Sci Rep. 2025 May 10;15(1):16275. doi: 10.1038/s41598-025-00607-0.

ABSTRACT

Climate change, which causes long-term temperature and weather changes, threatens natural ecosystems and cities. It has worldwide economic consequences. Climate change trends up to 2050 are predicted using the hybrid model that consists of Convolutional Neural Network-Gated Recurrent Unit-Long Short-Term Memory (CNN-GRU-LSTM), a unique deep learning architecture. With a focus on Al-Qassim Region, Saudi Arabia, the model assesses temperature, air temperature dew point, visibility distance, and atmospheric sea-level pressure. We used Synthetic Minority Over-sampling Technique for Regression with Gaussian Noise (SMOGN) to reduce dataset imbalance. The CNN-GRU-LSTM model was compared to 5 classic regression models: DTR, RFR, ETR, BRR, and K-Nearest Neighbors. Five main measures were used to evaluate model performance: MSE, MAE, MedAE, RMSE, and R². After Min-Max normalization, the dataset was split into training (70%), validation (15%), and testing (15%) sets. The paper shows that the CNN-GRU-LSTM model beats standard regression methods in all four climatic scenarios, with R² values of 99.62%, 99.15%, 99.71%, and 99.60%. Deep learning predicts climate change well and can guide environmental policy and urban development decisions.

PMID:40346151 | DOI:10.1038/s41598-025-00607-0

Sci Rep. 2025 May 9;15(1):16273. doi: 10.1038/s41598-025-00996-2.

ABSTRACT

Animal sounds can be recognised automatically by machine learning, and this has an important role to play in biodiversity monitoring. Yet despite increasingly impressive capabilities, bioacoustic species classifiers still exhibit imbalanced performance across species and habitats, especially in complex soundscapes. In this study, we explore the effectiveness of transfer learning in large-scale bird sound classification across various conditions, including single- and multi-label scenarios, and across different model architectures such as CNNs and Transformers. Our experiments demonstrate that both finetuning and knowledge distillation yield strong performance, with cross-distillation proving particularly effective in improving in-domain performance on Xeno-canto data. However, when generalizing to soundscapes, shallow finetuning exhibits superior performance compared to knowledge distillation, highlighting its robustness and constrained nature. Our study further investigates how to use multi-species labels, in cases where these are present but incomplete. We advocate for more comprehensive labeling practices within the animal sound community, including annotating background species and providing temporal details, to enhance the training of robust bird sound classifiers. These findings provide insights into the optimal reuse of pretrained models for advancing automatic bioacoustic recognition.

PMID:40346144 | DOI:10.1038/s41598-025-00996-2

Sci Rep. 2025 May 9;15(1):16239. doi: 10.1038/s41598-025-01270-1.

ABSTRACT

The accurate preoperative staging of laryngeal squamous cell carcinoma (LSCC) provides valuable guidance for clinical decision-making. The objective of this study was to establish a multiparametric MRI model using radiomics and deep learning (DL) to preoperatively distinguish between Stages I-II and III-IV of LSCC. Data from 401 histologically confirmed LSCC patients were collected from two centers (training set: 213; internal test set: 91; external test set: 97). Radiomics features were extracted from the MRI images, and seven radiomics models based on single and combined sequences were developed via random forest (RF). A DL model was constructed via ResNet 18, where DL features were extracted from its final fully connected layer. These features were fused with crucial radiomics features to create a combined model. The performance of the models was assessed using the area under the receiver operating characteristic (ROC) curve (AUC) and compared with the radiologist performances. The predictive capability of the combined model for Progression-Free Survival (PFS) was evaluated via Kaplan-Meier survival analysis and the Harrell's Concordance Index (C-index). In the external test set, the combined model had an AUC of 0.877 (95% CI 0.807-0.946), outperforming the DL model (AUC: 0.811) and the optimal radiomics model (AUC: 0.835). The combined model significantly outperformed both the DL (p = 0.017) and the optimal radiomics models (p = 0.039), and the radiologists (both p < 0.050). Moreover, the combined model demonstrated great prognostic predictive value in patients with LSCC, achieving a C-index of 0.624 for PFS. This combined model enhances preoperative LSCC staging, aiding in making more informed clinical decisions.

PMID:40346120 | DOI:10.1038/s41598-025-01270-1

Eur J Radiol. 2025 Apr 23;188:112132. doi: 10.1016/j.ejrad.2025.112132. Online ahead of print.

