IEEE J Biomed Health Inform. 2025 May 28;PP. doi: 10.1109/JBHI.2025.3574688. Online ahead of print.

ABSTRACT

The electrocardiogram (ECG) serves as a crucial tool for myocardial infarction (MI) localization, and deep learning methods have proven effective in assisting physicians with MI localization. Traditional MI localization methods are purely data-driven, and the quality of the data significantly affects the model's performance, particularly in the localization of rare MI. We propose a knowledgedriven graph representation learning (KD-GRL) framework which is designed to guide deep learning models in identifying key features for MI localization using prior knowledge. The MI localization knowledge graph (KG) is constructed by integrating medical knowledge about MI localization, including ECG leads and morphological manifestations, the correlations between MI localization labels, diagnostic rules, and patient demographic information. KG effectively represents the relationships among various entities, which include ECG signal entities, morphological feature entities, and demographic feature entities. The embeddings of these entities are obtained using parallel patient multi-feature extractors. Additionally, a KG aggregation method based on edge relation projection (ERP) is proposed to aggregate the relational information in the MI localization KG. Ultimately, the MI localization task is transformed into a link prediction task between patient entity and localization label entities within the KG. We conduct experiments on two public datasets, PTB and PTBXL, achieving F1-scores of 48.90% and 46.06%, respectively, both surpassing the comparison methods. Additionally, due to the incorporation of diagnostic knowledge, our method outperforms the comparison methods in localizing rare MIs.

PMID:40434860 | DOI:10.1109/JBHI.2025.3574688

Med Oncol. 2025 May 28;42(7):222. doi: 10.1007/s12032-025-02759-5.

ABSTRACT

Pancreatic cancer is a highly aggressive malignancy with rising incidence and mortality rates, often diagnosed at advanced stages. Conventional imaging methods, such as computed tomography (CT) and magnetic resonance imaging (MRI), struggle to assess tumor characteristics and vascular involvement, which are crucial for treatment planning. This paper explores the potential of diffusion magnetic resonance imaging (dMRI) in enhancing pancreatic cancer diagnosis and treatment. Diffusion-based techniques, such as diffusion-weighted imaging (DWI), diffusion tensor imaging (DTI), intravoxel incoherent motion (IVIM), and diffusion kurtosis imaging (DKI), combined with emerging AI‑powered analysis, provide insights into tissue microstructure, allowing for earlier detection and improved evaluation of tumor cellularity. These methods may help assess prognosis and monitor therapy response by tracking diffusion and perfusion metrics. However, challenges remain, such as standardized protocols and robust data analysis pipelines. Ongoing research, including deep learning applications, aims to improve reliability, and dMRI shows promise in providing functional insights and improving patient outcomes. Further clinical validation is necessary to maximize its benefits.

PMID:40434720 | DOI:10.1007/s12032-025-02759-5

Trop Anim Health Prod. 2025 May 28;57(5):230. doi: 10.1007/s11250-025-04475-8.

ABSTRACT

Lumpy Skin Disease (LSD) presents significant risks and economic challenges to global cattle farming. Effective and accurate classification of LSD is essential for managing the disease and reducing its impacts. Manual diagnosis is time-consuming, labor-intensive, and requires experienced personnel. Automated classification methods provide advantages by reducing labor and improving accuracy. This study proposes an automated algorithm for LSD classification using machine learning. The method uses a carefully curated dataset of images from both LSD-infected cattle and healthy cattle. Inception V3 was employed to extract features from complex lesion patterns in infected cattle images, comparing them to healthy cattle images. Support Vector Machines (SVM) were used to classify the extracted features. The results show the model achieved an 84% accuracy rate, with precision at 80%, recall at 83%, and an F1 score of 82%. These results were compared with other machine learning models, including Logistic Regression, Random Forest, Decision Tree, and AdaBoost. SVM outperformed other models, demonstrating consistent evaluation precision at 0.84. For further enhancement, expanding the dataset with high-quality images and applying advanced machine learning algorithms like Vision Transformers (ViTs), MobileNetV2, and Visual Geometry Group (VGG) could refine automated LSD classification. The aim is to improve disease management practices in the livestock industry through better classification systems.

PMID:40434587 | DOI:10.1007/s11250-025-04475-8

Indian J Ophthalmol. 2025 Jun 1;73(6):797-806. doi: 10.4103/IJO.IJO_1810_24. Epub 2025 May 28.

