Front Dent Med. 2025 Jun 26;6:1618246. doi: 10.3389/fdmed.2025.1618246. eCollection 2025.

ABSTRACT

OBJECT: In forensic dentistry, dental age estimation assists experts in determining the age of victims or suspects, which is vital for legal responsibility and sentencing. The traditional Demirjian method assesses the development of seven mandibular teeth in pediatric dentistry, but it is time-consuming and relies heavily on subjective judgment.

METHODS: This study constructed a largescale panoramic dental image dataset and applied various convolutional neural network (CNN) models for automated age estimation.

RESULTS: Model performance was evaluated using loss curves, residual histograms, and normal PP plots. Age prediction models were built separately for the total, female, and male samples. The best models yielded mean absolute errors of 1.24, 1.28, and 1.15 years, respectively.

DISCUSSION: These findings confirm the effectiveness of deep learning models in dental age estimation, particularly among northern Chinese adolescents.

PMID:40642202 | PMC:PMC12241049 | DOI:10.3389/fdmed.2025.1618246

Front Dent Med. 2025 Jun 26;6:1583455. doi: 10.3389/fdmed.2025.1583455. eCollection 2025.

ABSTRACT

INTRODUCTION: Accurate assessment of midpalatal suture (MPS) maturation is critical in orthodontics, particularly for planning treatment strategies in patients with maxillary transverse deficiency (MTD). Although cone-beam computed tomography (CBCT) provides detailed imaging suitable for MPS classification, manual interpretation is often subjective and time-consuming.

METHODS: This study aimed to develop and evaluate a lightweight two-dimensional convolutional neural network (2D CNN) for the automated classification of MPS maturation stages using axial CBCT slices. A retrospective dataset of CBCT images from 111 patients was annotated based on Angelieri's classification system and grouped into three clinically relevant categories: AB (Stages A and B), C, and DE (Stages D and E). A 9-layer CNN architecture was trained and evaluated using standard classification metrics and receiver operating characteristic (ROC) curve analysis.

RESULTS: The model achieved a test accuracy of 96.49%. Class-wise F1-scores were 0.95 for category AB, 1.00 for C, and 0.95 for DE. Area under the ROC curve (AUC) scores were 0.10 for AB, 0.62 for C, and 0.98 for DE. Lower AUC values in the early and transitional stages (AB and C) likely reflect known anatomical overlap and subjectivity in expert labeling.

DISCUSSION: These findings indicate that the proposed 2D CNN demonstrates high accuracy and robustness in classifying MPS maturation stages from CBCT images. Its compact architecture and strong performance suggest it is suitable for real-time clinical decision-making, particularly in identifying cases that may benefit from surgical intervention. Moreover, its lightweight design makes it adaptable for use in resource-limited settings. Future work will explore volumetric models to further enhance diagnostic reliability and confidence.

PMID:40642201 | PMC:PMC12241142 | DOI:10.3389/fdmed.2025.1583455

Phys Imaging Radiat Oncol. 2025 Jun 19;35:100793. doi: 10.1016/j.phro.2025.100793. eCollection 2025 Jul.

ABSTRACT

BACKGROUND AND PURPOSE: Radiotherapy targets DNA in cancer cell nuclei. Radiation dose, however, is prescribed to a macroscopic target volume assuming uniform distribution, failing to consider microscopic variations in dose absorbed by individual nuclei. This study investigated a potential link between pre-treatment tumour nuclear size distributions and post-radiotherapy outcomes in gynecological squamous cell carcinoma (SCC).

MATERIALS AND METHODS: Our multi-institutional cohort consisted of 191 non-metastatic gynecological SCC patients who had received radiotherapy with diagnostic whole slide images (WSIs) available. Tumour nuclear size distribution mean and standard deviation were extracted from WSIs using deep learning, and used to predict progression-free interval (PFI) and overall survival (OS) in multivariate Cox proportional hazards (CoxPH) analysis adjusted for age and clinical stage.

