Med Chem. 2025 Jun 3. doi: 10.2174/0115734064370188250527043536. Online ahead of print.

ABSTRACT

BACKGROUND: The advent of Severe Acute Respiratory Syndrome coronavirus 2 (SARS-CoV-2), the etiological agent of the Coronavirus Disease 2019 (COVID-19) pandemic, has impacted physical and mental health worldwide. The lack of effective antiviral drugs necessitates a robust therapeutic approach to develop anti-SARS-CoV-2 drugs. Various investigations have recognized ACE2 as the primary receptor of SARS-CoV-2, and this amalgamation of ACE2 with the spike protein of the coronavirus is paramount for viral entry into the host cells and inducing infection. Consequently, restricting the virus's accessibility to ACE2 offers an alternative therapeutic approach to averting this illness.

OBJECTIVE: The study aimed to identify potent inhibitors with enhanced affinity for the ACE2 protein and validate their stability and efficacy against established inhibitors via molecular docking, machine learning, and MD simulations.

METHODOLOGY: 202 ACE2 inhibitors (PDB ID and 6LZG), comprising repurposed antiviral compounds and specific ACE2 inhibitors, were selected for molecular docking. The two most effective compounds obtained from docking were further analyzed using machine learning to identify potential compounds with enhanced ACE2-binding affinity. To refine the dataset, molecular decoys were generated through the Database of Useful Decoys: Enhanced (DUD-E) server, and Singular Value Decomposition (SVD) was applied for data preprocessing. The Tree-based Pipeline Optimization Tool (TPOT) was then utilized to optimize the machine learning pipeline. The most promising ML-predicted compounds were re-evaluated through docking and subjected to Molecular Dynamics (MD) simulations to evaluate their structural stability and interactions with ACE2. Finally, these compounds were evaluated against the top two pre-established inhibitors using various computational tools.

RESULTS: The two best pre-established inhibitors were identified as Birinapant and Elbasvir, while the best machine-learning-predicted compounds were PubChem ID: 23658468 and PubChem ID: 117637105. Pharmacophore studies were conducted on the most effective machine-learning-predicted compounds, followed by a comparative ADME/T analysis between the best ML-screened and pre-established inhibitors. The results indicated that the top ML compound (PubChem ID: 23658468) demonstrated favorable BBB permeability and a high HIA index, highlighting its potential for therapeutic applications. The ML-screened ligand demonstrated structural stability with an RMSD (0.24 nm) and greater global stability (Rg: 2.08 nm) than Birinapant. Hydrogen bonding interactions further validated their strong binding affinity. MM/PBSA analysis confirmed the ML-screened compound's stronger binding affinity, with a binding free energy of - 132.90 kcal/mol, indicating enhanced stability in complex formation.

CONCLUSION: The results emphasize the efficacy of integrating molecular docking, machine learning, and molecular dynamics simulations in facilitating the rapid identification of novel inhibitors. PubChem ID: 23658468 demonstrates robust binding affinity to ACE2 and favorable pharmacokinetic properties, establishing it as a promising candidate for further investigation.

PMID:40464176 | DOI:10.2174/0115734064370188250527043536

Exp Physiol. 2025 Jun 4. doi: 10.1113/EP092313. Online ahead of print.

ABSTRACT

Proprioception is essential for perception and action. Like any other sense, proprioception is also subject to illusions. In this study, we model classic proprioceptive illusions in which tendon vibrations lead to biases in estimating the state of the body. We investigate these illusions with task-driven models that have been trained to infer the state of the body from distributed sensory muscle spindle inputs (primary and secondary afferents). Recent work has shown that such models exhibit representations similar to the neural code along the ascending proprioceptive pathway. Importantly, we did not train the models on illusion experiments and simulated muscle-tendon vibrations by considering their effect on primary afferents. Our results demonstrate that task-driven models are indeed susceptible to proprioceptive illusions, with the magnitude of the illusion depending on the vibration frequency. This work illustrates that primary afferents alone are sufficient to account for these classic illusions and provides a foundation for future theory-driven experiments.

PMID:40464159 | DOI:10.1113/EP092313

Front Plant Sci. 2025 May 20;16:1524630. doi: 10.3389/fpls.2025.1524630. eCollection 2025.

