bioRxiv [Preprint]. 2025 Feb 5:2025.01.31.635598. doi: 10.1101/2025.01.31.635598.

ABSTRACT

Deep learning excels at segmenting objects within noisy cryo-electron tomograms, but the approach is typically bottlenecked by access to ground truth training data. To address this issue we have developed CryoTomoSim (CTS), an open-source software package that builds coarse-grained models of macromolecular complexes embedded in vitreous ice and then simulates transmitted electron tilt series for tomographic reconstruction. Using CTS outputs, we demonstrate the effects of key microscope parameters (dose, defocus, and pixel size) on deep learning-based segmentation, and show that including both molecular crowding and diversity within synthetic datasets is key to training cellular segmentation networks from purely synthetic inputs. While very effective as initial models, the accuracy of these networks is currently limited, and real cellular data is necessary to train the most accurate and generalizable U-Nets. Using a co-training approach, we first segment over 100 tomograms from neuronal growth cones to quantify their cytoskeletal distributions and then we build a generalized cellular cryo-ET segmentation network called NeuralSeg that can segment a subset of cellular features in tomograms from all domains of life.

PMID:39975172 | PMC:PMC11838407 | DOI:10.1101/2025.01.31.635598

bioRxiv [Preprint]. 2025 Jan 30:2025.01.29.635570. doi: 10.1101/2025.01.29.635570.

ABSTRACT

Spatial correlation of functional connectivity profiles across matching anatomical locations in individuals is often calculated to delineate individual differences in functional networks. Likewise, spatial correlation is assessed across average functional connectivity profiles of groups to evaluate the maturity of functional networks during development. Despite its widespread use, spatial correlation is limited to comparing two samples at a time. In this study, we employed a variational autoencoder to embed functional connectivity profiles from various anatomical locations, individuals, and group averages for simultaneous comparison. We demonstrate that our variational autoencoder, with pre-trained weights, can project new functional connectivity profiles from the vertex space to a latent space with as few as two dimensions, yet still retain meaningful global and local structures in the data. Functional connectivity profiles from various functional networks occupy distinct compartments of the latent space. Moreover, the variability of functional connectivity profiles from the same anatomical location is readily captured in the latent space. We believe that this approach could be useful for visualization and exploratory analyses in precision functional mapping.

PMID:39975052 | PMC:PMC11838398 | DOI:10.1101/2025.01.29.635570

bioRxiv [Preprint]. 2025 Feb 5:2024.08.29.610300. doi: 10.1101/2024.08.29.610300.

ABSTRACT

The proto-oncogene Ras which governs diverse intracellular pathways has four major isoforms (KRAS4A, KRAS4B, HRAS, and NRAS) with substantial sequence homology and similar in vitro biochemistry. There is considerable interest in investigating the roles of these independently as their association with different cancers vary, but there are few Ras isoform-specific binding reagents as the only significant sequence differences are in their disordered and highly charged C-termini which have been difficult to elicit antibodies against. To overcome this limitation, we use deep learning-based methods to de novo design Ras isoform-specific binders (RIBs) for all major Ras isoforms that specifically target the Ras C-terminus. The RIBs bind to their target Ras isoforms both in vitro and in cells with remarkable specificity, disrupting their membrane localization and inhibiting Ras activity, and should contribute to dissecting the distinct roles of Ras isoforms in biology and disease.

PMID:39975043 | PMC:PMC11838417 | DOI:10.1101/2024.08.29.610300

bioRxiv [Preprint]. 2025 Feb 5:2025.02.04.636539. doi: 10.1101/2025.02.04.636539.

ABSTRACT

The development of therapies and vaccines targeting integral membrane proteins has been complicated by their extensive hydrophobic surfaces, which can make production and structural characterization difficult. Here we describe a general deep learning-based design approach for solubilizing native membrane proteins while preserving their sequence, fold, and function using genetically encoded de novo protein WRAPs ( W ater-soluble R Fdiffused A mphipathic P roteins) that surround the lipid-interacting hydrophobic surfaces, rendering them stable and water-soluble without the need for detergents. We design WRAPs for both beta-barrel outer membrane and helical multi-pass transmembrane proteins, and show that the solubilized proteins retain the binding and enzymatic functions of the native targets with enhanced stability. Syphilis vaccine development has been hindered by difficulties in characterizing and producing the outer membrane protein antigens; we generated soluble versions of four Treponema pallidum outer membrane beta barrels which are potential syphilis vaccine antigens. A 4.0 Å cryo-EM map of WRAPed TP0698 is closely consistent with the design model. WRAPs should be broadly useful for facilitating biochemical and structural characterization of integral membrane proteins, enabling therapeutic discovery by screening against purified soluble targets, and generating antigenically intact immunogens for vaccine development.