ABSTRACT

OBJECTIVE: This study aims to investigate the impact of adaptive statistical iterative reconstruction-Veo (ASIR-V) and deep learning image reconstruction (DLIR) algorithms on the quantification of pericoronary adipose tissue (PCAT) and epicardial adipose tissue (EAT). Furthermore, we propose to explore the feasibility of correcting the effects through fat threshold adjustment.

METHODS: A retrospective analysis was conducted on the imaging data of 134 patients who underwent coronary CT angiography (CCTA) between December 2023 and January 2024. These data were reconstructed into seven datasets using filtered back projection (FBP), ASIR-V at three different intensities (ASIR-V 30%, ASIR-V 50%, ASIR-V 70%), and DLIR at three different intensities (DLIR-L, DLIR-M, DLIR-H). Repeated-measures ANOVA was used to compare differences in fat, PCAT and EAT attenuation values among the reconstruction algorithms, and Bland-Altman plots were used to analyze the agreement between ASIR-V or DLIR and FBP algorithms in PCAT attenuation values.

RESULTS: Compared to FBP, ASIR-V 30 %, ASIR-V 50 %, ASIR-V 70 %, DLIR-L, DLIR-M, and DLIR-H significantly increased fat attenuation values (-103.91 ± 12.99 HU, -102.53 ± 12.68 HU, -101.14 ± 12.78 HU, -101.81 ± 12.41 HU, -100.87 ± 12.25 HU, -99.08 ± 12.00 HU vs. -105.95 ± 13.01 HU, all p < 0.001). When the fat threshold was set at -190 to -30 HU, ASIR-V and DLIR algorithms significantly increased PCAT and EAT attenuation values compared to FBP algorithm (all p < 0.05), with these values increasing as the reconstruction intensity level increased. After correction with a fat threshold of -200 to -35 HU for ASIR-V 30 %, -200 to -40 HU for ASIR-V 50 % and DLIR-L, and -200 to -45 HU for ASIR-V 70 %, DLIR-M, and DLIR-H, the mean differences in PCAT attenuation values between ASIR-V or DLIR and FBP algorithms decreased (-0.03 to 1.68 HU vs. 2.35 to 8.69 HU), and no significant difference was found in PCAT attenuation values between FBP and ASIR-V 30 %, ASIR-V 50 %, ASIR-V 70 %, DLIR-L, and DLIR-M (all p > 0.05).

CONCLUSION: Compared to the FBP algorithm, ASIR-V and DLIR algorithms increase PCAT and EAT attenuation values. Adjusting the fat threshold can mitigate the impact of ASIR-V and DLIR algorithms on PCAT attenuation values.

PMID:40344712 | DOI:10.1016/j.ejrad.2025.112132

PLoS Comput Biol. 2025 May 9;21(5):e1013074. doi: 10.1371/journal.pcbi.1013074. Online ahead of print.

ABSTRACT

With a burgeoning number of artificial intelligence (AI) applications in various fields, biomolecular science has also given a big welcome to advanced AI techniques in recent years. In this broad field, scoring a protein-ligand binding structure to output the binding strength is a crucial problem that heavily relates to computational drug discovery. Aiming at this problem, we have proposed an efficient scoring framework using deep learning techniques. This framework describes a binding structure by a high-resolution atomic graph, places a focus on the inter-molecular interactions and learns the graph in a rational way. For a protein-ligand binding complex, the generated atomic graph reserves key information of the atoms (as graph nodes), and focuses on inter-molecular interactions (as graph edges) that can be identified by introducing multiple distance ranges to the atom pairs within the binding area. To provide more confidence in the predicted binding strengths, we have interpreted the deep learning model from the model level and in a post-hoc analysis. The proposed learning framework has been demonstrated to have competitive performances in scoring and screening tasks, which will prospectively promote the development of related fields further.

PMID:40344574 | DOI:10.1371/journal.pcbi.1013074

JCO Clin Cancer Inform. 2025 May;9:e2400284. doi: 10.1200/CCI-24-00284. Epub 2025 May 9.

ABSTRACT

PURPOSE: Artificial intelligence (AI) tools could improve clinical decision making or exacerbate inequities because of bias. African American (AA) men reportedly have a worse prognosis for prostate cancer (PCa) and are underrepresented in the development genomic biomarkers. We assess the generalizability of tools developed using a multimodal AI (MMAI) deep learning system using digital histopathology and clinical data from NRG/Radiation Therapy Oncology Group PCa trials across racial subgroups.