ABSTRACT

The agricultural sector faces persistent threats from plant diseases and pests, with Tuta absoluta posing a severe risk to tomato farming by causing up to 100% crop loss. Timely pest detection is essential for effective intervention, yet traditional methods remain labor-intensive and inefficient. Recent advancements in deep learning offer promising solutions, with YOLOv8 emerging as a leading real-time detection model due to its speed and accuracy, outperforming previous models in on-field deployment. This study focuses on the early detection of Tuta absoluta-induced tomato leaf diseases in Sub-Saharan Africa. The first major contribution is the annotation of a dataset (TomatoEbola), which consists of 326 images and 784 annotations collected from three different farms and is now publicly available. The second key contribution is the proposal of a transfer learning-based approach to evaluate YOLOv8's performance in detecting Tuta absoluta. Experimental results highlight the model's effectiveness, with a mean average precision of up to 0.737, outperforming other state-of-the-art methods that achieve less than 0.69, demonstrating its capability for real-world deployment. These findings suggest that AI-driven solutions like YOLOv8 could play a pivotal role in reducing agricultural losses and enhancing food security.

PMID:40464016 | PMC:PMC12130032 | DOI:10.3389/fpls.2025.1524630

Front Plant Sci. 2025 May 20;16:1586268. doi: 10.3389/fpls.2025.1586268. eCollection 2025.

ABSTRACT

INTRODUCTION: Salvia miltiorrhiza Bunge (Danshen) is a traditional medicinal plant widely used in the treatment of cardiovascular and inflammatory diseases. Although various omics resources have been published, there remains a lack of an integrated platform to unify genomic, transcriptomic, proteomic, and metabolomic data.

METHODS: To address this gap, we constructed the S. miltiorrhiza Multi-omics Database (SmilODB, http://www.isage.top:56789/), which systematically integrates publicly available genome assemblies, transcriptome datasets, metabolic pathway annotations, and protein structural predictions. Protein structures were predicted using the RoseTTAFold algorithm, and all data were visualized using interactive heat maps, line charts, and histograms.

RESULTS: SmilODB includes: (i) two genome assemblies of S. miltiorrhiza, (ii) 48 tissue-specific transcriptome datasets from root, leaf, and other vegetative tissues, (iii) annotated biosynthetic pathways for bioactive compounds such as tanshinones and salvianolic acids, and (iv) 2,967 high-confidence protein models. The database also integrates bioinformatics tools such as genome browsers, BLAST, and gene heatmap generators.

DISCUSSION: SmilODB provides an accessible and comprehensive platform to explore multi-omics data related to S. miltiorrhiza. It serves as a valuable resource for both basic and applied research, facilitating advances in the understanding of this medicinal plant's molecular mechanisms and therapeutic potential.

PMID:40464010 | PMC:PMC12129996 | DOI:10.3389/fpls.2025.1586268

Front Physiol. 2025 May 20;16:1549380. doi: 10.3389/fphys.2025.1549380. eCollection 2025.

ABSTRACT

INTRODUCTION: Premature Ventricular Contractions (PVCs) can be warning signs for serious cardiac conditions, and early detection is essential for preventing complications. The use of deep learning models in electrocardiogram (ECG) analysis has aided more accurate and efficient PVC identification. These models automatically extract and analyze complex signal features, providing valuable clinical decision-making support. Here, we conducted a study focused on the practical applications of is technology.

METHODS: We first used the MIT-BIH arrhythmia database and a sparse low-rank algorithm to denoise ECG signals. We then transformed the one-dimensional time-series signals into two-dimensional images using Markov Transition Fields (MTFs), considering state transition probabilities and spatial location information to comprehensively capture signal features. Finally, we used the BiFormer classification model, which employs a Bi-level Routing Attention (BRA) mechanism to construct region-level affinity graphs, to retain only the regions highly relevant to our query. This approach filtered out redundant information, and optimized both computational efficiency and memory usage.

RESULTS: Our algorithm achieved a detection accuracy of 99.45%, outperforming other commonly-used PVC detection algorithms.