PMID:39975033 | PMC:PMC11838538 | DOI:10.1101/2025.02.04.636539

Sci Rep. 2025 Feb 19;15(1):6011. doi: 10.1038/s41598-025-89651-4.

ABSTRACT

Traffic congestion, particularly in rapidly expanding urban centers, significantly impacts the timely delivery of emergency medical services (EMS), where every minute can mean the difference between life and death. Traditional traffic signal control systems often lack real-time adaptability to prioritize emergency vehicles, resulting in delays caused by congestion around ambulances. To address this critical issue, this paper presents an AI-driven real-time traffic management system designed to reduce EMS response times. The proposed solution incorporates three core components: Raspberry Pi-based traffic signal prioritization, deep learning-enabled audio-visual ambulance detection, and an advanced intelligent traffic management framework. For audio detection, raw data is transformed into spectrograms using Mel Frequency Cepstral Coefficients (MFCCs) and classified using a Long Short-Term Memory (LSTM) network. Visual data is processed through a ResNet18 convolutional neural network, pre-trained on ImageNet using inductive transfer learning. The outputs from the auditory and visual streams are integrated using empirical risk minimization, enabling accurate ambulance detection through multimodal data fusion. Performance evaluation demonstrates the effectiveness of the proposed system, achieving 98.3% accuracy in audio classification, 98.1% accuracy in visual classification, and 99% accuracy with the fused model. Additional metrics, including precision, recall, F1-score, and a confusion matrix, confirm the model's reliability. This innovative system has the potential to transform urban traffic networks into intelligent, adaptive systems, reducing delays caused by traffic congestion, enhancing emergency medical care response times, and ultimately saving lives. The framework offers a scalable blueprint for future smart city traffic management solutions, meticulously designed to support urban growth and expansion.

PMID:39971977 | DOI:10.1038/s41598-025-89651-4

Sci Rep. 2025 Feb 19;15(1):6035. doi: 10.1038/s41598-025-89348-8.

ABSTRACT

Diet, stress, genetics, and a sedentary lifestyle may all contribute to heart disease rates. Although recent studies propose comprehensive automated diagnostic systems, these systems tend to focus on one aspect, such as feature selection, prioritization, or predictive accuracy. A more complete approach that considers all of these factors can improve the efficiency of a cardiac prediction system. This study uses an appropriate strategy to overcome potential network design problems, design challenges, overfitting, and lack of robustness that can interfere with system performance. The research introduces an ideally designed deep trust network called ID-DTN to improve system performance. The Ruzzo-Tompa method is used to eliminate noncontributory features. The Seagull Optimization Algorithm (SOA) is introduced to optimize the trust depth network to achieve optimal network design. The study scrutinizes the deep trust network (ID-DTN) and the restricted Boltzmann machine (RBM) and sheds light on the system's operation. This proposal can optimize both network architecture and feature selection, which is the main novelty. The proposed method is analyzed using the below-mentioned metrics: Matthew's correlation coefficient, F1 score, accuracy, sensitivity, specificity, and accuracy. ID-DTN performs well compared to other state-of-the-art methods. The validation results confirm that the proposed method improves the prediction accuracy to 97.11% and provides reliable recommendations for patients with cardiovascular disease.

PMID:39971944 | DOI:10.1038/s41598-025-89348-8

J Imaging Inform Med. 2025 Feb 19. doi: 10.1007/s10278-025-01440-7. Online ahead of print.