METHODS: In total, 5,708 patients from five randomized phase III trials were included. Two MMAI algorithms were evaluated: (1) the distant metastasis (DM) MMAI model optimized to predict risk of DM, and (2) the PCa-specific mortality (PCSM) MMAI model optimized to focus on prediction death in the presence of DM (DDM). The prognostic performance of the MMAI algorithms was evaluated in AA and non-AA subgroups using time to DM (primary end point) and time to DDM (secondary end point). Exploratory end points included time to biochemical failure and overall survival with Fine-Gray or Cox proportional hazards models. Cumulative incidence estimates were computed for time-to-event end points and compared using Gray's test.

RESULTS: There were 948 (16.6%) AA patients, 4,731 non-AA patients (82.9%), and 29 (0.5%) patients with unknown or missing race status. The DM-MMAI algorithm showed a strong prognostic signal for DM in the AA (subdistribution hazard ratio [sHR], 1.2 [95% CI, 1.0 to 1.3]; P = .007) and non-AA subgroups (sHR, 1.4 [95% CI, 1.3 to 1.5]; P < .001). Similarly, the PCSM-MMAI score showed a strong prognostic signal for DDM in both AA (sHR, 1.3 [95% CI, 1.1 to 1.5]; P = .001) and non-AA subgroups (sHR, 1.5 [95% CI, 1.4 to 1.6]; P < .001), with similar distributions of risk.

CONCLUSION: Using cooperative group data sets with a racially diverse population, the MMAI algorithm performed well across racial subgroups without evidence of algorithmic bias.

PMID:40344545 | DOI:10.1200/CCI-24-00284

Adv Sci (Weinh). 2025 May 8:e2501269. doi: 10.1002/advs.202501269. Online ahead of print.

ABSTRACT

The CRISPR-Cas12a system has gained significant attention as a rapid nucleic acid diagnostic tool due to its crRNA-guided trans-cleavage activity. Accurately predicting the activity of different targets is significant to facilitate the crRNA availability but remains challenging. In this study, a novel approach is presented that combines molecular dynamics simulations and neural network modeling to predict the trans-cleavage activity. Unlike conventional tools that rely solely on the base sequences, our method integrated sequence features and molecular interaction features of DNA in the CRISPR-Cas12a system, significantly improving prediction accuracy. Through feature importance analysis, key sequence features that influence Cas12a trans-cleavage activity are identified. Additionally, a crRNA-DNA library with over 23 456 feature sequences from representative viruses and bacteria is established, and validated the high predictive accuracy of the model (Pearson's r = 0.9328) by screening crRNAs from reference targets. This study offers new insights into the molecular interactions of Cas12a/crRNA-DNA and provides a reliable framework for optimizing crRNA design, facilitating the application of the CRISPR-Cas12a in rapid nucleic acid diagnostics.

PMID:40344384 | DOI:10.1002/advs.202501269

J Proteome Res. 2025 May 9. doi: 10.1021/acs.jproteome.4c00973. Online ahead of print.

ABSTRACT

Identifying detectable peptides, known as flyers, is key in mass spectrometry-based proteomics. Peptide detectability is strongly related to peptide sequences and their resulting physicochemical properties. Moreover, the high variability in MS data challenges the development of a generic model for detectability prediction, underlining the need for customizable tools. We present Pfly, a deep learning model developed to predict peptide detectability based solely on peptide sequence. Pfly is a versatile and reliable state-of-the-art tool, offering high performance, accessibility, and easy customizability for end-users. This adaptability allows researchers to tailor Pfly to specific experimental conditions, improving accuracy and expanding applicability across various research fields. Pfly is an encoder-decoder with an attention mechanism, classifying peptides as flyers or non-flyers, and providing both binary and categorical probabilities for four distinct classes defined in this study. The model was initially trained on a synthetic peptide library and subsequently fine-tuned with a biological dataset to mitigate bias toward synthesizability, improving predictive capacity and outperforming state-of-the-art predictors in benchmark comparisons across different human and cross-species datasets. The study further investigates the influence of protein abundance and rescoring, illustrating the negative impact on peptide identification due to misclassification. Pfly has been integrated into the DLOmix framework and is accessible on GitHub at https://github.com/wilhelm-lab/dlomix.