DISCUSSION: By integrating MTF and BiFormer, we effectively detected PVCs, facilitating an increased convergence between medicine and deep learning technology. We hope our model can help contribute to more accurate computational support for PVC diagnosis and treatment.

PMID:40463999 | PMC:PMC12129755 | DOI:10.3389/fphys.2025.1549380

Food Sci Nutr. 2025 Jun 2;13(6):e70377. doi: 10.1002/fsn3.70377. eCollection 2025 Jun.

ABSTRACT

The vine plant holds significant importance beyond grape farming due to its diverse products. Various grape-derived products, such as wine and molasses, highlight the vine plant's role as a valuable agricultural resource. Additionally, traditional cuisines around the world widely utilize grape leaves, contributing to their substantial economic value. However, diseases affecting grape leaves not only harm the plant and its yield but also render the leaves unsuitable for culinary use, leading to considerable economic losses for producers. Detecting diseases on grape leaves is a challenging and time-consuming task when performed manually. Thus, developing a deep learning-based model to automate the classification of grape leaf diseases is of critical importance. This study aims to classify the most common grape leaf diseases grape-scab (grape leaf blister mite) and downy mildew (grapevine downy mildew) alongside healthy leaves using deep learning techniques. Initially, we conducted a basic classification using pre-trained deep learning models. Subsequently, the Convolutional Block Attention Module (CBAM) and Squeeze-and-Excitation Networks (SE) were integrated into the most successful pre-trained classification model to enhance classification performance. As a result, the classification accuracy improved from 92.73% to 96.36%.

PMID:40463992 | PMC:PMC12129821 | DOI:10.1002/fsn3.70377

Front Oncol. 2025 May 20;15:1585891. doi: 10.3389/fonc.2025.1585891. eCollection 2025.

ABSTRACT

BACKGROUND AND OBJECTIVE: Accurate diagnosis of brain tumors significantly impacts patient prognosis and treatment planning. Traditional diagnostic methods primarily rely on clinicians' subjective interpretation of medical images, which is heavily dependent on physician experience and limited by time consumption, fatigue, and inconsistent diagnoses. Recently, deep learning technologies, particularly Convolutional Neural Networks (CNN), have achieved breakthrough advances in medical image analysis, offering a new paradigm for automated precise diagnosis. However, existing research largely focuses on single-task modeling, lacking comprehensive solutions that integrate tumor segmentation with classification diagnosis. This study aims to develop a multi-task deep learning model for precise brain tumor segmentation and type classification.

METHODS: The study included 485 pathologically confirmed cases, comprising T1-enhanced MRI sequence images of high-grade gliomas, metastatic tumors, and meningiomas. The dataset was proportionally divided into training (378 cases), testing (109 cases), and external validation (51 cases) sets. We designed and implemented BrainTumNet, a deep learning-based multi-task framework featuring an improved encoder-decoder architecture, adaptive masked Transformer, and multi-scale feature fusion strategy to simultaneously perform tumor region segmentation and pathological type classification. Five-fold cross-validation was employed for result verification.

RESULTS: In the test set evaluation, BrainTumNet achieved an Intersection over Union (IoU) of 0.921, Hausdorff Distance (HD) of 12.13, and Dice Similarity Coefficient (DSC) of 0.91 for tumor segmentation. For tumor classification, it attained a classification accuracy of 93.4% with an Area Under the ROC Curve (AUC) of 0.96. Performance remained stable on the external validation set, confirming the model's generalization capability.

CONCLUSION: The proposed BrainTumNet model achieves high-precision diagnosis of brain tumor segmentation and classification through a multi-task learning strategy. Experimental results demonstrate the model's strong potential for clinical application, providing objective and reliable auxiliary information for preoperative assessment and treatment decision-making in brain tumor cases.

PMID:40463867 | PMC:PMC12129765 | DOI:10.3389/fonc.2025.1585891

Front Dent Med. 2025 May 20;6:1534406. doi: 10.3389/fdmed.2025.1534406. eCollection 2025.