ABSTRACT

Segmentation and classification of breast tumors are two critical tasks since they provide significant information for computer-aided breast cancer diagnosis. Combining these tasks leverages their intrinsic relevance to enhance performance, but the variability and complexity of tumor characteristics remain challenging. We propose a novel multi-task deep learning network (NMTNet) for the joint segmentation and classification of breast tumors, which is based on a convolutional neural network (CNN) and U-shaped architecture. It mainly comprises a shared encoder, a multi-scale fusion channel refinement (MFCR) module, a segmentation branch, and a classification branch. First, ResNet18 is used as the backbone network in the encoding part to enhance the feature representation capability. Then, the MFCR module is introduced to enrich the feature depth and diversity. Besides, the segmentation branch combines a lesion region enhancement (LRE) module between the encoder and decoder parts, aiming to capture more detailed texture and edge information of irregular tumors to improve segmentation accuracy. The classification branch incorporates a fine-grained classifier that reuses valuable segmentation information to discriminate between benign and malignant tumors. The proposed NMTNet is evaluated on both ultrasound and magnetic resonance imaging datasets. It achieves segmentation dice scores of 90.30% and 91.50%, and Jaccard indices of 84.70% and 88.10% for each dataset, respectively. And the classification accuracy scores are 87.50% and 99.64% for the corresponding datasets, respectively. Experimental results demonstrate the superiority of NMTNet over state-of-the-art methods on breast tumor segmentation and classification tasks.

PMID:39971818 | DOI:10.1007/s10278-025-01440-7

J Imaging Inform Med. 2025 Feb 19. doi: 10.1007/s10278-025-01446-1. Online ahead of print.

ABSTRACT

The rapid advancements of deep learning technology have revolutionized medical imaging diagnosis. However, training these models is often challenged by label imbalance and the scarcity of certain diseases. Most models fail to recognize multiple coexisting diseases, which are common in real-world clinical scenarios. Moreover, most radiological models rely solely on image data, which contrasts with radiologists' comprehensive approach, incorporating both images and other clinical information such as clinical history and laboratory results. In this study, we introduce a Multimodal Chest X-ray Network (MCX-Net) that integrates chest X-ray images and clinical history texts for multi-label disease diagnosis. This integration is achieved by combining a pretrained text encoder, a pretrained image encoder, and a pretrained image-text cross-modal encoder, fine-tuned on the public MIMIC-CXR-JPG dataset, to diagnose 13 diverse lung diseases on chest X-rays. As a result, MCX-Net achieved the highest macro AUROC of 0.816 on the test set, significantly outperforming unimodal baselines such as ViT-base and ResNet152, which scored 0.747 and 0.749, respectively (p < 0.001). This multimodal approach represents a substantial advancement over existing image-based deep-learning diagnostic systems for chest X-rays.

PMID:39971817 | DOI:10.1007/s10278-025-01446-1

BJOG. 2025 Feb 19. doi: 10.1111/1471-0528.18097. Online ahead of print.

ABSTRACT

Cardiotocography (CTG), introduced in the 1960s, was initially expected to prevent hypoxia-related deaths and neurological injuries. However, more than five decades later, evidence supporting the evidence of intrapartum CTG in preventing neonatal and long-term childhood morbidity and mortality remains inconclusive. At the same time, shortcomings in CTG interpretation have been recognised as important contributory factors to rising caesarean section rates and missed opportunities for timely interventions. An important limitation is its high false-positive rate and poor specificity, which undermines reliably identifying foetuses at risk of hypoxia-related injuries. These shortcomings are compounded by the technology's significant intra- and interobserver variability, as well as the subjective and complex nature of fetal heart rate interpretation. However, human factors and other environmental factors are equally significant. Advancements in fetal heart rate monitoring are crucial to support clinicians in improving health outcomes for newborns and their mothers, while at the same time avoiding unnecessary operative deliveries. These limitations highlight the clinical need to enhance neonatal outcomes while minimising unnecessary interventions, such as instrumental deliveries or caesarean sections. We believe that achieving this requires a paradigm shift from subjective interpretation of complex and nonspecific fetal heart rate patterns to evidence-based, quantifiable solutions that integrate hardware, engineering and clinical perspectives. Such transformation necessitates an international, multidisciplinary effort encompassing the entire continuum of pregnancy care and the broader healthcare ecosystem, with emphasis on well-defined, actionable health outcomes. Achieving this will depend on collaborations between researchers, clinicians, medical device manufacturers and other relevant stakeholders. This expert review paper outlines the most relevant and promising directions for research and strategic initiatives to address current challenges in fetal heart rate monitoring. Key themes include advancements in computerised fetal heart rate monitoring, the application of big data and artificial intelligence, innovations in home and remote monitoring and consideration of human factors.

PMID:39971749 | DOI:10.1111/1471-0528.18097

Magn Reson Med Sci. 2025 Feb 20. doi: 10.2463/mrms.mp.2024-0126. Online ahead of print.