PMID:40344201 | DOI:10.1021/acs.jproteome.4c00973

Med Phys. 2025 May 8. doi: 10.1002/mp.17871. Online ahead of print.

ABSTRACT

BACKGROUND: Cardiac perfusion PET is commonly used to assess ischemia and cardiovascular risk, which enables quantitative measurements of myocardial blood flow (MBF) through kinetic modeling. However, the estimation of kinetic parameters is challenging due to the noisy nature of short dynamic frames and limited sample data points.

PURPOSE: This work aimed to investigate the errors in MBF estimation in PET through a simulation study and to evaluate different parameter estimation approaches, including a deep learning (DL) method.

MATERIALS AND METHODS: Simulated studies were generated using digital phantoms based on cardiac segmentations from 55 clinical CT images. We employed the irreversible 2-tissue compartmental model and simulated dynamic 13N-ammonia PET scans under both rest and stress conditions (220 cases each). The simulations covered a rest K1 range of 0.6 to 1.2 and a stress K1 range of 1.2 to 3.6 (unit: mL/min/g) in the myocardium. A transformer-based DL model was trained on the simulated dataset to predict parametric images (PIMs) from noisy PET image frames and was validated using 5-fold cross-validation. We compared the DL method with the voxel-wise nonlinear least squares (NLS) fitting applied to the dynamic images, using either Gaussian filter (GF) smoothing (GF-NLS) or a dynamic nonlocal means (DNLM) algorithm for denoising (DNLM-NLS). Two patients with coronary CT angiography (CTA) and fractional flow reserve (FFR) were enrolled to test the feasibility of applying DL models on clinical PET data.

RESULTS: The DL method showed clearer image structures with reduced noise compared to the traditional NLS-based methods. In terms of mean absolute relative error (MARE), as the rest K1 values increased from 0.6 to 1.2 mL/min/g, the overall bias in myocardium K1 estimates decreased from approximately 58% to 45% for the NLS-based methods while the DL method showed a reduction in MARE from 42% to 18%. For stress data, as the stress K1 decreased from 3.6 to 1.2 mL/min/g, the MARE increased from 30% to 70% for the GF-NLS method. In contrast, both the DNLM-NLS (average: 42%) and the DL methods (average: 20%) demonstrated significantly smaller MARE changes as stress K1 varied. Regarding the regional mean bias (±standard deviation), the GF-NLS method had a bias of 6.30% (±8.35%) of rest K1, compared to 1.10% (±8.21%) for DNLM-NLS and 6.28% (±14.05%) for the DL method. For the stress K1, the GF-NLS showed a mean bias of 10.72% (±9.34%) compared to 1.69% (±8.82%) for DNLM-NLS and -10.55% (±9.81%) for the DL method.

SIGNIFICANCE: This study showed that an increase in the tracer uptake rate (K1) corresponded to improved accuracy and precision in MBF quantification, whereas lower tracer uptake resulted in higher noise in dynamic PET and poorer parameter estimates. Utilizing denoising techniques or DL approaches can mitigate noise-induced bias in PET parametric imaging.

PMID:40344168 | DOI:10.1002/mp.17871

PLoS One. 2025 May 9;20(5):e0322624. doi: 10.1371/journal.pone.0322624. eCollection 2025.

ABSTRACT

Brain tumors pose a significant medical challenge, necessitating early detection and precise classification for effective treatment. This study aims to address this challenge by introducing an automated brain tumor classification system that utilizes deep learning (DL) and Magnetic Resonance Imaging (MRI) images. The main purpose of this research is to develop a model that can accurately detect and classify different types of brain tumors, including glioma, meningioma, pituitary tumors, and normal brain scans. A convolutional neural network (CNN) architecture with pretrained VGG16 as the base model is employed, and diverse public datasets are utilized to ensure comprehensive representation. Data augmentation techniques are employed to enhance the training dataset, resulting in a total of 17,136 brain MRI images across the four classes. The accuracy of this model was 99.24%, a higher accuracy than other similar works, demonstrating its potential clinical utility. This higher accuracy was achieved mainly due to the utilization of a large and diverse dataset, the improvement of network configuration, the application of a fine-tuning strategy to adjust pretrained weights, and the implementation of data augmentation techniques in enhancing classification performance for brain tumor detection. In addition, a web application was developed by leveraging HTML and Dash components to enhance usability, allowing for easy image upload and tumor prediction. By harnessing artificial intelligence (AI), the developed system addresses the need to reduce human error and enhance diagnostic accuracy. The proposed approach provides an efficient and reliable solution for brain tumor classification, facilitating early diagnosis and enabling timely medical interventions. This work signifies a potential advancement in brain tumor classification, promising improved patient care and outcomes.