ABSTRACT

Recent advances in deep learning and machine learning have greatly increased the capabilities of extracting features for evaluating the response to anti VEGF treatment in patients with Diabetic Macular Edema (DME). In this review, we explore how these algorithms can be used for discriminating between responders and non-responders to anti vascular endothelial growth factor (VEGF) injections. Electronic databases, including PubMed, IEEE Xplore, BioMed, JAMA, and Google Scholar, were searched, and reference lists from relevant publications were also considered from inception till August 31, 2023, based on the inclusion and exclusion criteria. Data extraction was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The results focus on keywords such as DME, OCT, anti VEGF, and patient responses after anti VEGF injections. The article measures the effectiveness of different machine learning and deep learning algorithms, including linear discriminant analysis (LDA), ResNet-50, CNN with attention, quadratic discriminant analysis (QDA), random forest (RF), and support vector machines (SVM), in analyzing eyes that could tolerate extended interval dosing. According to a review of 50 relevant papers published between 2016 and 2023, the algorithms achieved an average automated sensitivity of 74% (95% CI: 0.55-0.92) in detecting treatment responses.

PMID:40434455 | DOI:10.4103/IJO.IJO_1810_24

Am J Physiol Lung Cell Mol Physiol. 2025 May 28. doi: 10.1152/ajplung.00058.2025. Online ahead of print.

ABSTRACT

Intracellular calcium (Ca2+) release via phospholipase C (PLC) following G-protein coupled receptor (GPCR) activation is typically linked to membrane depolarization and airway smooth muscle (ASM) contraction. However, recent findings show that bitter taste receptor agonists, such as chloroquine (CQ), induce a paradoxical and potent relaxation response despite activating the Ca2+ signaling pathway. This relaxation has been hypothesized to be driven by a distinct compartmentalization of calcium ions toward the cellular periphery, subsequently leading to membrane hyperpolarization, in contrast to the contractile effects of histamine. In this study, we further investigate the spatio-temporal dynamics of Ca2+ signaling in ASM cells using single-cell microscopy and deep learning-based segmentation, integrating the results into a comprehensive model of ASM ion channel dynamics to compare the effects of histamine, CQ, and flufenamic acid (FFA). Our results show that histamine induces a strong, synchronized calcium release, nearly twice as high as that of CQ, which produces a sustained but lower- magnitude response. Per-cell analysis reveals more variable and asynchronous Ca2+ signaling for CQ and FFA, with higher entropy compared to histamine. Integrating these findings into an ASM ion channel model, we observed that histamine-mediated Ca2+ release activates voltage-gated Ca2+ and Na+ channels (leading to depolarization). In contrast, CQ preferentially engages BKCa, SKCa, and chloride channels (promoting hyperpolarization). These findings provide insights into the unique mechanisms by which bitter taste receptor agonists can modulate ASM tone, offering potential therapeutic strategies for relaxing ASM and alleviating airway hyperresponsiveness in conditions such as asthma.

PMID:40434402 | DOI:10.1152/ajplung.00058.2025

Radiol Artif Intell. 2025 May 28:e240484. doi: 10.1148/ryai.240484. Online ahead of print.

ABSTRACT

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To estimate the total lung volume (TLV) from real and synthetic frontal chest radiographs (CXR) on a pixel level using lung thickness maps generated by a U-Net deep learning model. Materials and Methods This retrospective study included 5,959 chest CT scans from two public datasets: the lung nodule analysis 2016 (n = 656) and the Radiological Society of North America (RSNA) pulmonary embolism detection challenge 2020 (n = 5,303). Additionally, 72 participants were selected from the Klinikum Rechts der Isar dataset (October 2018 to December 2019), each with a corresponding chest radiograph taken within seven days. Synthetic radiographs and lung thickness maps were generated using forward projection of CT scans and their lung segmentations. A U-Net model was trained on synthetic radiographs to predict lung thickness maps and estimate TLV. Model performance was assessed using mean squared error (MSE), Pearson correlation coefficient (r), and two-sided Student's t-distribution. Results The study included 72 participants (45 male, 27 female, 33 healthy: mean age 62 years [range 34-80]; 39 with chronic obstructive pulmonary disease: mean age 69 years [range 47-91]). TLV predictions showed low error rates (MSEPublic-Synthetic = 0.16 L2, MSEKRI-Synthetic = 0.20 L2, MSEKRI-Real = 0.35 L2) and strong correlations with CT-derived reference standard TLV (nPublic-Synthetic = 1,191, r = 0.99, P < .001; nKRI-Synthetic = 72, r = 0.97, P < .001; nKRI-Real = 72, r = 0.91, P < .001). When evaluated on different datasets, the U-Net model achieved the highest performance for TLV estimation on the Luna16 test dataset, with the lowest mean squared error (MSE = 0.09 L2) and strongest correlation (r = 0.99, P <.001) compared with CT-derived TLV. Conclusion The U-Net-generated pixel-level lung thickness maps successfully estimated TLV for both synthetic and real radiographs. ©RSNA, 2025.

