically, an unsupervised convolutional autoencoder (CAE), the scale structure perception convolutional autoencoder (SSP_CAE), was constructed, whose Squeeze-and-Excitation Attention (SE Attention) and skip connections avoid the blurring of image structure information and the loss of pixel-level details in the encoding and decoding process. Furthermore, to balance the global and local information of the image, a scale perception loss function called SP_SSIM was proposed for model training. The defect segmentation was achieved by using Otsu thresholding method to binarize the obtained Mean Absolute Error (MSE) residual heat image in the testing stage of the model. Finally, the experiments were performed on the test dataset and the experimental results showed that the proposed SSP_CAE can effectively detect PV cell defects. The experimentally obtained defect detection performance metrics Precision, Recall, IoU, F1-score and AUROC values were 0.739, 0.886, 0.723, 0.764 and 0.841, respectively. Compared with other classical methods, the proposed SSP_CAE had a better comprehensive performance for defect detection.

IchthyNet: An Ensemble Method for the Classification of In Situ Marine Zooplankton Shadowgraph Images

This study explores the use of machine learning for the automated classification of the ten most abundant groups of marine organisms (in the size range of 5–12 cm) plus marine snow found in the ecosystem of the U.S. east coast. Images used in this process were collected using a shadowgraph imaging system on a towed, undulating platform capable of collecting continuous imagery over large spatiotemporal scales. As a large quantity (29,818,917) of images was collected, the task of locating and identifying all imaged organisms could not be efficiently achieved by human analysis alone. Several tows of data were collected off the coast of Delaware Bay. The resulting images were then cleaned, segmented into regions of interest (ROIs), and fed through three convolutional neural networks (CNNs): VGG-16, ResNet-50, and a custom model created to find more high-level features in this dataset. These three models were used in a Random Forest Classifier-based ensemble approach to reach the best identification fidelity. The networks were trained on a training set of 187,000 ROIs augmented with random rotations and pixel intensity thresholding to increase data variability and evaluated against two datasets. While the performance of each individual model is examined, the best approach is to use the ensemble, which performed with an F1-score of 98% and an area under the curve (AUC) of 99% on both test datasets while its accuracy, precision, and recall fluctuated between 97% and 98%.

Deep learning analyses of splicing variants identify the link of PCP4 with amyotrophic lateral sclerosis.

Amyotrophic lateral sclerosis (ALS) is a severe motor neuron disease, with most sporadic cases lacking clear genetic causes. Abnormal pre-mRNA splicing is a fundamental mechanism in neurodegenerative diseases. For example, TAR DNA-binding protein 43 (TDP-43) loss-of-function (LOF) causes widespread RNA mis-splicing events in ALS. Additionally, splicing mutations are major contributors to neurological disorders. However, the role of intronic variants driving RNA mis-splicing in ALS remains poorly understood. To address this, we developed Spliformer to predict RNA splicing. Spliformer is a transformer-based deep learning model trained and tested on splicing events from the GENCODE database, as well as RNA-seq data from blood and central nervous system tissues. We benchmarked Spliformer against SpliceAI and Pangolin using testing datasets and paired whole-genome sequencing (WGS) with RNA-seq data. We further developed the Spliformer-motif model to identify splicing regulatory motifs. We analyzed Clinvar dataset to identify the link of splicing variants with disease pathogenicity. Additionally, we analyzed WGS data of ALS patients and controls to identify common intronic splicing variants linked to ALS risk or disease phenotypes. We also profiled rare intronic splicing variants in ALS patients to identify known or novel ALS-associated genes. Minigene assays were employed to validate candidate splicing variants. Finally, we measured spine density in neurons with a specific gene knockdown or those expressing a TDP-43 disease-causing mutant. Spliformer accurately predicts the possibilities of a nucleotide within a pre-mRNA sequence being a splice donor, acceptor, or neither. Spliformer outperformed SpliceAI and Pangolin in both speed and accuracy in tested splicing events and/or paired WGS/RNA-seq data. Spliformer-motif successfully identified canonical and novel splicing regulatory motifs. In Clinvar dataset, splicing variants are highly related to disease pathogenicity. Genome-wide analyses of common intronic splicing variants nominated one variant linked to ALS progression. Deep learning analyses of WGS data from 1,370 ALS patients revealed rare splicing variants in reported ALS genes (such as PTPRN2 and CFAP410, validated through minigene assays and RNA-seq), and TDP-43 LOF related RNA mis-splicing genes (such as PTPRD). Further genetic analysis and minigene assays nominated PCP4 and TMEM63A as ALS-associated genes. Functional assays demonstrated that PCP4 is critical for maintaining spine density and can rescue spine loss in neurons expressing a disease-causing TDP-43 mutant. In summary, we developed Spliformer and Spliformer-motif that accurately predict and interpret pre-mRNA splicing. Our findings highlight an intronic genetic mechanism driving RNA mis-splicing in ALS and nominate PCP4 as an ALS-associated gene.