PMID:40344143 | DOI:10.1371/journal.pone.0322624

PLoS One. 2025 May 9;20(5):e0322365. doi: 10.1371/journal.pone.0322365. eCollection 2025.

ABSTRACT

PURPOSE: Radiomics allows for the quantification of medical images and facilitates precision medicine. Many radiomic features derived from computed tomography (CT) are sensitive to variations across scanners, reconstruction settings, and acquisition protocols. In this phantom study, eight different CT reconstruction parameters were varied to explore image- and feature-level harmonization approaches to improve tissue classification.

METHODS: Varying reconstructions of an anthropomorphic radiopaque phantom containing three lesion categories (metastasis, hemangioma, and benign cyst) and normal liver tissue were used for evaluating two harmonization methods and their combination: (i) generative adversarial networks (GANs) at the image level; (ii) ComBat at the feature level, and (iii) a combination of (i) and (ii). A total of 93 texture and intensity features were extracted from each tissue class before and after image-level harmonization and were also harmonized at the feature level. Reproducibility and stability were assessed via the Concordance Correlation Coefficient (CCC) and pairwise comparisons using paired stability tests. The ability of features to discriminate between tissue classes was assessed by measuring the area under the receiver operating characteristic curve. The global reproducibility and discriminative power were assessed by averaging over the entire dataset and across all tissue types.

RESULTS: ComBat improved reproducibility by 31.58% and stability by 5.24%, while GAN increased reproducibility by 8% it reduced stability by 4.33%. Classification analysis revealed that ComBat increased average AUC by 15.19%, whereas GAN decreased AUC by 2.56%.

CONCLUSION: While GAN qualitatively enhances image harmonization, ComBat provides superior statistical improvements in feature stability and classification performance, highlighting the importance of robust feature-level harmonization in radiomics.

PMID:40344028 | DOI:10.1371/journal.pone.0322365

J Xray Sci Technol. 2025 May;33(3):501-515. doi: 10.1177/08953996241304987. Epub 2025 Feb 20.

ABSTRACT

BackgroundLung disease is the crucial disease that affects the breathing conditions and even causes death. There are various approaches for the lung disease classification; still the inefficiency in accurate detection, computational complexity and over-fitting issues limits the performance of the model. To overcome the challenges, a deep learning model is proposed in this research. Initially, the input is acquired and is pre-processed using three various techniques like data augmentation, filtering and image re-sizing. Then, the threshold based segmentation is employed for obtaining the required region.ObjectiveFrom the segmented image, various categories of lung diseases like COVID, lung Opacity, Pneumonia and normal are identified using the proposed Optimal Cross Stage Partial Bidirectional Long short term memory (OCBiNet).MethodsThe proposed OCBiNet is designed using Bidirectional Long short-term memory (BiNet) with Cross Stage Partial connection in its hidden state. Besides, the adjustable parameters are modified using the proposed Improved Mother Optimization (ImMO) algorithm.ResultsThe ImMO algorithm is designed by integrating the Logistic Chaotic Mapping within the conventional Mother Optimization algorithm for enhancing the convergence rate in obtaining the global best solution.ConclusionsThe proposed OCBiNet is evaluated based on Accuracy, Recall, Precision, and F-Score and acquired the values of 99.11%, 98.98%, 99.18%, and 99.08% respectively.

PMID:40343884 | DOI:10.1177/08953996241304987

IEEE Trans Biomed Eng. 2025 May 9;PP. doi: 10.1109/TBME.2025.3568282. Online ahead of print.