ABSTRACT

PURPOSE: Recently, a novel deep learning (DL)-based technique for reconstructing highly undersampled MR data (DL-Speed, DLS) has been developed, which demonstrated superior performance over compressed sensing. This study aimed to achieve high-resolution double inversion recovery (DIR) imaging using DLS (DLS-DIR) and compare its diagnostic performance in the detection of juxtacortical multiple sclerosis (MS) lesions with that of conventional DIR (C-DIR).

METHODS: We retrospectively analyzed MRI data from 25 patients with MS who underwent a comprehensive imaging protocol, including 3D fluid-attenuated inversion recovery (FLAIR), C-DIR, and DLS-DIR. A voxel size of 1.3 × 1.3 × 1.4 mm3 with a scan duration of 3 mins 55s were used for C-DIR, and isotropic 0.7 mm voxels with a scan time of 4 mins 23s were employed for DLS-DIR. Two neuroradiologists assessed the juxtacortical MS lesions during 2 separate reading sessions (one with C-DIR and the other with DLS-DIR). Lesions were categorized as subcortical white matter lesions, intracortical lesions, or mixed lesions involving both subcortical white and gray matter. The lesion counts per region were compared between the imaging techniques using the Wilcoxon signed-rank test.

RESULTS: DLS-DIR detected a significantly higher number of juxtacortical MS lesions compared to C-DIR (Radiologist A: 211 lesions vs. 164 lesions; Radiologist B: 209 lesions vs. 157 lesions, P < 0.05). DLS-DIR also identified more intracortical lesions (Radiologist A: 22 additional lesions, Radiologist B: 34 additional lesions, P < 0.05) and more mixed lesions (Radiologist A: 46 additional lesions, Radiologist B: 42 additional lesions, P < 0.05).

CONCLUSION: The DLS technology enables high-resolution, whole-brain DLS-DIR imaging within a 5 mins acquisition time, which can be seamlessly incorporated into routine clinical workflows. This approach substantially enhances the detection and evaluation of juxtacortical MS lesions.

PMID:39971311 | DOI:10.2463/mrms.mp.2024-0126

JMIR Ment Health. 2025 Feb 19. doi: 10.2196/72803. Online ahead of print.

ABSTRACT

BACKGROUND: Heart rate variability (HRV) is a physiological marker of the cardiac autonomic modulation and related emotional regulation. Electroencephalography (EEG) is reflective of brain cortical activities and related psychopathology. The HRV and EEG have been employed in machine learning- and deep learning-based algorithms either alone or with other wearable device-based features to classify patients with psychiatric disorder (PT) and healthy controls (HC). Little study examined the utility of wearable device-based physiological markers to discern PT with various psychiatric diagnosis versus HC.

OBJECTIVE: This study examined the HRV and prefrontal EEG features most frequently selected in the support vector machine (SVM) having the highest classification accuracy of PT versus HC, contributing to the individual-level initial screening of PT and minimized duration of untreated psychiatric illness.

METHODS: A simultaneous acquisition of 5 minute-length PPG (measured on right ear lobe) and resting-state EEG (with eye-closed; using two left/right forehead-located electrodes) of 182 participants [87 PT (including major depressive disorder (70.1%) and panic disorder (12.6%)) and 95 HC] were performed. The PPG-based HRV features were quantified for both time- and frequency-domains. The time-varying EEG signals were converted into frequency-domain signals of the power spectral density. In the feature selection of the Gaussian radial basis function kernel-based support vector machine (SVM) models, estimators were comprised of top N (1£N£22) highest scored HRV/EEG features based on the one-way ANOVA F-value. Classification performance of SVM model (PT vs. HC) having N estimators was assessed using the Leave-one-out cross-validation (LOOCV; N = 182), to confirm those showing the highest balanced accuracy and area under the receiver operating characteristic curve (AUROC) as final classification model.

RESULTS: The final SVM model having 13 estimators showed balanced accuracy of 0.76 and AUROC of 0.78. Power spectral density of HRV in the high frequency, very low frequency, low frequency (LF) bands, and total power, a product of the mean of the 5-minute standard deviation of all NN intervals (SDNN) and normalized LF power of HRV, power spectral density of frontal EEG in the high alpha and alpha peak frequency comprised the top 13-scored classification features in > 90% of the LOOCV.

CONCLUSIONS: This study showed a possible synergic effect of combining the HRV and prefrontal EEG features in machine learning-based mental health screening. Future studies to predict the treatment response and to propose the preferred treatment regimen based on the baseline physiological markers are required.