ABSTRACT

BACKGROUND: Third molar extraction, a common dental procedure, often involves complications, such as alveolar nerve injury. Accurate preoperative assessment of the extraction difficulty and nerve injury risk is crucial for better surgical planning and patient outcomes. Recent advancements in deep learning (DL) have shown the potential to enhance the predictive accuracy using panoramic radiographic (PR) images. This systematic review evaluated the accuracy and reliability of DL models for predicting third molar extraction difficulty and inferior alveolar nerve (IAN) injury risk.

METHODS: A systematic search was conducted across PubMed, Scopus, Web of Science, and Embase until September 2024, focusing on studies assessing DL models for predicting extraction complexity and IAN injury using PR images. The inclusion criteria required studies to report predictive performance metrics. Study selection, data extraction, and quality assessment were independently performed by two authors using the PRISMA and QUADAS-2 guidelines.

RESULTS: Six studies involving 12,419 PR images met the inclusion criteria. DL models demonstrated high accuracy in predicting extraction difficulty (up to 96%) and IAN injury (up to 92.9%), with notable sensitivity (up to 97.5%) for specific classifications, such as horizontal impactions. Geographically, three studies originated in South Korea and one each from Turkey and Thailand, limiting generalizability. Despite high accuracy, demographic data were sparsely reported, with only two studies providing patient sex distribution.

CONCLUSION: DL models show promise in improving the preoperative assessment of third molar extraction. However, further validation in diverse populations and integration with clinical workflows are necessary to establish its real-world utility, as limitations such as limited generalizability, potential selection bias and lack of long-term follow up remain challenges.

PMID:40463825 | PMC:PMC12129997 | DOI:10.3389/fdmed.2025.1534406

Front Artif Intell. 2025 May 20;8:1517986. doi: 10.3389/frai.2025.1517986. eCollection 2025.

ABSTRACT

Artificial Intelligence (AI) is revolutionizing climate modeling by enhancing predictive accuracy, computational efficiency, and multi-source data integration, playing a crucial role in sustainable urban planning. This Mini Review examines recent advancements in machine learning (ML) and deep learning (DL) techniques that improve climate risk assessment, resource optimization, and infrastructure resilience. Despite these innovations, significant challenges persist, including data quality inconsistencies, model interpretability limitations, ethical concerns, and the scalability of AI models across diverse urban contexts. To bridge these gaps, this review highlights key research directions, emphasizing the development of interpretable AI models, robust data governance frameworks, and scalable AI-driven solutions that help climate adaptation. By addressing these challenges, AI-based climate modeling can provide actionable insights for policymakers, urban planners, and researchers fostering climate-resilient and sustainable urban environments.

PMID:40463823 | PMC:PMC12129934 | DOI:10.3389/frai.2025.1517986

Proc SPIE Int Soc Opt Eng. 2025 Feb;13406:134060B. doi: 10.1117/12.3047410. Epub 2025 Apr 11.

ABSTRACT

Placenta volume measured from 3D ultrasound (3DUS) images is an important tool for tracking the growth trajectory and is associated with pregnancy outcomes. Manual segmentation is the gold standard, but it is time-consuming and subjective. Although fully automated deep learning algorithms perform well, they do not always yield high-quality results for each case. Interactive segmentation models could address this issue. However, there is limited work on interactive segmentation models for the placenta. Despite their segmentation accuracy, these methods may not be feasible for clinical use as they require relatively large computational power which may be especially prohibitive in low-resource environments, or on mobile devices. In this paper, we propose a lightweight interactive segmentation model aiming for clinical use to interactively segment the placenta from 3DUS images in real-time. The proposed model adopts the segmentation from our fully automated model for initialization and is designed in a human-in-the-loop manner to achieve iterative improvements. The Dice score and normalized surface Dice are used as evaluation metrics. The results show that our model can achieve superior performance in segmentation compared to state-of-the-art models while using significantly fewer parameters. Additionally, the proposed model is much faster for inference and robust to poor initial masks. The code is available at https://github.com/MedICL-VU/PRISM-placenta.

PMID:40463735 | PMC:PMC12128914 | DOI:10.1117/12.3047410

Commun Med (Lond). 2025 Jun 3;5(1):212. doi: 10.1038/s43856-025-00935-2.