Spatial transcriptome reveals histology-correlated immune signature learnt by deep learning attention mechanism on H&E-stained images for ovarian cancer prognosis

BackgroundThe ability to predict the prognosis of patients with ovarian cancer can greatly improve disease management. However, the knowledge on the mechanism of the prediction is limited. We sought to deconvolute the attention feature learnt by a deep learning convolutional neural networks trained with whole-slide images (WSIs) of hematoxylin-and-eosin (H&E)–stained tumor samples using spatial transcriptomic data.MethodsIn this study, 773 WSIs of H&E-stained tumor sections from 335 patients with treatment naïve high-grade serous ovarian cancer who were included in The Cancer Genome Atlas (TCGA) Pan-Cancer study were used to train, and validate, and to test a ResNet101 CNN model modified with attention mechanism. WSIs from patients in an independent cohort were used to further evaluate the model.ResultsThe prognostic value of the predicted H&E-based survival scores from the trained model on patient survival was evaluated. The attention signals learnt by the model were then examined their correlation with immune signatures using spatial transcriptome. After validating the model with the testing datasets, pathway enrichment analysis showed that the H&E—based survival score significantly correlated with certain immune signatures and this was validated spatially using spatial transcriptome data generated from ovarian cancer FFPE samples by correlating the selected signature and attention signal.ConclusionsIn conclusion, attention mechanism might be useful to identify regions for their specific immune activities. This could guide future pathological study for the useful immunological features that are important in modulating the prognosis of ovarian cancer patients.

Optimization of dried garlic physicochemical properties using a self-organizing map and the development of an artificial intelligence prediction model

The experiments were conducted at different levels of infrared power, airflow, and temperature. The relationships between the input process factors and response factors’ physicochemical properties of dried garlic were optimized by a self-organizing map (SOM), and the model was developed using machine learning. Artificial Neural Network (ANN) with 99% predicting accuracy and Self-Organizing Maps (SOM) with 97% clustering accuracy were used to determine the quality characteristics of garlic. Specifically, five key areas were identified, and valuable insights were offered for optimizing garlic production and improving its overall quality. The (aw) values for the sample ranged from 0.43 to 0.48. The maximum vitamin C content was 0.112 mg/g, followed by an air temperature of 40 °C and 0.7 m/s air velocity under 1500 W/m². The total color change values increased with IR and higher air temperature but declined with higher air velocity. Also, the garlic’s flavor strength, allicin content, water activity, and vitamin C levels decreased as the IR and air temperature increased. The results demonstrated a significant impact of the independent parameters on the response parameters (P < 0.01). Interestingly, the machine learning predictions closely matched the test data sets, providing valuable insights for understanding and controlling the factors affecting garlic drying performances.

Comparative analysis of machine learning approaches for predicting the risk of vaginal laxity