PMID:40434310 | DOI:10.1148/ryai.240484

Radiol Artif Intell. 2025 May;7(3):e250260. doi: 10.1148/ryai.250260.

NO ABSTRACT

PMID:40434277 | DOI:10.1148/ryai.250260

Small Methods. 2025 May 28:e2500106. doi: 10.1002/smtd.202500106. Online ahead of print.

ABSTRACT

Advances in organoid models, as ex vivo mini-organs, and the development of screening imaging technologies have continuously driven each other forward. A complete understanding of organoids requires detailed insights into the intertwined intraorganoid and extraorganoid activities and how they change across time and space. This study introduces a multiplexed imaging platform that integrates label-free nanoplasmonic biosensing with fluorescence microscopy to offer simultaneous monitoring of dynamics occurring within and around arrays of single spheroids with spatiotemporal resolution. The label-free module employs nanoplasmonic biosensors with extraordinary optical transmission to track biomolecular secretions into the surroundings, while concurrent fluorescence imaging enables structural analysis and viability assessment. To perform multiparametric interrogation of the data from different channels over extended periods, a deep-learning-augmented image analysis is incorporated. The platform is applied to tumor spheroids to investigate vascular endothelial growth factor A secretion alongside morphometric changes and viability, showcasing its ability to capture variations in secretion and growth dynamics between untreated and drug-treated groups. This integrated approach advances comprehensive insights into organoid models and can complement existing technologies to accelerate discoveries in disease modeling and drug development.

PMID:40434268 | DOI:10.1002/smtd.202500106

Comput Methods Biomech Biomed Engin. 2025 May 28:1-14. doi: 10.1080/10255842.2025.2511222. Online ahead of print.

ABSTRACT

Major Depressive Disorder (MDD) is known as a widespread illness and needs a timely treatment. The treatment procedure is currently based on the trial and error between various treatments. An individualized treatment selection is crucial for saving time and financial resources and preventing possible side effects. Because of the complex nature of this problem, a Deep Learning (DL) approach, as a promising method for the precision medicine, was utilized to identify the responders to the treatment using pre-treatment EEG signals. Eighty-three patients with MDD participated in this study to receive treatment using Repetitive Transcranial Magnetic Stimulation (rTMS). A deep hybrid neural network was developed based on three pre-trained convolutional neural networks named DenseNet121, EfficientNetB0, and Xception. The training of each model was performed by feeding three types of EEG representations as the inputs into the models including the Wavelet Transform (WT) images, the connectivity matrix between electrode pairs, and the raw EEG signals. The performance of the proposed models were assessed for the three different input types and achieved the highest accuracy of 94.7% in classifying patients as responders or non-responders when utilizing a sequence of raw EEG images. For the WT and connectivity inputs the best accuracy of model was 94.38% and 94.25% respectively. Therefore, the proposed model can be a step forward towards the clinical implementation of an end-to-end treatment selection framework using raw EEG signals.

PMID:40434017 | DOI:10.1080/10255842.2025.2511222

Integr Org Biol. 2024 Sep 23;6(1):obae036. doi: 10.1093/iob/obae036. eCollection 2024.

