JMIR Med Inform. 2025 Aug 6;13:e73631. doi: 10.2196/73631.

ABSTRACT

BACKGROUND: Seasonal influenza is a major global public health concern, leading to escalated morbidity and mortality rates. Traditional early warning models rely on binary (0/1) classification methods, which issue alerts only when predefined thresholds are crossed. However, these models exhibit inflexibility, often leading to false alarms or missed warnings and failing to provide granular risk assessments essential for decision-making. Therefore, we propose a probability-based early warning system using machine learning to mitigate these limitations and to offer continuous risk estimations of alerts (0-1 variable) instead of rigid threshold-based alerts. Based on probabilistic prediction, public health experts can make more flexible decisions in combination with the actual situation, significantly reducing the uncertainty and pressure in the decision-making process and reducing the waste of public health resources and the risk of social panic.

OBJECTIVE: The main aim of this study is to devise an innovative approach for early warning systems focused on influenza-like cases. Therefore, a Dense Residual Network (Dense ResNet), a supervised deep learning model, was developed. The model's training involved fitting the influenza-like illness positive rate, which enabled the early detection and warning of signals of changes occurring in the activity level of influenza-like cases. This departure from conventional methodologies underscores the transformative potential of machine learning, particularly in providing advanced capabilities for timely and proactive warnings in the context of influenza outbreaks.

METHODS: We developed a Dense ResNet machine learning model trained on influenza surveillance data from Northern and Southern China (2014-2024). This model generates early warning signals 3, 5, and 7 days in advance, providing a probability-based risk assessment represented as a continuous variable ranging from 0 to 1, in contrast to the traditional binary (0/1) warning systems. We evaluated the performance of this model using area under the curve scores, accuracy, recall, and F1-scores, then compared it with support vector machine (SVM), random forests, XGBoost (Extreme Gradient Boosting), and LSTM (long short-term memory) models.

RESULTS: The Dense ResNet model demonstrated the best performance, characterized by 5-day lead warnings and a 50th percentile probability threshold, achieving area under the curve scores of 0.94 (Northern China) and 0.95 (Southern China). Relative to traditional models, probability-based warning signals improved early detection, reduced false alarms, and facilitated tiered public health responses.

CONCLUSIONS: This study presented a novel probability-based machine learning model essential for early warning signals of influenza, demonstrating superior accuracy, flexibility, and practical applicability compared to other techniques. This approach enhances preparedness for influenza among the population and promotes the use of automated artificial intelligence-driven public health responses by replacing binary warnings with probability-driven risk assessments. Future research should integrate real-time surveillance data and dynamic transmission models to improve the precision of early warning.

PMID:40769217 | DOI:10.2196/73631

J Neural Eng. 2025 Aug 6. doi: 10.1088/1741-2552/adf888. Online ahead of print.

ABSTRACT

Surface electromyography (sEMG) and pressure-based force myography (pFMG) are two complementary modalities adopted in hand gesture recognition due to their ability to capture muscle electrical and mechanical activity, respectively. While sEMG carries rich neural information about the intended gestures and has long been established as the primary control signal in myoelectric interfaces, pFMG has recently emerged as a stable modality that is less sensitive to sweat and can indicate motion onset earlier than sEMG, making their fusion promising for robust pattern recognition. However, gesture classification systems based on these signals often suffer from performance degradation due to limb position changes, which affect signal characteristics. To address this, we introduce MyoPose, a novel and lightweight spatial synergy-based feature set for enhancing neuromechanical control. MyoPose works on effectively decoding colocated sEMG-pFMG information to improve hand gesture recognition under limb position variability while remaining computationally efficient for resource-constrained hardware. The proposed MyoPose feature combined with Linear Discriminant Analysis (LDA), achieved 87.7% accuracy in a nine-hand gesture recognition task, outperforming standard myoelectric feature sets and comparable to a state-of-the-art decision-level multimodal fusion parallel CNN. Notably, MyoPose maintained computational efficiency, achieving real-time feasibility with an estimated controller delay of 110.62 ms, well within the operational requirement of 100-125 ms, as well as ultra-light memory requirement of 0.011 KB. The novelty of this study lies in providing an effective feature set for multimodal driven hand gesture recognition, handling limb position variations with robust accuracy, and showing potential for real-time feasibility for human-machine interfaces without the need for deep learning.