ABSTRACT

Deep learning has significantly enhanced the research on the emerging issue of Electroencephalogram (EEG)-based visual classification and reconstruction, which has gained a growth of attention and concern recently. To promote the research progress, at this critical moment, a review work on the deep learning methodology for the issue becomes necessary and important. However, such a work seems absent in the literature. This paper provides the first review on EEG-based visual classification and reconstruction, whose contents can be categorized into the following four main parts: 1) comprehensively summarizing and systematically analyzing the representative deep learning methods from both feature encoding and decoding perspectives; 2) introducing the available benchmark datasets, describing the experimental paradigms, and displaying the method performances; 3) proposing the methodological essences and neuroscientific insights as well as the dynamic closed-loop interaction and promotion between them, which are potentially beneficial for technological innovations and academic progress; 4) discussing the potential challenges of current research and the prospective opportunities in future trends. We expect that this work can shed light on the technological directions and also enlighten the academic breakthroughs for the issue in the not-so-far future.

PMID:40343828 | DOI:10.1109/TBME.2025.3568282

Sci Rep. 2025 May 8;15(1):16001. doi: 10.1038/s41598-025-98852-w.

ABSTRACT

Accurate diagnosis of orthopedic injuries, especially pelvic and hip fractures, is vital in trauma management. While pelvic radiographs (PXRs) are widely used, misdiagnosis is common. This study proposes an automated system that uses convolutional neural networks (CNNs) to detect potential fracture areas and predict fracture conditions, aiming to outperform traditional object detection-based systems. We developed two deep learning models for hip fracture detection and prediction, trained on PXRs from three hospitals. The first model utilized automated hip area detection, cropping, and classification of the resulting patches. The images were preprocessed using the Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm. The YOLOv5 architecture was employed for the object detection model, while three different pre-trained deep neural network (DNN) architectures were used for classification, applying transfer learning. Their performance was evaluated on a test dataset, and compared with 35 clinicians. YOLOv5 achieved a 92.66% accuracy on regular images and 88.89% on CLAHE-enhanced images. The classifier models, MobileNetV2, Xception, and InceptionResNetV2, achieved accuracies between 94.66% and 97.67%. In contrast, the clinicians demonstrated a mean accuracy of 84.53% and longer prediction durations. The DNN models showed significantly better accuracy and speed compared to human evaluators (p < 0.0005, p < 0.01). These DNN models highlight promising utility in trauma diagnosis due to their high accuracy and speed. Integrating such systems into clinical practices may enhance the diagnostic efficiency of PXRs.

PMID:40341645 | DOI:10.1038/s41598-025-98852-w

Sci Rep. 2025 May 8;15(1):16090. doi: 10.1038/s41598-025-00017-2.

ABSTRACT

In architectural and engineering design, minimizing weight, deflection, and strain energy requires navigating complex, non-linear interactions among competing objectives, making the optimization of concrete thin shell constructions particularly challenging. Traditional multi-objective optimization (MOO) methods frequently encounter difficulties in effectively exploring design spaces, which often necessitate substantial computational resources and result in suboptimal solutions. This paper presents a novel approach for enhancing topology and thickness optimization. Graph-constrained conditional Generative Adversarial Networks (GANs) and the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) are used in the study. The hybrid approach addresses fundamental limitations in current optimization techniques by combining the generative capabilities of deep learning with the refinement processes of evolutionary algorithms. NSGA-II enhances the algorithm by employing evolutionary processes to generate various structural designs that adhere to topological constraints. Specialized graph-constrained GANs accomplish this. The implementation of the system in a concrete thin shell structure at the Shenzhen Qianhai Smart Community resulted in significant performance improvements: a 33.3% reduction in total weight, a 50% decrease in maximum deflection, and a 20% reduction in strain energy compared to baseline models. A comparative comparison of traditional NSGA-II techniques shows substantial benefits, including a 50% enhancement in convergence speed and notable advancements in solution diversity and quality. We confirmed structural integrity through extensive finite element analysis and practical prototyping, achieving performance variations under 3.5%. This work illustrates the potential of sophisticated machine learning and evolutionary algorithms to produce innovative, high-performance architectural solutions, thereby providing a new methodology for structural optimization.

PMID:40341580 | DOI:10.1038/s41598-025-00017-2

Sci Rep. 2025 May 8;15(1):16087. doi: 10.1038/s41598-025-00374-y.