RESULTS: Multivariate CoxPH analysis revealed that a larger nuclear size distribution mean results in more favorable outcomes for PFI (HR = 0.45, 95% CI: 0.19 - 1.09, p = 0.084) and OS (HR = 0.55, 95% CI: 0.24 - 1.25, p = 0.16), and that a larger nuclear size standard deviation results in less favorable outcomes for PFI (HR = 7.52, 95% CI: 1.43 - 39.52, p = 0.023) and OS (HR = 4.67, 95% CI: 0.96 - 22.57, p = 0.063). The bootstrap-validated C-statistic was 0.56 for PFI and 0.57 for OS.

CONCLUSION: Despite low accuracy, tumour nuclear size heterogeneity aided prognostication over standard clinical variables and was associated with outcomes following radiotherapy in gynecological SCC. This highlights the potential importance of personalized multiscale dosimetry and warrants further large-scale pan-cancer studies.

PMID:40642183 | PMC:PMC12242011 | DOI:10.1016/j.phro.2025.100793

Afr J Lab Med. 2024 Nov 21;13(1):2516. doi: 10.4102/ajlm.v13i1.2516. eCollection 2024.

ABSTRACT

BACKGROUND: Artificial intelligence (AI) and digital technology, as advanced human-created tools, are influencing the healthcare sector.

AIM: This review provides a comprehensive and structured exploration of the opportunities presented by AI and digital technology to laboratory diagnostics and management of communicable and non-communicable diseases in Africa.

METHODS: The study employed the Preferred Reporting Items for Systematic Reviews, Meta-Analyses guidelines and Bibliometric analysis as its methodological approach. Peer-reviewed publications from 2000 to 2024 were retrieved from PubMed®, Web of Science™ and Google Scholar databases.

RESULTS: The study incorporated a total of 1563 peer-reviewed scientific documents and, after filtration, 37 were utilised for systematic review. The findings revealed that AI and digital technology play a key role in patient management, quality assurance and laboratory operations, including healthcare decision-making, disease monitoring and prognosis. Metadata reflected the disproportionate research outputs distribution across Africa. In relation to non-communicable diseases, Egypt, South Africa, and Morocco lead in cardiovascular, diabetes and cancer research. Representing communicable diseases research, Algeria, Egypt, and South Africa were prominent in HIV/AIDS research. South Africa, Nigeria, Ghana, and Egypt lead in malaria and tuberculosis research.

CONCLUSION: Facilitation of widespread adoption of AI and digital technology in laboratory diagnostics across Africa is critical for maximising patient benefits. It is recommended that governments in Africa allocate more funding for infrastructure and research on AI to serve as a catalyst for innovation.

WHAT THIS STUDY ADDS: This review provides a comprehensive and context-specific analysis of AI's application in African healthcare.

PMID:40642055 | PMC:PMC12242046 | DOI:10.4102/ajlm.v13i1.2516

Front Med (Lausanne). 2025 Jun 26;12:1589707. doi: 10.3389/fmed.2025.1589707. eCollection 2025.

ABSTRACT

Accurate segmentation of the prostate in T2-weighted MRI is critical for effective prostate diagnosis and treatment planning. Existing methods often struggle with the complex textures and subtle variations in the prostate. To address these challenges, we propose RaNet (Residual Attention Network), a novel framework based on ResNet50, incorporating three key modules: the DilatedContextNet (DCNet) encoder, the Multi-Scale Attention Fusion (MSAF), and the Feature Fusion Module (FFM). The encoder leverages residual connections to extract hierarchical features, capturing both fine-grained details and multi-scale patterns in the prostate. The MSAF enhances segmentation by dynamically focusing on key regions, refining feature selection and minimizing errors, while the FFM optimizes the handling of spatial hierarchies and varying object sizes, improving boundary delineation. The decoder mirrors the encoder's structure, using deconvolutional layers and skip connections to retain essential spatial details. We evaluated RaNet on a prostate MRI dataset PROMISE12 and ProstateX , achieving a DSC of 98.61 and 96.57 respectively. RaNet also demonstrated robustness to imaging artifacts and MRI protocol variability, confirming its applicability across diverse clinical scenarios. With a balance of segmentation accuracy and computational efficiency, RaNet is well suited for real-time clinical use, offering a powerful tool for precise delineation and enhanced prostate diagnostics.