ABSTRACT

BACKGROUND: During radiation therapy, the natural movement of organs can lead to underdosing the cancer and overdosing the healthy tissue, compromising treatment efficacy. Real-time image-guided adaptive radiation therapy can track the tumour and account for the motion. Typically, fiducial markers are implanted as a surrogate for the tumour position due to the low radiographic contrast of soft tissues in kilovoltage (kV) images. A segmentation approach that does not require markers would eliminate the costs, delays, and risks associated with marker implantation.

METHODS: We trained patient-specific conditional Generative Adversarial Networks for prostate segmentation in kV images. The networks were trained using synthetic kV images generated from each patient's own imaging and planning data, which are available prior to the commencement of treatment. We validated the networks on two treatment fractions from 30 patients using multi-centre data from two clinical trials.

RESULTS: Here, we present a large-scale proof-of-principle study of x-ray-based markerless prostate segmentation for globally available cancer therapy systems. Our results demonstrate the feasibility of a deep learning approach using kV images to track prostate motion across the entire treatment arc for 30 patients with prostate cancer. The mean absolute deviation is 1.4 and 1.6 mm in the anterior-posterior/lateral and superior-inferior directions, respectively.

CONCLUSIONS: Markerless segmentation via deep learning may enable real-time image guidance on conventional cancer therapy systems without requiring implanted markers or additional hardware, thereby expanding access to real-time adaptive radiation therapy.

PMID:40461695 | DOI:10.1038/s43856-025-00935-2

Sci Rep. 2025 Jun 3;15(1):19462. doi: 10.1038/s41598-025-04720-y.

ABSTRACT

Recently, with the growth of electronic cars and the advancement of modern vehicles using portable equipment and embedded systems, several in-vehicle networks like the CAN (Controller Area Network) encountered novel risks of security. Because the portal of CAN does not have systems of security, like encryption and authentication in order to contend with cyber-attacks, the necessity for a system of intrusion detection for identifying attacks on the portal of CAN is really essential. In this study, Triple-attention Mechanism (TAN) has been used to recognize different kinds of security intrusions in portals of CAN. The purpose of TAN here is to identify intrusion within 3 steps. Within the initial phase, the major features have been extracted, and TAN functions as a descriptor of feature. Then, the discriminating categorizer classifies the current features. Eventually, with the help of adversarial learning, intrusion has been recognized. The current work utilizes a novel Greylag Goose Optimization algorithm for optimal selection of the network hyperparameters. For checking the effectiveness of the suggested method, an open-source dataset was applied, which recorded the traffic of CAN using a real vehicle throughout injection attacks of message. The results show that this method outperforms certain machine learning algorithms in error rate and false negative for DoS and drive gear and RPM spoofing attack with accuracy of 96.3%, recall of 96.1%, F1-Score of 96.2%, specificity of 97.2%, accuracy of 96.3%, AUC-ROC of 0.97, and MCC of 0.92 for DoS attacks. Therefore, the phase attack is minimized.

PMID:40461686 | DOI:10.1038/s41598-025-04720-y

Sci Rep. 2025 Jun 3;15(1):19495. doi: 10.1038/s41598-025-03932-6.