ABSTRACT

Indoor activity monitoring methods assurance the well-being and security of disabled and aging individuals living in their homes. These models utilize numerous technologies and sensors to monitor day-to-day work like movement, medication adherence, and sleep patterns, and provide valued perceptions of the user's everyday life and entire health. Internet of Things (IoT) based health systems have an important part in medical assistance and help in enhancing data processing and its prediction. Communicating data or reports requires more time and energy, in addition to causing energy problems and greater latency. Currently, numerous studies focus on human activity recognition (HAR) using deep learning (DL) and machine learning (ML) methods, but more effort is needed to enhance HAR models for disabled individuals. Therefore, this article presents a Smart Indoor Monitoring for Disabled People Using an Ensemble of Deep Learning Models in an Internet of Things Environment (SIMDP-EDLIoT) technique. The SIMDP-EDLIoT model is designed to monitor and detect various conditions and activities within indoor spaces for disabled people. Initially, the SIMDP-EDLIoT approach uses linear scaling normalization (LSN) to ensure that the input data is scaled appropriately. Besides, the Improved Osprey Optimization Algorithm (IOOA)-based feature selection is employed to classify the most relevant features, enhancing the efficiency of the system by reducing dimensionality. For monitoring indoor activities, an ensemble of three DL techniques such as bidirectional long short-term memory (BiLSTM), gated recurrent unit (GRU), and conditional variational autoencoder (CVAE) are employed. Experimental study is performed to underscore the importance of the SIMDP-EDLIoT method under the HAR dataset. The comparative analysis of the SIMDP-EDLIoT method demonstrated a superior performance with an accuracy of 98.85%, precision of 97.71%, sensitivity of 97.70%, specificity of 99.24%, and F-measure of 97.70%, outperforming existing approaches across all metrics.

PMID:40341573 | DOI:10.1038/s41598-025-00374-y

J Microsc. 2025 May 8. doi: 10.1111/jmi.13423. Online ahead of print.

ABSTRACT

Atomic force microscopy (AFM) plays a pivotal role in cell biology research. It enables scientists to observe the morphology of cell surfaces at the nanoscale, providing essential data for understanding cellular functions, including cell-cell interactions and responses to the microenvironment. Nevertheless, AFM-captured cell images frequently suffer from artefacts, which significantly hinder detailed analyses of cell structures. In this study, we developed a cross-module resolution enhancement method for post-processing AFM cell images. The method leverages the AFM topological deep learning neural network. We propose an enhanced spatial fusion structure and an optimised back-projection mechanism within an adversarial-based super-resolution network to detect weak signals and complex textures unique to AFM cell images. Furthermore, we designed a crossover-based frequency division module, capitalising on the distinct frequency characteristics of AFM images. This module effectively separates and enhances features pertinent to cell structure. In this paper, experiments were conducted using AFM images of various cells, and the results demonstrated the model's superiority. It substantially enhances image quality compared to existing methods. Specifically, the peak signal-to-noise ratio (PSNR) of the reconstructed image increased by 1.65 decibels, from 28.121 to 29.771, the structural similarity (SSIM) increased by 0.041, from 0.746 to 0.787, the Learned Perceptual Image Patch Similarity (LPIPS) decreased by 0.205, from 0.437 to 0.232, the Fréchet Inception Distance (FID) decreased by 6.996, from 55.442 to 48.446 and the Natural Image Quality Evaluator (NIQE) decreased by 0.847, from 4.296 to 3.449. Lay abstract: This study proposes a deep learning-based cross-module method for super-resolving AFM cell images, integrating frequency division and adaptive fusion modules. It boosts PSNR by 1.65 dB and SSIM by 0.041, accurately recovering cellular microstructures, thus significantly aiding cell biology research and biomedicine applications.

PMID:40341533 | DOI:10.1111/jmi.13423

Nat Commun. 2025 May 8;16(1):4276. doi: 10.1038/s41467-025-59523-6.

ABSTRACT

Human-machine voice interaction based on speech recognition offers an intuitive, efficient, and user-friendly interface, attracting wide attention in applications such as health monitoring, post-disaster rescue, and intelligent control. However, conventional microphone-based systems remain challenging for complex human-machine collaboration in noisy environments. Herein, an anti-noise triboelectric acoustic sensor (Anti-noise TEAS) based on flexible nanopillar structures is developed and integrated with a convolutional neural network-based deep learning model (Anti-noise TEAS-DLM). This highly synergistic system enables robust acoustic signal recognition for human-machine collaboration in complex, noisy scenarios. The Anti-noise TEAS directly captures acoustic fundamental frequency signals from laryngeal mixed-mode vibrations through contact sensing, while effectively suppressing environmental noise by optimizing device-structure buffering. The acoustic signals are subsequently processed and semantically decoded by the DLM, ensuring high-fidelity interpretation. Evaluated in both simulated virtual and real-life noisy environments, the Anti-noise TEAS-DLM demonstrates near-perfect noise immunity and reliably transmits various voice commands to guide robotic systems in executing complex post-disaster rescue tasks with high precision. The combined anti-noise robustness and execution accuracy endow this DLM-enhanced Anti-noise TEAS as a highly promising platform for next-generation human-machine collaborative systems operating in challenging noisy environments.