CLINICALTRIAL: N/A.

PMID:39971280 | DOI:10.2196/72803

Brain Res. 2025 Feb 17:149507. doi: 10.1016/j.brainres.2025.149507. Online ahead of print.

ABSTRACT

Brain tumors pose a significant threat to human health, require a precise and quick diagnosis for effective treatment. However, achieving high diagnostic accuracy with traditional methods remains challenging due to the complex nature of brain tumors. Recent advances in deep learning have showed potential in automating brain tumor classification using brain MRI images, offering the potential to enhance diagnostic result. This paper present FusionNet, a novel approach that utilizing normal and segmented MRI images to achieve better classification accuracy. Segmented images are generated using a Dual Residual Blocks based pre-trained model. Secondly, the model uses attention based mechanism and ensemble feature selection to prioritize the relevant features for improving the classification performance. Thirdly, proposed model incorporates the feature fusion of both the images (normal and segmented) to increase the selected feature for better classification. The proposed model achieved high accuracy across multiple datasets, with an accuracy of 99.62%, 99.54%, 99.39%, and 99.57% on the Figshare, Kaggle, Sartaj, combined dataset respectively. The proposed model demonstrates notable improvements in performance on both datasets. It achieves higher accuracy, precision, recall, and F1-score compared to existing models on the both datasets. The proposed FusionNet demonstrates significant improvements in brain tumor classification performance. The utility of this study lies in its contribution to the scientific community as a robust, efficient tool that advances brain tumor classification, supporting medical professionals in achieving superior diagnostic outcomes.

PMID:39970997 | DOI:10.1016/j.brainres.2025.149507

BMC Res Notes. 2025 Feb 19;18(1):77. doi: 10.1186/s13104-025-07142-1.

ABSTRACT

The nonlinear Telegraph equation appears in a variety of engineering and science problems. This paper presents a deep learning algorithm termed physics-informed neural networks to resolve a hyperbolic nonlinear telegraph equation with Dirichlet, Neumann, and Periodic boundary conditions. To include physical information about the issue, a multi-objective loss function consisting of the residual of the governing partial differential equation and initial conditions and boundary conditions is defined. Using multiple densely connected neural networks, termed feedforward deep neural networks, the proposed scheme has been trained to minimize the total loss results from the multi-objective loss function. Three computational examples are provided to demonstrate the efficacy and applications of our suggested method. Using a Python software package, we conducted several tests for various model optimizations, activation functions, neural network architectures, and hidden layers to choose the best hyper-parameters representing the problem's physics-informed neural network model with the optimal solution. Furthermore, using graphs and tables, the results of the suggested approach are contrasted with the analytical solution in literature based on various relative error analyses and statistical performance measure analyses. According to the results, the suggested computational method is effective in resolving difficult non-linear physical issues with various boundary conditions.

PMID:39972356 | DOI:10.1186/s13104-025-07142-1

Plant Methods. 2025 Feb 19;21(1):21. doi: 10.1186/s13007-025-01341-4.

ABSTRACT

Accurate detection of missed tassels is crucial for maintaining the purity of hybrid maize seed production. This study introduces the MT-YOLO model, designed to replace or assist manual detection by leveraging deep learning and unmanned aerial systems (UASs). A comprehensive dataset was constructed, informed by an analysis of the agronomic characteristics of missed tassels during the detasseling period, including factors such as tassel visibility, plant height variability, and tassel development stages. The dataset captures diverse tassel images under varying lighting conditions, planting densities, and growth stages, with special attention to early tasseling stages when tassels are partially wrapped in leaves-a critical yet underexplored challenge for accurate detasseling. The MT-YOLO model demonstrates significant improvements in detection metrics, achieving an average precision (AP) of 93.1%, precision of 93.3%, recall of 91.6%, and an F1-score of 92.4%, outperforming Faster R-CNN, SSD, and various YOLO models. Compared to the baseline YOLO v5s, the MT-YOLO model increased recall by 1.1%, precision by 4.9%, and F1-score by 3.0%, while maintaining a detection speed of 124 fps. Field tests further validated its robustness, achieving a mean missed rate of 9.1%. These results highlight the potential of MT-YOLO as a reliable and efficient solution for enhancing detasseling efficiency in hybrid maize seed production.