This study develops predictive models for Chinese female patients with VL utilizing machine learning techniques. The aim is to create an effective model that can assist in clinical diagnosis and treatment of vaginal relaxation, thereby enhancing women’s pelvic floor health. In total, 1184 women with VL have been randomly selected and categorized into groups using the finger measurement method. Among them, there are 383 cases of mild VL, 405 cases of moderate VL, and 396 cases of severe VL. Concurrently, 396 healthy women without VL who underwent routine health examinations have been chosen at random and assigned to the non-VL group. Based on 1580 cases, we have established LightGBM, Random Forest, XGBoost, and AdaBoost models based on training dataset using 5-fold cross-validation and GridSearch, and analyzed the performance of the models on the hold-out test dataset. The confusion matrix, precision, recall, F1 score, overall accuracy, and ROC curve of the models on the hold-out test dataset are compared. The overall accuracy of LightGBM model, RF model, XGBoost model, and AdaBoost model are 0.8987, 0.8987, 0.8987, and 0.8457, respectively. The average AUC of LightGBM model is 0.976, the one of RF model is 0.9763, the one of XGBoost model is 0.9775, and the one of AdaBoost model is 0.928. The XGBoost model has the more comprehensive and reasonable performance among the four prediction models, which can accurately distinguish between healthy, mild VL, as well as moderate VL and severe VL, which can assist doctors in diagnosing persons’ conditions more accurately, devising personalized treatment plans, avoiding unnecessary surgeries, reducing persons’ psychological stress, improving patient compliance and treatment outcomes, thus enhancing overall treatment results.

Deep learning for kidney trauma detection: CT image algorithm performance and external validation: experimental study.

Detecting kidney trauma on CT scans can be challenging and is sometimes overlooked. While deep learning (DL) has shown promise in medical imaging, its application to kidney injuries remains underexplored. This study aims to develop and validate a DL algorithm for detecting kidney trauma, using institutional trauma data and the Radiological Society of North America (RSNA) dataset for external validation. We developed RenoTrNet, a DL model trained on institutional data. We evaluated the model's performance through external validation on randomly selected cases from the RSNA dataset. Performance metrics included the accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Heatmap visualizations were used to aid interpretability. In the internal testing dataset, the model achieved an accuracy of 0.88 (95% CI: 0.82-0.92), with a sensitivity of 0.75 (95% CI: 0.62-0.85) and a specificity of 0.95 (95% CI: 0.89-0.98). PPV and NPV were 0.89 (95% CI: 0.76-0.95) and 0.88 (95% CI: 0.81-0.93), respectively. In external RSNA validation, the algorithm c demonstrated robust performance with an accuracy of 0.93 (0.91-0.95), a sensitivity of 0.73 (0.60-0.83), a specificity of 0.94 (0.93-0.96), a PPV of 0.45 (0.35-0.56), and an NPV of 0.98 (0.97-0.99). The RenoTrNet DL algorithm demonstrated high accuracy in detecting kidney trauma on CT scans, both in internal and external validation. By optimizing image segmentation and computational efficiency, this model has potential for clinical deployment, potentially aiding in trauma diagnosis in real-world clinical scenarios.

Hybrid recurrent neural network using custom activation function for COVID-19 short term time series prediction

Introduction: The nonlinear behaviour of activation functions is vital in Artificial Neural Networks (ANNs) for exploring the complex relationship between the input and output features. However, these are probably going to encounter vanishing gradient problems due to small gradients that lead to training instability, expensive exponent operations, and slow convergence. Objectives: The primary objective of this study is to develop Taylor expansion of the second order to realize the hyperbolic tangent and sigmoid functions. In particular, long short term memory network make extensive use of these functions as well as gating mechanism to control the flow of information and gradients. Both the custom functions can reduce the vanishing gradient issues in recurrent neural networks. Methods: Taylor expansion hyperbolic tangent and sigmoid activation functions based parallel heterogeneous Long Short Term Memory Network integrated with Bayesian hyperparameter Optimization is being proposed for coronavirus multi step time series prediction. a Min-Max Normalization is applied, which produces scaled data in the range (0, 1). The normalized dataset is partitioned into training and testing datasets, with 80% and 20%, respectively. Furthermore, both train and test datasets are prepared as input and target series using a window size of 5-7.The further proposed model is tuned with key hyperparameters such as the number of neurons, learning rate, dropout, and type of optimizer. The remaining model parameters are epochs, batch size, and loss, which are 200, 32, and mean square error, respectively. Results: The proposed model efficacy is evaluated on coronavirus daily cumulative cases, cumulative deaths, daily new cases, and total recovery cases in India. The Analysis reveals that the current model achieves remarkable performance in terms of Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of determination (R2 Score) when compared to existing models. Conclusions: The study reveals that the proposed framework with the Taylor approximation activation function produces more consistency in prediction than the default activation functions, including Tanh and sigmoid. In spite of that, gradients of Taylor Tanh and sigmoid activation function traits indicate a decline in the possibility of vanishing issue.