ABSTRACT

Artificial intelligence (AI) is poised to revolutionize many aspects of science, including the study of evolutionary morphology. While classical AI methods such as principal component analysis and cluster analysis have been commonplace in the study of evolutionary morphology for decades, recent years have seen increasing application of deep learning to ecology and evolutionary biology. As digitized specimen databases become increasingly prevalent and openly available, AI is offering vast new potential to circumvent long-standing barriers to rapid, big data analysis of phenotypes. Here, we review the current state of AI methods available for the study of evolutionary morphology, which are most developed in the area of data acquisition and processing. We introduce the main available AI techniques, categorizing them into 3 stages based on their order of appearance: (1) machine learning, (2) deep learning, and (3) the most recent advancements in large-scale models and multimodal learning. Next, we present case studies of existing approaches using AI for evolutionary morphology, including image capture and segmentation, feature recognition, morphometrics, and phylogenetics. We then discuss the prospectus for near-term advances in specific areas of inquiry within this field, including the potential of new AI methods that have not yet been applied to the study of morphological evolution. In particular, we note key areas where AI remains underutilized and could be used to enhance studies of evolutionary morphology. This combination of current methods and potential developments has the capacity to transform the evolutionary analysis of the organismal phenotype into evolutionary phenomics, leading to an era of "big data" that aligns the study of phenotypes with genomics and other areas of bioinformatics.

PMID:40433986 | PMC:PMC12082097 | DOI:10.1093/iob/obae036

Adv Sci (Weinh). 2025 May 28:e14732. doi: 10.1002/advs.202414732. Online ahead of print.

ABSTRACT

Accurate decoding of peripheral nerve signals is essential for advancing neuroscience research, developing therapeutics for neurological disorders, and creating reliable human-machine interfaces. However, the poor mechanical compliance of conventional metal electrodes and limited generalization of existing decoding models have significantly hindered progress in understanding peripheral nerve function. This study introduces low-impedance, soft-conducting polymer electrodes capable of forming stable, intimate contacts with peripheral nerve tissues, allowing for continuous and reliable recording of neural activity in awake animals. Using this unique dataset of neurophysiological signals, a neural network model that integrates both handcrafted and deep learning-derived features, while incorporating parameter-sharing and adaptation training strategies, is developed. This approach significantly improves the generalizability of the decoding model across subjects, reducing the reliance on extensive training data. The findings not only deepen the understanding of peripheral nerve function but also open avenues for developing advanced interventions to augment and restore neurological functions.

PMID:40433949 | DOI:10.1002/advs.202414732

Respirology. 2025 May 28. doi: 10.1111/resp.70057. Online ahead of print.

ABSTRACT

BACKGROUND AND OBJECTIVE: Iriscope, a 1.3 mm video endoscopic probe introduced through an r-EBUS catheter, allows for the direct visualisation of small peripheral pulmonary nodules (PPNs). This study assessed the ability of physicians with different levels of experience in bronchoscopy, and the ability of artificial intelligence (AI) to predict the malignant nature of small PPNs during Iriscope peripheral endoscopy.

METHODS: Patients undergoing bronchoscopy with r-EBUS and Iriscope for peripheral PPNs < 20 mm with a definite diagnosis were analysed. Senior and Junior physicians independently interpreted video-recorded Iriscope sequences, classifying them as tumoral (malignant) or non-tumoral, blind to the final diagnosis. A deep learning (DL) model was also trained on Iriscope images and tested on a different set of patients for comparison with human interpretation. Diagnostic accuracy, sensitivity, specificity, and F1 score were calculated.

RESULTS: Sixty-one patients with small PPNs (median size 15 mm, IQR: 11-20 mm) were included. The technique allowed for the direct visualisation of the lesions in all cases. The final diagnosis was cancer for 37 cases and a benign lesion in 24 cases. Senior physicians outperformed junior physicians in recognising tumoral Iriscope images, with a balanced accuracy of 85.4% versus 66.7%, respectively, when compared with the final diagnosis. The DL model outperformed junior physicians with a balanced accuracy of 71.5% but was not superior to senior physicians.

CONCLUSION: Iriscope could be a valuable tool in PPNs management, especially for experienced operators. Applied to Iriscope images, DL could enhance overall performance of less experienced physicians in diagnosing malignancy.

PMID:40433758 | DOI:10.1111/resp.70057

Front Neurol. 2025 May 13;16:1536751. doi: 10.3389/fneur.2025.1536751. eCollection 2025.

ABSTRACT

BACKGROUND: Accurate preoperative grading of meningiomas is crucial for selecting the most suitable treatment strategies and predicting patient outcomes. Traditional MRI-based assessments are often insufficient to distinguish between low- and high-grade meningiomas reliably. Deep learning (DL) models have emerged as promising tools for automated histopathological grading using imaging data. This systematic review and meta-analysis aimed to comprehensively evaluate the diagnostic performance of deep learning (DL) models for meningioma grading.