PMID:40769169 | DOI:10.1088/1741-2552/adf888

Med Image Anal. 2025 Jul 29;106:103715. doi: 10.1016/j.media.2025.103715. Online ahead of print.

ABSTRACT

In pediatric orthodontics, accurate estimation of growth potential is essential for developing effective treatment strategies. Our research aims to predict this potential by identifying the growth peak and analyzing cervical vertebra morphology solely through lateral cephalometric radiographs. We accomplish this by comprehensively analyzing cervical vertebral maturation (CVM) features from these radiographs. This methodology provides clinicians with a reliable and efficient tool to determine the optimal timings for orthodontic interventions, ultimately enhancing patient outcomes. A crucial aspect of this approach is the meticulous annotation of keypoints on the cervical vertebrae, a task often challenged by its labor-intensive nature. To mitigate this, we introduce Attend-and-Refine Network (ARNet), a user-interactive, deep learning-based model designed to streamline the annotation process. ARNet features Interaction-guided recalibration network, which adaptively recalibrates image features in response to user feedback, coupled with a morphology-aware loss function that preserves the structural consistency of keypoints. This novel approach substantially reduces manual effort in keypoint identification, thereby enhancing the efficiency and accuracy of the process. Extensively validated across various datasets, ARNet demonstrates remarkable performance and exhibits wide-ranging applicability in medical imaging. In conclusion, our research offers an effective AI-assisted diagnostic tool for assessing growth potential in pediatric orthodontics, marking a significant advancement in the field.

PMID:40769097 | DOI:10.1016/j.media.2025.103715

Sci Rep. 2025 Aug 6;15(1):28804. doi: 10.1038/s41598-025-05574-0.

ABSTRACT

Steel defect detection is a crucial aspect of steel production and quality control. Therefore, focusing on small-scale defects in complex production environments remains a critical challenge. To address this issue, we propose an innovative perception-efficient network designed for the fast and accurate detection of multi-scale surface defects. First, we introduce the Defect Capture Path Aggregation Network, which enhances the feature fusion network's ability to learn multi-scale representations. Second, we design a Perception-Efficient Head (PEHead) to effectively mitigate local aliasing issues, thereby reducing the occurrence of missed detections. Finally, we propose the Receptive Field Extension Module (RFEM) to strengthen the backbone network's ability to capture global features and address extreme aspect ratio variations. These three modules can be seamlessly integrated into the YOLO framework. The proposed method is evaluated on three public steel defect datasets: NEU-DET, GC10-DET, and Severstal. Compared to the original YOLOv8n model, PEYOLO achieves mAP50 improvements of 3.5%, 9.1%, and 3.3% on these datasets, respectively. While maintaining similar detection accuracy, PEYOLO retains a high inference speed, making it suitable for real-time applications. Experimental results demonstrate that the proposed PEYOLO can be effectively applied to real-time steel defect detection.

PMID:40770273 | DOI:10.1038/s41598-025-05574-0

Sci Data. 2025 Aug 6;12(1):1366. doi: 10.1038/s41597-025-05718-x.