ABSTRACT

Drug-Target Interaction (DTI) prediction is a vital task in drug discovery, yet it faces significant challenges such as data imbalance and the complexity of biochemical representations. This study makes several contributions to address these issues, introducing a novel hybrid framework that combines advanced machine learning (ML) and deep learning (DL) techniques. The framework leverages comprehensive feature engineering, utilizing MACCS keys to extract structural drug features and amino acid/dipeptide compositions to represent target biomolecular properties. This dual feature extraction method enables a deeper understanding of chemical and biological interactions, enhancing predictive accuracy. To address data imbalance, Generative Adversarial Networks (GANs) are employed to create synthetic data for the minority class, effectively reducing false negatives and improving the sensitivity of the predictive model. The Random Forest Classifier (RFC) is utilized to make precise DTI predictions, optimized for handling high-dimensional data. The proposed framework's scalability and robustness were validated across diverse datasets, including BindingDB-Kd, BindingDB-Ki, and BindingDB-IC50. For the BindingDB-Kd dataset, the GAN+RFC model achieved remarkable performance metrics: accuracy of 97.46%, precision of 97.49%, sensitivity of 97.46%, specificity of 98.82%, F1-score of 97.46%, and ROC-AUC of 99.42%. Similarly, for the BindingDB-Ki dataset, the model attained an accuracy of 91.69%, precision of 91.74%, sensitivity of 91.69%, specificity of 93.40%, F1-score of 91.69%, and ROC-AUC of 97.32%. On the BindingDB-IC50 dataset, the model achieved an accuracy of 95.40%, precision of 95.41%, sensitivity of 95.40%, specificity of 96.42%, F1-score of 95.39%, and ROC-AUC of 98.97%. These results demonstrate the efficacy of the GAN-based approach in capturing complex patterns, significantly improving DTI prediction outcomes. In conclusion, the proposed GAN-based hybrid framework sets a new benchmark in computational drug discovery by addressing critical challenges in DTI prediction. Its robust performance, scalability, and generalizability contribute substantially to therapeutic development and pharmaceutical research.

PMID:40461636 | DOI:10.1038/s41598-025-03932-6

Sci Rep. 2025 Jun 3;15(1):19458. doi: 10.1038/s41598-025-05116-8.

ABSTRACT

Electrical impedance spectroscopy (EIS) is a powerful tool used to investigate the properties of materials and biological tissues. This study presents one of the first applications of EIS for the detection and classification of oral potentially malignant disorders (OPMDs) and oral cancer. We aimed to apply EIS in conjunction with deep learning to assist the clinical diagnosis of OPMD and oral cancer as a non-invasive diagnostic technology. Currently, the diagnosis of OPMD and oral cancer relies on clinical examination and histopathological analysis of invasive scalpel tissue biopsies, which is stressful for patients, time-consuming for clinicians and subject to histopathological interobserver variation in diagnosis, although recent advances in artificial intelligence may circumvent discrepancy. Here we developed a novel deep learning convolutional neural network (CNN)-based method to automatically differentiate normal, OPMD and malignant oral tissues using EIS measurements. EIS readings were initially taken from untreated or glacial acetic acid-treated porcine oral mucosa and analyzed via CNN to determine if this method could discriminate between normal and damaged oral epithelium. CNN models achieved area under the curve (AUC) values of 0.92 ± 0.03, with specificity 0.95 and sensitivity 0.84, showing good discrimination. EIS data from ventral tongue and floor-of-the-mouth were collected from 51 healthy humans and 11 patients with OPMD and oral cancer. When a binary classification (low or high risk of malignancy) was applied, the best CNN model achieved an AUC 0.91 ± 0.1, with accuracy 0.91 ± 0.05, specificity 0.97 and sensitivity 0.74. These results demonstrate the considerable potential of EIS in combination with CNN models as an adjunctive non-invasive diagnostic tool for OPMD and oral cancer.

PMID:40461631 | DOI:10.1038/s41598-025-05116-8

Sci Rep. 2025 Jun 3;15(1):19419. doi: 10.1038/s41598-025-98653-1.

ABSTRACT

EEG is widely applied in emotion recognition, brain disease detection, and other fields due to its high temporal resolution and non-invasiveness. However, artifact removal remains a crucial issue in EEG signal processing. Recently, with the rapid development of deep learning, there has been a significant transformation in the methods of EEG artifact removal. Nonetheless, existing research still exhibits some limitations: (1) insufficient capability to remove unknown artifacts; (2) inability to adapt to tasks where artifact removal needs to be applied to the overall input of multi-channel EEG data. Therefore, this study proposes CLEnet by integrating dual-scale CNN (Convolutional Neural Networks) and LSTM (Long Short-Term Memory), and incorporating an improved EMA-1D (One-Dimensional Efficient Multi-Scale Attention Mechanism). CLEnet can extract the morphological features and temporal features of EEG, thereby separating EEG from artifacts. We conducted experiments on three datasets, and the results showed that CLEnet performed best. Specifically, in the task of removing artifacts from multi-channel EEG data containing unknown artifacts, CLEnet shows improvements of 2.45% and 2.65% in SNR(signal-to-noise ratio) and CC(average correlation coefficient). Moreover, RRMSEt(relative root mean square error in the temporal domain) and RRMSEf (relative root mean square error in the frequency domain) decrease by 6.94% and 3.30%.