METHODS: This study was conducted in accordance with the PRISMA-DTA guidelines and was prospectively registered on the Open Science Framework. A systematic search of PubMed, Scopus, and Web of Science was performed up to March 2025. Studies using DL models to classify meningiomas based on imaging data were included. A random-effects meta-analysis was used to pool sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUC). A bivariate random-effects model was used to fit the summary receiver operating characteristic (SROC) curve. Study quality was assessed using the Newcastle-Ottawa Scale, and publication bias was evaluated using Egger's test.

RESULTS: Twenty-seven studies involving 13,130 patients were included. The pooled sensitivity was 92.31% (95% CI: 92.1-92.52%), specificity 95.3% (95% CI: 95.11-95.48%), and accuracy 97.97% (95% CI: 97.35-97.98%), with an AUC of 0.97 (95% CI: 0.96-0.98). The bivariate SROC curve demonstrated excellent diagnostic performance, characterized by a relatively narrow 95% confidence interval despite moderate to high heterogeneity (I2 = 79.7%, p < 0.001).

CONCLUSION: DL models demonstrate high diagnostic accuracy for automatic meningioma grading and could serve as valuable clinical decision-support tools.

SYSTEMATIC REVIEW REGISTRATION: DOI: 10.17605/OSF.IO/RXEBM.

PMID:40433621 | PMC:PMC12108801 | DOI:10.3389/fneur.2025.1536751

Front Artif Intell. 2025 May 13;8:1583459. doi: 10.3389/frai.2025.1583459. eCollection 2025.

ABSTRACT

This review provides a thorough and organized overview of machine learning (ML) applications in predicting heart disease, covering technological advancements, challenges, and future prospects. As cardiovascular diseases (CVDs) are the leading cause of global mortality, there is an urgent demand for early and precise diagnostic tools. ML models hold considerable potential by utilizing large-scale healthcare data to enhance predictive diagnostics. To systematically investigate this field, the literature is organized into five thematic categories such as "Heart Disease Detection and Diagnostics," "Machine Learning Models and Algorithms for Healthcare," "Feature Engineering and Optimization Techniques," "Emerging Technologies in Healthcare," and "Applications of AI Across Diseases and Conditions." The review incorporates performance benchmarking of various ML models, highlighting that hybrid deep learning (DL) frameworks, e.g., convolutional neural network-long short-term memory (CNN-LSTM) consistently outperform traditional models in terms of sensitivity, specificity, and area under the curve (AUC). Several real-world case studies are presented to demonstrate the successful deployment of ML models in clinical and wearable settings. This review showcases the progression of ML approaches from traditional classifiers to hybrid DL structures and federated learning (FL) frameworks. It also discusses ethical issues, dataset limitations, and model transparency. The conclusions provide important insights for the development of artificial intelligence (AI) powered, clinically applicable heart disease prediction systems.

PMID:40433606 | PMC:PMC12106346 | DOI:10.3389/frai.2025.1583459

Front Artif Intell. 2025 May 13;8:1561281. doi: 10.3389/frai.2025.1561281. eCollection 2025.

ABSTRACT

High-resolution digital elevation models (HRDEMs) from LiDAR and InSAR technologies have significantly improved the accuracies of mapping hydrographic features such as river boundaries, streamlines, and waterbodies over large areas. However, drainage crossings that facilitate the passage of drainage flows beneath roads are not often represented in HRDEMs, resulting in erratic or distorted hydrographic features. At present, drainage crossing datasets are largely missing or available with variable quality. While previous studies have investigated basic convolutional neural network (CNN) models for drainage crossing characterization, it remains unclear if advanced deep learning models will improve the accuracy of drainage crossing classification. Although HRDEM-derived geomorphological features have been identified to enhance feature extraction in other hydrography applications, the contributions of these features to drainage crossing image classification have yet to be sufficiently investigated. This study develops advanced CNN models, EfficientNetV2, using four co-registered 1-meter resolution geomorphological data layers derived from HRDEMs for drainage crossing classification. These layers include positive openness (POS), geometric curvature, and two topographic position index (TPI) layers utilizing 3 × 3 and 21 × 21 cell windows. The findings reveal that the advanced CNN models with HRDEM, TPI (21 × 21), and a combination of HRDEM, POS, and TPI (21 × 21) improve classification accuracy in comparison to the baseline model by 3.39, 4.27, and 4.93%, respectively. The study culminates in explainable artificial intelligence (XAI) for evaluating those most critical image segments responsible for characterizing drainage crossings.