ABSTRACT

Artificial intelligence advancements have significantly enhanced computer-aided intervention, learning among surgeons, and analysis of surgical videos post-operation, substantially elevating surgical expertise and patient outcomes. Recognition systems for endoscopic surgical phases using deep learning algorithms heavily rely on comprehensive annotated datasets. Our research presents the Renji dataset featuring videos of endoscopic submucosal dissection (ESD) for colorectal neoplastic lesions (CNLs), which includes 30 procedural recordings with 130,298 phase-specific annotations collaboratively labeled by a team of three specialists in endoscopy. To our knowledge, this represents the first openly accessible collection of ESD videos specifically targeting CNLs treatment, and we anticipate this work will help establish standards for constructing similar ESD databases. Both the video collection and corresponding annotations have been made publicly accessible through the Figshare platform.

PMID:40770268 | DOI:10.1038/s41597-025-05718-x

Sci Rep. 2025 Aug 6;15(1):28814. doi: 10.1038/s41598-025-09739-9.

ABSTRACT

Conventional approaches to material decomposition in spectral CT face challenges related to precise algorithm calibration across imaged conditions and low signal quality caused by variable object size and reduced dose. In this proof-of-principle study, a deep learning approach to multi-material decomposition was developed to quantify iodine, gadolinium, and calcium in spectral CT. A dual-phase network architecture was trained using synthetic datasets containing computational models of cylindrical and virtual patient phantoms. Classification and quantification performance was evaluated across a range of patient size and dose parameters. The model was found to accurately classify (accuracy: cylinders - 98%, virtual patients - 97%) and quantify materials (mean absolute percentage difference: cylinders - 8-10%, virtual patients - 10-15%) in both datasets. Performance in virtual patient phantoms improved as the hybrid training dataset included a larger contingent of virtual patient phantoms (accuracy: 48% with 0 virtual patients to 97% with 8 virtual patients). For both datasets, the algorithm was able to maintain strong performance under challenging conditions of large patient size and reduced dose. This study shows the validity of a deep-learning based approach to multi-material decomposition trained with in-silico images that can overcome the limitations of conventional material decomposition approaches.

PMID:40770244 | DOI:10.1038/s41598-025-09739-9

Sci Rep. 2025 Aug 6;15(1):28787. doi: 10.1038/s41598-025-12679-z.

ABSTRACT

Optimizing clinical pathways is pivotal for enhancing healthcare delivery, yet traditional methods are increasingly insufficient in the face of complex, personalized medical demands. This paper introduces an innovative optimization framework that fuses Latent Dirichlet Allocation (LDA) topic modeling with Bidirectional Long Short-Term Memory (BiLSTM) networks to address the complexities of modern healthcare. The LDA component elucidates key diagnostic and treatment patterns from clinical narratives, while the BiLSTM network adeptly captures the temporal progression of patient care. Our model was validated against a real-world medical dataset, achieving remarkable results with an accuracy of over 90%, precision exceeding 28% improvement, recall with a 21% enhancement, and an F1 score that reflects a 25% increase over existing models. These results were obtained through comparative analysis with established models such as DeepCare, Doctor AI, and LSTM variants, showcasing the superior predictive capabilities of our LDA-BiLSTM integrated approach. This study not only advances the academic discourse on clinical pathway management but also presents a tangible tool for healthcare practitioners, promising a significant impact on the customization and efficacy of clinical pathways, thereby enhancing patient care and satisfaction.

PMID:40770243 | DOI:10.1038/s41598-025-12679-z

Sci Data. 2025 Aug 6;12(1):1369. doi: 10.1038/s41597-025-05708-z.

ABSTRACT

Deep learning models can accelerate the processing of image-based biodiversity data and provide educational value by giving direct feedback to citizen scientists. However, the training of such models requires large amounts of labelled data and not all species are equally suited for identification from images alone. Most butterfly and many moth species (Lepidoptera) which play an important role as biodiversity indicators are well-suited for such approaches. This dataset contains over 540.000 images of 185 butterfly and moth species that occur in Austria. Images were collected by citizen scientists with the application "Schmetterlinge Österreichs" and correct species identification was ensured by an experienced entomologist. The number of images per species ranges from one to nearly 30.000. Such a strong class imbalance is common in datasets of species records. The dataset is larger than other published dataset of butterfly and moth images and offers opportunities for the training and evaluation of machine learning models on the fine-grained classification task of species identification.