PMID:39972352 | DOI:10.1186/s13007-025-01341-4

Nature. 2025 Feb 19. doi: 10.1038/s41586-025-08598-8. Online ahead of print.

ABSTRACT

The recognition of ligands by transmembrane proteins is essential for the exchange of materials, energy and information across biological membranes. Progress has been made in the de novo design of transmembrane proteins1-6, as well as in designing water-soluble proteins to bind small molecules7-12, but de novo design of transmembrane proteins that tightly and specifically bind to small molecules remains an outstanding challenge13. Here we present the accurate design of ligand-binding transmembrane proteins by integrating deep learning and energy-based methods. We designed pre-organized ligand-binding pockets in high-quality four-helix backbones for a fluorogenic ligand, and generated a transmembrane span using gradient-guided hallucination. The designer transmembrane proteins specifically activated fluorescence of the target fluorophore with mid-nanomolar affinity, exhibiting higher brightness and quantum yield compared to those of enhanced green fluorescent protein. These proteins were highly active in the membrane fraction of live bacterial and eukaryotic cells following expression. The crystal and cryogenic electron microscopy structures of the designer protein-ligand complexes were very close to the structures of the design models. We showed that the interactions between ligands and transmembrane proteins within the membrane can be accurately designed. Our work paves the way for the creation of new functional transmembrane proteins, with a wide range of applications including imaging, ligand sensing and membrane transport.

PMID:39972138 | DOI:10.1038/s41586-025-08598-8

Sci Rep. 2025 Feb 19;15(1):6070. doi: 10.1038/s41598-025-90363-y.

ABSTRACT

This work introduces a novel method for estimating hydrological loading displacement using 3D Convolutional Neural Networks (3D-CNN). This approach utilizes vertical displacement time series data from 41 Global Navigation Satellite System (GNSS) stations across Yunnan Province, China, and its adjacent areas, coupled with spatiotemporal variations in terrestrial water storage derived from the Gravity Recovery and Climate Experiment satellites (GRACE). The 3D-CNN method demonstrates markedly higher inversion precision compared to conventional load Green's function inversion techniques. This improvement is evidenced by substantial reductions in deviations from GNSS observations across various statistical metrics: the maximum deviation decreased by 1.34 millimeters, the absolute minimum deviation by 1.47 millimeters, the absolute mean deviation by 79.6%, and the standard deviation by 31.4%. An in-depth analysis of terrestrial water storage and loading displacement from 2019 to 2022 in Yunnan Province revealed distinct seasonal fluctuations, primarily driven by dominant annual and semi-annual cycles, and these periodic signals accounted for over 90% of the variance. The spatial distribution of terrestrial water loading displacement is strongly associated with regional precipitation patterns, showing smaller amplitudes in the northeast and northwest and larger amplitudes in the southwest. The research findings presented in this paper offer a novel perspective on the spatiotemporal variations of environmental load effects, particularly those related to the terrestrial water loading deformation with significant spatial heterogeneity. Accurate assessment of the effects of terrestrial water loading displacement (TWLD) is of considerable importance for precise geodetic observations, as well as for the establishment and maintenance of high-precision dynamic reference frames. Furthermore, the development of TWLD model that integrates GRACE and GNSS data provides valuable data support for the higher-precision inversion of changes in terrestrial water storage.

PMID:39972111 | DOI:10.1038/s41598-025-90363-y

Sci Rep. 2025 Feb 19;15(1):6099. doi: 10.1038/s41598-025-89738-y.

ABSTRACT

Convolutional neural networks (CNNs) and mammalian visual systems share architectural and information processing similarities. We leverage these parallels to develop an in-silico CNN model simulating diseases affecting the visual system. This model aims to replicate neural complexities in an experimentally controlled environment. Therefore, we examine object recognition and internal representations of a CNN under neurodegeneration and neuroplasticity conditions simulated through synaptic weight decay and retraining. This approach can model neurodegeneration from events like tau accumulation, reflecting cognitive decline in diseases such as posterior cortical atrophy, a condition that can accompany Alzheimer's disease and primarily affects the visual system. After each degeneration iteration, we retrain unaffected synapses to simulate ongoing neuroplasticity. Our results show that with significant synaptic decay and limited retraining, the model's representational similarity decreases compared to a healthy model. Early CNN layers retain high similarity to the healthy model, while later layers are more prone to degradation. The results of this study reveal a progressive decline in object recognition proficiency, mirroring posterior cortical atrophy progression. In-silico modeling of neurodegenerative diseases can enhance our understanding of disease progression and aid in developing targeted rehabilitation and treatments.