PMID:40433605 | PMC:PMC12106317 | DOI:10.3389/frai.2025.1561281

Uncertain Safe Util Mach Learn Med Imaging (2024). 2025;15167:3-13. doi: 10.1007/978-3-031-73158-7_1. Epub 2024 Oct 3.

ABSTRACT

Supervised learning has become the dominant paradigm in computer-aided diagnosis. Generally, these methods assume that the training labels represent "ground truth" information about the target phenomena. In actuality, the labels, often derived from human annotations, are noisy/unreliable. This aleoteric uncertainty poses significant challenges for modalities such as electroencephalography (EEG), in which "ground truth" is difficult to ascertain without invasive experiments. In this paper, we propose a novel Bayesian framework to mitigate the effects of aleoteric label uncertainty in the context of supervised deep learning. Our target application is EEG-based epileptic seizure detection. Our framework, called BUNDL, leverages domain knowledge to design a posterior distribution for the (unknown) "clean labels" that automatically adjusts based on the data uncertainty. Crucially, BUNDL can be wrapped around any existing detection model and trained using a novel KL divergence-based loss function. We validate BUNDL on both a simulated EEG dataset and the Temple University Hospital (TUH) corpus using three state-of-the-art deep networks. In all cases, BUNDL improves seizure detection performance over existing noise mitigation strategies.

PMID:40433566 | PMC:PMC12107695 | DOI:10.1007/978-3-031-73158-7_1

Front Neurorobot. 2025 May 13;19:1568811. doi: 10.3389/fnbot.2025.1568811. eCollection 2025.

ABSTRACT

The number of music of different genres is increasing year by year, and manual classification is costly and requires professionals in the field of music to manually design features, some of which lack the generality of music genre classification. Deep learning has had a large number of scientific research results in the field of music classification, but the existing deep learning methods still have the problems of insufficient extraction of music feature information, low accuracy rate of music genres, loss of time series information, and slow training. To address the problem that different music durations affect the accuracy of music genre classification, we form a Log Mel spectrum with music audio data of different cut durations. After discarding incomplete audio, we design data enhancement with different slicing durations and verify its effect on accuracy and training time through comparison experiments. Based on this, the audio signal is divided into frames, windowed and short-time Fourier transformed, and then the Log Mel spectrum is obtained by using the Mel filter and logarithmic compression. Aiming at the problems of loss of time information, insufficient feature extraction, and low classification accuracy in music genre classification, firstly, we propose a Res-Transformer model that fuses the residual network with the Transformer coding layer. The model consists of two branches, the left branch is an improved residual network, which enhances the spectral feature extraction ability and network expression ability and realizes the dimensionality reduction; the right branch uses four Transformer coding layers to extract the time-series information of the Log Mel spectrum. The output vectors of the two branches are spliced and input into the classifier to realize music genre classification. Then, to further improve the classification accuracy of the model, we propose the TS-Resformer model based on the Res-Transformer model, combined with different attention mechanisms, and design the time-frequency attention mechanism, which employs different scales of filters to fully extract the low-level music features from the two dimensions of time and frequency as the input to the time-frequency attention mechanism, respectively. Finally, experiments show that the accuracy of this method is 90.23% on the FMA-small dataset, which is an improvement in classification accuracy compared with the classical model.

PMID:40433555 | PMC:PMC12106318 | DOI:10.3389/fnbot.2025.1568811

Front Plant Sci. 2025 May 13;16:1550302. doi: 10.3389/fpls.2025.1550302. eCollection 2025.

ABSTRACT

Global climate change and ecological degradation highlight the urgency of dealing with agricultural waste and ecological restoration. Traditional pollutant monitoring and ecological restoration methods face challenges in accuracy and adaptability, especially when dealing with complex environmental data. This paper proposes the Bio-DANN model, which combines biogeochemical models and deep learning techniques to improve the accuracy of pollutant monitoring and ecological restoration prediction. The model uses deep neural networks (DNNs) and attention mechanisms to process multidimensional environmental data in various agricultural and ecological scenarios in real time. Experimental results based on Open Soil Data and NEON datasets show that Bio-DANN performs well in pollutant prediction, with mean square errors (MSE) of 0.012 and 0.018, root mean square errors (RMSE) of 0.109 and 0.134, and accuracy of 0.92 and 0.90, respectively. In terms of ecological restoration assessment, Bio-DANN achieved ΔF and PIPGR of 0.15 and 18%, and 0.20 and 22%, respectively, and H' values of 1.5 and 1.7, which are better than other models. Bio-DANN provides a promising technical solution for environmental protection, resource recovery and sustainable agriculture, especially showing significant potential in pollutant monitoring, soil health assessment and ecological restoration evaluation.