PMID:40770239 | DOI:10.1038/s41597-025-05708-z

Sci Rep. 2025 Aug 6;15(1):28723. doi: 10.1038/s41598-025-14406-0.

ABSTRACT

Alzheimer's disease (AD) presents a pressing global health challenge, demanding improved strategies for early detection and understanding its progression. In this study, we address this need by employing survival analysis techniques to predict transition time from Cognitive Normal (CN) to Mild Cognitive Impairment (MCI) in elderly individuals, considering the predictive value of baseline comorbidities. Leveraging data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and Australian Imaging, Biomarker & Lifestyle Flagship Study of Ageing (AIBL) databases, we construct feature sets encompassing demographics, cognitive scores, and comorbidities. Various machine learning and deep learning methods for survival analysis are employed. Our top-performing model, fast random forest, achieves a concordance index of 0.84 when considering all feature modalities, with comorbidity data emerging as a significant predictor. The top features identified by the best-performing model include one demographic feature (age), seven cognitive scores (ADAS13, RAVLT learning, FAQ, ADAS11, RAVLT immediate, CDRSB, ADASQ4), and two comorbidities (Endocrine & Metabolic, Renal & Genitourinary). Age is highlighted as the most influential predictor, while cognitive scores are crucial indicators of Alzheimer's disease. External validation against the AIBL dataset affirms the robustness of our approach. Overall, our study contributes to a deeper understanding of the role of baseline comorbidities in AD risk prediction and emphasizes the importance of incorporating comprehensive feature assessment in clinical practice for early diagnosis and personalized treatment planning.

PMID:40770222 | DOI:10.1038/s41598-025-14406-0

Eur Radiol. 2025 Aug 6. doi: 10.1007/s00330-025-11801-z. Online ahead of print.

ABSTRACT

OBJECTIVES: This study aimed to evaluate the potential additional value of deep radiomics for assessing residual cancer burden (RCB) in locally advanced breast cancer, after neoadjuvant chemotherapy (NAC) but before surgery, compared to standard predictors: tumor volume and subtype.

MATERIALS AND METHODS: This retrospective study used a 105-patient single-institution training set and a 41-patient external test set from three institutions in the LIMA trial. DCE-MRI was performed before and after NAC, and RCB was determined post-surgery. Three networks (nnU-Net, Attention U-net and vector-quantized encoder-decoder) were trained for tumor segmentation. For each network, deep features were extracted from the bottleneck layer and used to train random forest regression models to predict RCB score. Models were compared to (1) a model trained on tumor volume and (2) a model combining tumor volume and subtype. The potential complementary performance of combining deep radiomics with a clinical-radiological model was assessed. From the predicted RCB score, three metrics were calculated: area under the curve (AUC) for categories RCB-0/RCB-I versus RCB-II/III, pathological complete response (pCR) versus non-pCR, and Spearman's correlation.

RESULTS: Deep radiomics models had an AUC between 0.68-0.74 for pCR and 0.68-0.79 for RCB, while the volume-only model had an AUC of 0.74 and 0.70 for pCR and RCB, respectively. Spearman's correlation varied from 0.45-0.51 (deep radiomics) to 0.53 (combined model). No statistical difference between models was observed.

CONCLUSIONS: Segmentation network-derived deep radiomics contain similar information to tumor volume and subtype for inferring pCR and RCB after NAC, but do not complement standard clinical predictors in the LIMA trial.

KEY POINTS: Question It is unknown if and which deep radiomics approach is most suitable to extract relevant features to assess neoadjuvant chemotherapy response on breast MRI. Findings Radiomic features extracted from deep-learning networks yield similar results in predicting neoadjuvant chemotherapy response as tumor volume and subtype in the LIMA study. However, they do not provide complementary information. Clinical relevance For predicting response to neoadjuvant chemotherapy in breast cancer patients, tumor volume on MRI and subtype remain important predictors of treatment outcome; deep radiomics might be an alternative when determining tumor volume and/or subtype is not feasible.