PMID:39972104 | DOI:10.1038/s41598-025-89738-y

Sci Rep. 2025 Feb 19;15(1):6124. doi: 10.1038/s41598-025-90572-5.

ABSTRACT

This research presents a novel ensemble fuzzy deep learning approach for brain Magnetic Resonance Imaging (MRI) analysis, aiming to improve the segmentation of brain tissues and abnormalities. The method integrates multiple components, including diverse deep learning architectures enhanced with volumetric fuzzy pooling, a model fusion strategy, and an attention mechanism to focus on the most relevant regions of the input data. The process begins by collecting medical data using sensors to acquire MRI images. These data are then used to train several deep learning models that are specifically designed to handle various aspects of brain MRI segmentation. To enhance the model's performance, an efficient ensemble learning method is employed to combine the predictions of multiple models, ensuring that the final decision accounts for different strengths of each individual model. A key feature of the approach is the construction of a knowledge base that stores data from training images and associates it with the most suitable model for each specific sample. During the inference phase, this knowledge base is consulted to quickly identify and select the best model for processing new test images, based on the similarity between the test data and previously encountered samples. The proposed method is rigorously tested on real-world brain MRI segmentation benchmarks, demonstrating superior performance in comparison to existing techniques. Our proposed method achieves an Intersection over Union (IoU) of 95% on the complete Brain MRI Segmentation dataset, demonstrating a 10% improvement over baseline solutions.

PMID:39972098 | DOI:10.1038/s41598-025-90572-5

Sci Rep. 2025 Feb 19;15(1):6029. doi: 10.1038/s41598-025-90084-2.

ABSTRACT

The limited availability of labeled ECG data restricts the application of supervised deep learning methods in ECG detection. Although existing self-supervised learning approaches have been applied to ECG analysis, they are predominantly image-based, which limits their effectiveness. To address these limitations and provide novel insights, we propose a Temporal-Spatial Self-Supervised Learning (TSSL) method specifically designed for ECG detection. TSSL leverages the intrinsic temporal and spatial characteristics of ECG signals to enhance feature representation. Temporally, ECG signals retain consistent identity information over time, enabling the model to generate stable representations for the same individual across different time points while isolating representations of different leads to preserve their unique features. Spatially, ECG signals from various leads capture the heart's activity from different perspectives, revealing both commonalities and distinct patterns. TSSL captures these correlations by maintaining consistency in the relationships between signals and their representations across different leads. Experimental results on the CPSC2018, Chapman, and PTB-XL databases demonstrate that TSSL introduces new capabilities by effectively utilizing temporal and spatial information, achieving superior performance compared to existing methods and approaching the performance of full-label training with only 10% of the labeled data. This highlights TSSL's ability to provide deeper insights and enhanced feature extraction beyond mere performance improvements. We make our code publicly available on https://github.com/cwp9731/temporal-spatial-self-supervised-learning.

PMID:39972080 | DOI:10.1038/s41598-025-90084-2

Sci Rep. 2025 Feb 19;15(1):6116. doi: 10.1038/s41598-025-90418-0.

ABSTRACT

Skin diseases, a significant category in the medical field, have always been challenging to diagnose and have a high misdiagnosis rate. Deep learning for skin disease classification has considerable value in clinical diagnosis and treatment. This study proposes a skin disease classification model based on multi-scale channel attention. The network architecture of the model consists of three main parts: an input module, four processing blocks, and an output module. Firstly, the model has improved the pyramid segmentation attention module to extract multi-scale features of the image entirely. Secondly, the reverse residual structure is used to replace the residual structure in the backbone network, and the attention module is integrated into the reverse residual structure to achieve better multi-scale feature extraction. Finally, the output module consists of an adaptive average pool and a fully connected layer, which convert the aggregated global features into several categories to generate the final output for the classification task. To verify the performance of the proposed model, this study used two commonly used skin disease datasets, ISIC2019 and HAM10000, for validation. The experimental results showed that the accuracy of this study was 77.6% on the ISIC2019 skin disease series dataset and 88.2% on the HAM10000 skin disease dataset. External validation data was added for evaluation to validate the model further, and the comprehensive evaluation results proved the effectiveness of the proposed model in this paper.

PMID:39972014 | DOI:10.1038/s41598-025-90418-0