PMID:40461599 | DOI:10.1038/s41598-025-98653-1

Commun Eng. 2025 Jun 3;4(1):100. doi: 10.1038/s44172-025-00431-4.

ABSTRACT

A large number of in-service reinforced concrete structures are now entering the mid-to-late stages of their service life. Efficient detection of damage characteristics and accurate prediction of material performance degradation have become essential for ensuring the safety of these structures. Traditional damage detection methods, which primarily rely on manual inspections and sensor monitoring, are inefficient and lack accuracy. Similarly, performance prediction models for reinforced concrete materials, which are often based on limited experimental data and polynomial fitting, oversimplify the influencing factors. In contrast, partial differential equation models that account for degradation mechanisms are computationally intensive and difficult to solve. Recent advancements in deep learning and machine learning, as part of artificial intelligence, have introduced innovative approaches for both damage detection and material performance prediction in reinforced concrete structures. This paper provides a comprehensive overview of machine learning and deep learning theories and models, and reviews the current research on their application to the durability of reinforced concrete structures, focusing on two main areas: intelligent damage detection and predictive modeling of material durability. Finally, the article discusses future trends and offers insights into the intelligent innovation of concrete structure durability.

PMID:40461590 | DOI:10.1038/s44172-025-00431-4

Sci Rep. 2025 Jun 3;15(1):19425. doi: 10.1038/s41598-025-04552-w.

ABSTRACT

The automatic detection of pain through the analysis of facial expressions is indeed one of the most critical challenges in the healthcare system. One of the significant challenges in automatic pain detection from facial expressions is the variability in how individuals express pain and other emotions through their facial deformations. This research aims to solve this issue by presenting ErAS-Net, an Enhanced Residual Attention-Based Subject-Specific Network that employs various attention mechanisms. Through transfer learning and multiple attention mechanisms, the proposed deep learning model is designed to mimic human perception of facial expressions, thereby enhancing its pain recognition ability and capturing the unique features of each individual's facial expressions based on their specific patterns. The UNBC-McMaster Shoulder Pain dataset is used to demonstrate the effectiveness of the proposed deep learning algorithm, which achieves impressive values of 98.77% accuracy for binary classification and 94.21% for four-level pain intensity classification using tenfold cross-validation. Additionally, the model attained 89.83% accuracy for binary classification with the Leave-One-Subject-Out (LOSO) validation method. To further evaluate generalizability, a cross-dataset experiment was conducted using the BioVid Heat Pain Database, where ErAS-Net achieved 78.14% accuracy for binary pain detection on unseen data without fine-tuning. The fact that this finding supports the attention mechanism and human perception is why the proposed model proves to be a powerful and reliable tool for automatic pain detection.

PMID:40461564 | DOI:10.1038/s41598-025-04552-w

Sci Rep. 2025 Jun 3;15(1):19379. doi: 10.1038/s41598-025-03979-5.

ABSTRACT

Systematic literature review (SLR) is an important tool for Health Economics and Outcomes Research (HEOR) evidence synthesis. SLRs involve the identification and selection of pertinent publications and extraction of relevant data elements from full-text articles, which can be a manually intensive procedure. Previously we developed machine learning models to automatically identify relevant publications based on pre-specified inclusion and exclusion criteria. This study investigates the feasibility of applying Natural Language Processing (NLP) approaches to automatically extract data elements from the relevant scientific literature. First, 239 full-text articles were collected and annotated for 12 important variables including study cohort, lab technique, and disease type, for proper SLR summary of Human papillomavirus (HPV) Prevalence, Pneumococcal Epidemiology, and Pneumococcal Economic Burden. The three resulting annotated corpora are shared publicly at [ https://github.com/Merck/NLP-SLR-corpora ], to provide training data and a benchmark baseline for the NLP community to further research this challenging task. We then compared three classic Named Entity Recognition (NER) algorithms, namely Conditional Random Fields (CRF), Long Short-Term Memory (LSTM), and the Bidirectional Encoder Representations from Transformers (BERT) models, to assess performance on the data element extraction task. The annotation corpora contain 4,498, 579, and 252 annotated entity mentions for HPV Prevalence, Pneumococcal Epidemiology, and Pneumococcal Economic Burden tasks respectively. Deep learning algorithms achieved superior performance in recognizing the targeted SLR data elements, compared to conventional machine learning algorithms. LSTM models have achieved 0.890, 0.646 and 0.615 micro-averaged F1 scores for three tasks respectively. CRF models could not provide comparable performance on most of the elements of interest. Although BERT-based models are known to generally achieve superior performance on many NLP tasks, we did not observe improvement in our three tasks. Deep learning algorithms have achieved superior performance compared with machine learning models on multiple SLR data element extraction tasks. LSTM model, in particular, is more preferable for deployment in supporting HEOR SLR data element extraction, due to its better performance, generalizability, and scalability as it's cost-effective in our SLR benchmark datasets.