PMID:40770139 | DOI:10.1007/s00330-025-11801-z

Commun Biol. 2025 Aug 6;8(1):1159. doi: 10.1038/s42003-025-08590-y.

ABSTRACT

Deep learning for microbiome analysis has shown potential for understanding microbial communities and human phenotypes. Here, we propose an approach, Transformer-based Robust Principal Component Analysis(TRPCA), which leverages the strengths of transformer architectures and interpretability of Robust Principal Component Analysis. To investigate benefits of TRPCA over conventional machine learning models, we benchmarked performance on age prediction from three body sites(skin, oral, gut), with 16S rRNA gene amplicon(16S) and whole-genome sequencing(WGS) data. We demonstrated prediction of age from longitudinal samples and combined classification and regression tasks via multi-task learning(MTL). TRPCA improves age prediction accuracy from human microbiome samples, achieving the largest reduction in Mean Absolute Error for WGS skin (MAE: 8.03, 28% reduction) and 16S skin (MAE: 5.09, 14% reduction) samples, compared to conventional approaches. Additionally, TRPCA's MTL approach achieves an accuracy of 89% for birth country prediction across 5 countries, while improving age prediction from WGS stool samples. Notably, TRPCA uncovers a link between subject and error prediction through residual analysis for paired samples across sequencing method (16S/WGS) and body site(oral/gut). These findings highlight TRPCA's utility in improving age prediction while maintaining feature-level interpretability, and elucidating connections between individuals and microbiomes.

PMID:40770074 | DOI:10.1038/s42003-025-08590-y

Sci Rep. 2025 Aug 7;15(1):28847. doi: 10.1038/s41598-025-14576-x.

ABSTRACT

Data binary encoding has proven to be a versatile tool for optimizing data processing and memory efficiency in various machine learning applications. This includes deep barcoding, generating barcodes from deep learning feature extraction for image retrieval of similar cases among millions of indexed images. Despite the recent advancement in barcode generation methods, converting high-dimensional feature vectors (e.g., deep features) to compact and discriminative binary barcodes is still an urgent necessity and remains an unresolved problem. Difference-based binarization of features is one of the most efficient binarization methods, transforming continuous feature vectors into binary sequences and capturing trend information. However, the performance of this method is highly dependent on the ordering of the input features, leading to a significant combinatorial challenge. This research addresses this problem by optimizing feature sequences based on retrieval performance metrics. Our approach identifies optimal feature orderings, leading to substantial improvements in retrieval effectiveness compared to arbitrary or default orderings. We assess the performance of the proposed approach in various medical and non-medical image retrieval tasks. This evaluation includes medical images from The Cancer Genome Atlas (TCGA), a comprehensive publicly available dataset, as well as COVID-19 Chest X-rays dataset. In addition, we evaluate the proposed approach on non-medical benchmark image datasets, such as CIFAR-10, CIFAR-100, and Fashion-MNIST. Our findings demonstrate the importance of optimizing binary barcode representation to significantly enhance accuracy for fast image retrieval across a wide range of applications, highlighting the applicability and potential of barcodes in various domains.

PMID:40770058 | DOI:10.1038/s41598-025-14576-x

Sci Rep. 2025 Aug 6;15(1):28744. doi: 10.1038/s41598-025-13646-4.