PMID:40433163 | PMC:PMC12106413 | DOI:10.3389/fpls.2025.1550302

Front Plant Sci. 2025 May 13;16:1543986. doi: 10.3389/fpls.2025.1543986. eCollection 2025.

ABSTRACT

In order to enhance the accuracy of rice leaf disease detection in complex farmland environments, and facilitate the deployment of the deep learning model onto mobile terminals for rapid real-time inference, this paper introduces a disease detection network titled YOLOv11 Multi-scale Dynamic Feature Fusion for Rice Disease Detection (YOLOv11-MSDFF-RiceD). The model adopts the concept of ParameterNet to design the FlexiC3k2Net module, which replaces the neck feature extraction network, thereby bolstering the model's feature learning capabilities without significantly increasing computational complexity. Additionally, an efficient multi-scale feature fusion module (EMFFM) is devised, improving both the computational efficiency and feature extraction capabilities of the model, while simultaneously reducing the number of parameters and memory footprint. The bounding box regression loss function, inner-WIoU, utilizes auxiliary bounding boxes and scale factors. Finally, the Dependency Graph (DepGraph) pruning model is employed to minimize the model's size, computational load, and parameter count, with only a moderate sacrifice in accuracy. Compared to the original YOLOv11n model, the optimized model achieves reductions in computational complexity, parameter scale, and memory usage by 50.7%, 49.6%, and 36.9%, respectively, with only a 1.7% improvement in mAP@0.5:0.9. These optimizations enable efficient deployment on resource-constrained mobile devices, making the model highly suitable for real-time disease detection in practical agricultural scenarios where hardware limitations are critical. Consequently, the improved model proposed in this paper effectively detects rice disease targets in complex environments, providing theoretical and technical support for the deployment and application of mobile terminal detection devices, such as rice disease detectors, in practical scenarios.

PMID:40433155 | PMC:PMC12106424 | DOI:10.3389/fpls.2025.1543986

Front Med (Lausanne). 2025 May 13;12:1587417. doi: 10.3389/fmed.2025.1587417. eCollection 2025.

ABSTRACT

OBJECTIVE: In non-clinical safety evaluation of drugs, pathological result is one of the gold standards for determining toxic effects. However, pathological diagnosis might be challenging and affected by pathologist expertise. In carcinogenicity studies, drug-induced squamous cell carcinoma (SCC) of the mouse stomach represents a diagnostic challenge for toxicopathologists. This study aims to establish a detection model for mouse gastric squamous cell carcinoma (GSCC) using deep learning algorithms, to improve the accuracy and consistency of pathological diagnoses.

METHODS: A total of 93 cases of drug-induced mouse GSCC and 56 cases of normal mouse stomach tissue from carcinogenicity studies were collected. After scanning into digital slides, semi-automated data annotation was performed. All images underwent preprocessing, including tissue extraction, artifact removal, and exclusion of normal epithelial regions. The images were then randomly divided into training, validation, and test sets in an 8:1:1 ratio. Five different convolutional neural networks (CNNs)-FCN, LR-ASPP, DeepLabv3+, U-Net, and DenseNet were applied to identify GSCC and non-GSCC regions. Tumor prediction images (algorithm results shown as overlays) derived from the slide images were compared, and the performance of the constructed models was evaluated using Precision, Recall, and F1-score.

RESULTS: The Precision, Recall, and F1-scores of DenseNet, U-Net, and DeepLabv3 + algorithms were all above 90%. Specifically, the DenseNet model achieved an overall Precision of 0.9044, Recall of 0.9291, and F1-score of 0.9157 in the test set. Compared to the other algorithms, DenseNet exhibited the highest F1-score and Recall, demonstrating superior generalization ability.

CONCLUSION: The DenseNet algorithm model developed in this study shown promising application potential for assisting in the diagnosis of mouse GSCC. As artificial intelligence (AI) technology continues to advance in non-clinical safety evaluation of drugs, CNN-based toxicological pathology detection models will become essential tools to assist pathologists in precise diagnosis and consistency evaluation.

PMID:40432719 | PMC:PMC12106445 | DOI:10.3389/fmed.2025.1587417