PMID:40461545 | DOI:10.1038/s41598-025-03979-5

Sci Rep. 2025 Jun 3;15(1):19407. doi: 10.1038/s41598-025-01791-9.

ABSTRACT

Accurate and timely diagnosis of leukemia, a cancer characterized by an excessive number of abnormal white blood cells (WBCs), is crucial for effective treatment. Manual examination of blood smear images for leukemia diagnosis is often laborious and costly. Computer-aided classification of WBCs has the potential to assist hematologists in improving diagnostic accuracy. However, the subtle visual differences among the five types of immature neutrophils pose a significant challenge, even for experienced professionals. The study proposes a method called densely connected co-occurrence network (DCONN). The method first detects white blood cells in blood smear images using Yolact. Then, the images are pre-processed to minimize the correlation between image channels by transforming RGB color space to LAB color space. Finally, DCONN extracts spatial texture information using a co-occurrence matrix to improve classification accuracy. DCONN achieved 93.46% accuracy in classifying five types of immature neutrophils: myeloblast, promyelocyte, myelocyte, metamyelocyte, and band cells. The results indicate that using a combination of densely connected convolutional layers and a co-occurrence layer improves classification accuracy while using fewer trainable parameters than other deep learning methods such as ResNet and Inception. Additionally, the model is less demanding in terms of training hardware than attentional mechanism-based models that also have local feature representation. DCONN achieves advanced performance based on small-scale models without requiring much training time. The proposed method can be extended to other pathological image analyses in the future.

PMID:40461529 | DOI:10.1038/s41598-025-01791-9

Sci Rep. 2025 Jun 3;15(1):19369. doi: 10.1038/s41598-025-02015-w.

ABSTRACT

Cancer is among the most dangerous diseases contributing to rising global mortality rates. Lung cancer, particularly adenocarcinoma, is one of the deadliest forms and severely impacts human life. Early diagnosis and appropriate treatment significantly increase patient survival rates. Computed Tomography (CT) is a preferred imaging modality for detecting lung cancer, as it offers detailed visualization of tumor structure and growth. With the advancement of deep learning, the automated identification of lung cancer from CT images has become increasingly effective. This study proposes a novel lung cancer detection framework using a Flower Pollination Algorithm (FPA)-based weighted ensemble of three high-performing pretrained Convolutional Neural Networks (CNNs): VGG16, ResNet101V2, and InceptionV3. Unlike traditional ensemble approaches that assign static or equal weights, the FPA adaptively optimizes the contribution of each CNN based on validation performance. This dynamic weighting significantly enhances diagnostic accuracy. The proposed FPA-based ensemble achieved an impressive accuracy of 98.2%, precision of 98.4%, recall of 98.6%, and an F1 score of 0.985 on the test dataset. In comparison, the best individual CNN (VGG16) achieved 94.6% accuracy, highlighting the superiority of the ensemble approach. These results confirm the model's effectiveness in accurate and reliable cancer diagnosis. The proposed study demonstrates the potential of deep learning and neural networks to transform cancer diagnosis, helping early detection and improving treatment outcomes.

PMID:40461493 | DOI:10.1038/s41598-025-02015-w