ABSTRACT

Decoding of continuous speech from electroencephalography (EEG) presents a promising avenue for understanding neural mechanisms of auditory processing and developing applications in hearing diagnostics. Recent advances in deep learning have improved decoding accuracy. However, challenges remain due to the low signal-to-noise ratio of the recorded brain signals. This study explores the application of contrastive learning, a self-supervised learning technique, to learn robust latent representations of EEG signals. We introduce a novel model architecture that leverages contrastive learning and transformer networks to capture relationships between auditory stimuli and EEG responses. Our model is evaluated on two tasks from the ICASSP 2023 Auditory EEG Decoding Challenge: a binary stimulus classification task (match-mismatch) and stimulus envelope decoding. We achieve state-of-the-art performance on both tasks, significantly outperforming previous winners with 87% accuracy in match-mismatch classification and a 0.176 Pearson correlation in envelope regression. Furthermore, we investigate the impact of model architecture, training set size, and finetuning on decoding performance, providing insights into the factors influencing model generalizability and accuracy. Our findings underscore the potential of contrastive learning for advancing the field of auditory EEG decoding and its potential applications in clinical settings.

PMID:40770040 | DOI:10.1038/s41598-025-13646-4

Phytomedicine. 2025 Aug 5;146:157127. doi: 10.1016/j.phymed.2025.157127. Online ahead of print.

ABSTRACT

BACKGROUND: Epidermal growth factor receptor-tyrosine kinase inhibitors (EGFR-TKI) are widely regarded as the most promising strategy to treat non-small cell lung cancer (NSCLC). While the existing development model cannot meet the requirements.

METHODS: Here, we developed a drug prediction and screening platform based on state-of-the-art deep learning (DL) algorithms and cell membrane chromatography (CMC). Using this platform, we discovered several novel EGFR antagonists from massive molecules. The interactions between target compounds and EGFR were investigated using frontal analysis, Surface plasmon resonance analysis, cellular thermal shift assay, and molecule docking. Xenografts models were established to study the anti-tumor activity of most promising compound. Proteomics, immunofluorescence, and Western blots were conducted to explore its mechanism.

RESULTS: The deep neural network had good performance, with an area under the curve of 0.950 ± 0.012. Among the top 100 molecules predicted by the model, 25 compounds had retention on SNAP-tagged (ST)-EGFR/CMC, of which 12 molecules exhibited significant anti-tumor activity. Interaction analysis revealed that Geraniin, Brazilin, and 7-Epitaxol would directly bind to EGFR and inhibit its activation. 7-Epitaxol exhibited significant anti-tumor activity in vivo and in vitro. 7-Epitaxol combined with EGFR and inhibited its phosphorylation, blocked PI3K/AKT pathways, thereby exerting its anti-tumor activity by promoting autophagy in A549 cells.

CONCLUSION: The results provide a novel and powerful platform for drug discovery and development in NSCLC research. By using this platform, Geraniin, Brazilin, and 7-epitaxol were identified as novel EGFR antagonists. We also innovatively demonstrated that 7-epitaxol promotes autophagy in NSCLC.

PMID:40768807 | DOI:10.1016/j.phymed.2025.157127

Cogn Emot. 2025 Aug 6:1-20. doi: 10.1080/02699931.2025.2543569. Online ahead of print.

ABSTRACT

Deep neural networks (DNNs) provide a useful computational framework for constructing cognitive models of emotion understanding. This paper provides a focused discussion of the use of DNNs in this context. It begins by defining three key components of emotion understanding - perception, prediction, and regulation - and discussing how each can be modelling using different deep learning architectures. It continues by positioning what DNN models can contribute to affective science in relation to important existing theoretical perspectives, including both domain-general frameworks like Bayesian cognitive modelling, and domain-specific frameworks, such as the theory of constructed emotion. The paper highlights both the strengths and limitations of DNNs as cognitive models and provides guidance for how to capitalise on the former while mitigating the latter.

PMID:40768739 | DOI:10.1080/02699931.2025.2543569

JMIR Med Inform. 2025 Aug 6;13:e66727. doi: 10.2196/66727.

ABSTRACT

BACKGROUND: The assessment of osteonecrosis of the femoral head (ONFH) often presents challenges in accuracy and efficiency. Traditional methods rely on imaging studies and clinical judgment, prompting the need for advanced approaches. This study aims to use deep learning algorithms to enhance disease assessment and prediction in ONFH, optimizing treatment strategies.