PMID:40341503 | DOI:10.1038/s41467-025-59523-6

Int J Comput Assist Radiol Surg. 2025 May 8. doi: 10.1007/s11548-025-03382-5. Online ahead of print.

ABSTRACT

PURPOSE: Identifying and quantifying coronary artery calcification (CAC) is crucial for preoperative planning, as it helps to estimate both the complexity of the 2D coronary angiography (2DCA) procedure and the risk of developing intraoperative complications. Despite the relevance, the actual practice relies upon visual inspection of the 2DCA image frames by clinicians. This procedure is prone to inaccuracies due to the poor contrast and small size of the CAC; moreover, it is dependent on the physician's experience. To address this issue, we developed a workflow to assist clinicians in identifying CAC within 2DCA using data from 44 image acquisitions across 14 patients.

METHODS: Our workflow consists of three stages. In the first stage, a classification backbone based on ResNet-18 is applied to guide the CAC identification by extracting relevant features from 2DCA frames. In the second stage, a U-Net decoder architecture, mirroring the encoding structure of the ResNet-18, is employed to identify the regions of interest (ROI) of the CAC. Eventually, a post-processing step refines the results to obtain the final ROI. The workflow was evaluated using a leave-out cross-validation.

RESULTS: The proposed method outperformed the comparative methods by achieving an F1-score for the classification step of 0.87 (0.77 - 0.94) (median ± quartiles), while for the CAC identification step the intersection over minimum (IoM) was 0.64 (0.46 - 0.86) (median ± quartiles).

CONCLUSION: This is the first attempt to propose a clinical decision support system to assist the identification of CAC within 2DCA. The proposed workflow holds the potential to improve both the accuracy and efficiency of CAC quantification, with promising clinical applications. As future work, the concurrent use of multiple auxiliary tasks could be explored to further improve the segmentation performance.

PMID:40341465 | DOI:10.1007/s11548-025-03382-5

Sci Rep. 2025 May 8;15(1):16042. doi: 10.1038/s41598-025-01120-0.

ABSTRACT

With the increasing urbanization in China, monitoring and predicting the deformation of deep excavations has become increasingly critical. Concurrently, as neural network models find application and development in deep excavation displacement prediction, traditional models face challenges such as insufficient accuracy and weak generalization capabilities, failing to meet the high-precision warning demands of practical engineering. Therefore, research into hybrid models is necessary. This study proposes a combined neural network model integrating a Convolutional Neural Network, Long Short-Term Memory network, and Self-Attention Mechanism (CNN-LSTM-SAM), which utilizes time-series monitoring data as input. The CNN-LSTM-SAM model merges the data feature extraction capabilities of CNN, the long-term memory function of LSTM, and the information weighting capacity of the self-attention mechanism, synthesizing the advantages of various deep excavation displacement prediction models to enhance prediction accuracy and provide more effective support for construction practice. Furthermore, given the limited application of the CNN-LSTM-SAM model in deep excavation displacement analysis, this research contributes to addressing gaps in this field. Applied to an internally braced deep excavation project in the Donggang Business District of Dalian, displacement data acquired through Distributed Fiber Optic Sensing (DFOS) technology were used as training data. The CNN-LSTM-SAM model was employed to predict the horizontal displacement at the pile top. The resulting deformation predictions were compared and analyzed against those from Back Propagation (BP) neural network, Long Short-Term Memory (LSTM) network, and a combined Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model. Results indicate that at monitoring point S5, the coefficient of determination (R2) for the CNN-LSTM-SAM model's predictions increased by 12.42%, 10.85%, and 5.63% compared to the BP, LSTM, and CNN-LSTM models, respectively, demonstrating higher accuracy than the other three models. Similar patterns were observed when training and predicting using data from other monitoring points, proving the applicability and robustness of the CNN-LSTM-SAM model. The findings of this study offer valuable references for the design and construction of similar deep excavation projects.

PMID:40341437 | DOI:10.1038/s41598-025-01120-0