OBJECTIVE: The primary objective of this research is to analyze pathological images of ONFH using advanced deep learning algorithms to evaluate treatment response, vascular reconstruction, and disease progression. By identifying the most effective algorithm, this study seeks to equip clinicians with precise tools for disease assessment and prediction.

METHODS: Magnetic resonance imaging (MRI) data from 30 patients diagnosed with ONFH were collected, totaling 1200 slices, which included 675 slices with lesions and 225 normal slices. The dataset was divided into training (630 slices), validation (135 slices), and test (135 slices) sets. A total of 10 deep learning algorithms were tested for training and optimization, and MobileNetV3_Large was identified as the optimal model for subsequent analyses. This model was applied for quantifying vascular reconstruction, evaluating treatment responses, and assessing lesion progression. In addition, a long short-term memory (LSTM) model was integrated for the dynamic prediction of time-series data.

RESULTS: The MobileNetV3_Large model demonstrated an accuracy of 96.5% (95% CI 95.1%-97.8%) and a recall of 94.8% (95% CI 93.2%-96.4%) in ONFH diagnosis, significantly outperforming DenseNet201 (87.3%; P<.05). Quantitative evaluation of treatment responses showed that vascularized bone grafting resulted in an average increase of 12.4 mm in vascular length (95% CI 11.2-13.6 mm; P<.01) and an increase of 2.7 in branch count (95% CI 2.3-3.1; P<.01) among the 30 patients. The model achieved an AUC of 0.92 (95% CI 0.90-0.94) for predicting lesion progression, outperforming traditional methods like ResNet50 (AUC=0.85; P<.01). Predictions were consistent with clinical observations in 92.5% of cases (24/26).

CONCLUSIONS: The application of deep learning algorithms in examining treatment response, vascular reconstruction, and disease progression in ONFH presents notable advantages. This study offers clinicians a precise tool for disease assessment and highlights the significance of using advanced technological solutions in health care practice.

PMID:40768653 | DOI:10.2196/66727

Sci Adv. 2025 Aug 8;11(32):eadr8901. doi: 10.1126/sciadv.adr8901. Epub 2025 Aug 6.

ABSTRACT

In this study, postimplantation human epiblast and amnion development are modeled using a stem cell-based embryoid system. A dataset of 3697 fluorescent images, along with tissue, cavity, and cell masks, is generated from experimental data. A computational pipeline analyzes morphological and marker expression features, revealing key developmental processes such as tissue growth, cavity expansion, and cell differentiation. To uncover hidden developmental dynamics, a deep manifold learning framework is introduced. This framework uses an autoencoder to project embryoid images into a twenty-dimensional (20D) latent space and models the dynamics using a mean-reverting stochastic process of mixed Gaussians. The approach accurately captures phenotypic changes observed at discrete experimental time points. Moreover, it enables the generation of artificial yet realistic embryoid images at finer temporal resolutions, providing deeper insights into the progression of early human development.

PMID:40768579 | DOI:10.1126/sciadv.adr8901

PLoS One. 2025 Aug 6;20(8):e0327315. doi: 10.1371/journal.pone.0327315. eCollection 2025.

ABSTRACT

Land cover and plant species identification using satellite images and deep learning approaches have recently been a widely addressed area of research. However, mangroves, a specific species that have significantly declined in quantity and quality worldwide despite their numerous benefits, have not been the subject of attention. The novelty of this research is to deal with this species based on an advanced deep learning solution (a proposed ensemble U-Net model) and a high-resolution Planet satellite imagery (5 m x 5 m) in a case study of Ngoc Hien district, Ca Mau province, Vietnam. Twelve single U-Net backbone models were trained, and three quantitative metrics (Intersection over Union, F1-score, and Overall Accuracy) were used to evaluate. The findings indicate that three out of twelve models (MobileNet, SEResNeXt-101 and Efficientnet-B7) experienced the most efficient assessment results for identifying all classes, in which the MobileNet model was the best. These models were applied for the ensemble model's development. The ensemble model's quantitative assessment metrics increased considerably by about 3-10% compared to the single-component models. The IoU, F1-score, and OA values of this model were 80.08%, 95.82%, and 95.90%, respectively. Three classes of mangrove species (Avicennia alba, Rhizophora apiculate, and mixed mangroves) in the ensemble model had more uniform assessment results. In conclusion, to achieve optimal classification outcomes, a land-cover map comprising mangrove species is possibly established using the proposed ensemble model, while a distribution map of mangrove species enables to be developed using the MobileNet model.

PMID:40768506 | DOI:10.1371/journal.pone.0327315

PLoS One. 2025 Aug 6;20(8):e0321239. doi: 10.1371/journal.pone.0321239. eCollection 2025.

ABSTRACT

Mass spectrometry (MS) analysis plays a crucial role in the biomedical field; however, the high dimensionality and complexity of MS data pose significant challenges for feature extraction and classification. Deep learning has become a dominant approach in data analysis, and while some deep learning methods have achieved progress in MS classification, their feature representation capabilities remain limited. Most existing methods rely on single-channel representations, which struggle to effectively capture structural information within MS data. To address these limitations, we propose a Multi-Channel Embedding Representation Module (MSMCE), which focuses on modeling inter-channel dependencies to generate multi-channel representations of raw MS data. Additionally, we implement a feature fusion mechanism by concatenating the initial encoded representation with the multi-channel embeddings along the channel dimension, significantly enhancing the classification performance of subsequent models. Experimental results on four public datasets demonstrate that the proposed MSMCE module not only achieves substantial improvements in classification performance but also enhances computational efficiency and training stability, highlighting its effectiveness in raw MS data classification and its potential for robust application across diverse datasets.

PMID:40768503 | DOI:10.1371/journal.pone.0321239

IEEE Trans Neural Syst Rehabil Eng. 2025 Aug 6;PP. doi: 10.1109/TNSRE.2025.3596488. Online ahead of print.

ABSTRACT

OBJECTIVES: After stroke, hand function assessments are used as outcome measures to evaluate new rehabilitation therapies, but do not reflect true performance in natural environments. Wearable (egocentric) cameras provide a way to capture hand function information during activities of daily living (ADLs). However, while clinical assessments involve observing multiple functional tasks, existing deep learning methods developed to analyze hands in egocentric video are only capable of considering single ADLs. This study presents a novel multi-video architecture that processes multiple task videos to make improved estimations about hand impairment.

METHODS: An egocentric video dataset of ADLs performed by stroke survivors in a home simulation lab was used to develop single and multi-input video models for binary impairment classification. Using SlowFast as a base feature extractor, late fusion (majority voting, fully-connected network) and intermediate fusion (concatenation, Markov chain) were investigated for building multi-video architectures.

RESULTS: Through evaluation with Leave-One-Participant-Out-Cross-Validation, using intermediate concatenation fusion to build multi-video models was found to achieve the best performance out of the fusion techniques. The resulting multi-video model for cropped inputs achieved an F1-score of 0.778±0.129 and significantly outperformed its single-video counterpart (F1-score of 0.696±0.102). Similarly, the multi-video model for full-frame inputs (F1-score of 0.796±0.102) significantly outperformed its single-video counterpart (F1-score of 0.708±0.099).

CONCLUSION: Multi-video architectures are beneficial for estimating hand impairment from egocentric video after stroke.

SIGNIFICANCE: The proposed deep learning solution is the first of its kind in multi-video analysis, and opens the door to further applications in automating other multi-observation assessments for clinical use.

PMID:40768474 | DOI:10.1109/TNSRE.2025.3596488