PMID:40641983 | PMC:PMC12241084 | DOI:10.3389/fmed.2025.1589707

Sci Rep. 2025 Jul 10;15(1):24888. doi: 10.1038/s41598-025-08226-5.

ABSTRACT

Bug detection and resolution are pivotal to maintaining the quality, reliability, and performance of software systems. Manual debugging, along with traditional static rule-based methods, proves inefficient when applied to complex software structures in contemporary times. SynergyBug combines BERT and GPT-3 to autonomously detect and repair bugs across multiple sources. It resolves essential requirements by implementing an automated system that diagnoses and resolves software bugs automatically, thus minimising human involvement. The framework unites BERT as a contextual machinery with GPT-3 to produce bug fix generation capabilities. The semantic pattern within bug reports, together with error logs and documentation, feeds into BERT for contextual embedding generation. GPT-3 applies the generated embeddings to produce code fixes, code snippets, as well as detailed explanations that address detected problems. The system achieves continuous automatic debugging by enhancing both detection and resolution steps into one unified process. The experimental outcomes prove that it achieves superior performance than conventional bug detection methods by reaching 98.79% accuracy alongside 97.23% precision and 96.56% recall. The system demonstrated exceptional detection strength for functional and performance, and security bugs, where the detection rates reached 94% and 90% and 92%, respectively. SynergyBug showed its ability to expand as it processed bug reports exceeding 100,000 cases without noticeably impacting system performance. This proposed system provides faster debugging capabilities to improve the quality of the complete software development process. This paper discusses as a tool that can revolutionise bug management through proactive instead of just reactive strategies. The implementation of human monitoring within safety programs and managing training system biases represent essential organisational factors. The study terminates by recognising SynergyBug as a crucial development leading toward automated debugging tools that maintain operational safety within intricate software systems.

PMID:40640256 | DOI:10.1038/s41598-025-08226-5

Sci Rep. 2025 Jul 10;15(1):24840. doi: 10.1038/s41598-025-09120-w.

ABSTRACT

Super-resolution imaging has emerged as a rapidly advancing field in diagnostic ultrasound. Ultrasound Localization Microscopy (ULM) achieves sub-wavelength precision in microvasculature imaging by tracking gas microbubbles (MBs) flowing through blood vessels. However, MB localization faces challenges due to dynamic point spread functions (PSFs) caused by harmonic and sub-harmonic emissions, as well as depth-dependent PSF variations in ultrasound imaging. Additionally, deep learning models often struggle to generalize from simulated to in vivo data due to significant disparities between the two domains. To address these issues, we propose a novel approach using the DEformable DEtection TRansformer (DE-DETR). This object detection network tackles object deformations by utilizing multi-scale feature maps and incorporating a deformable attention module. We further refine the super-resolution map by employing a KDTree algorithm for efficient MB tracking across consecutive frames. We evaluated our method using both simulated and in vivo data, demonstrating improved precision and recall compared to current state-of-the-art methodologies. These results highlight the potential of our approach to enhance ULM performance in clinical applications.

PMID:40640235 | DOI:10.1038/s41598-025-09120-w

Sci Rep. 2025 Jul 10;15(1):24830. doi: 10.1038/s41598-025-05958-2.

ABSTRACT

Predicting time-series data is inherently complex, spurring the development of advanced neural network approaches. Monitoring and predicting PM2.5 levels is especially challenging due to the interplay of diverse natural and anthropogenic factors influencing its dispersion, making accurate predictions both costly and intricate. A key challenge in predicting PM2.5 concentrations lies in its variability, as the data distribution fluctuates significantly over time. Meanwhile, neural networks provide a cost-effective and highly accurate solution in managing such complexities. Deep learning models like Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM) have been widely applied to PM2.5 prediction tasks. However, prediction errors increase as the forecasting window expands from 1 to 72 hours, underscoring the rising uncertainty in longer-term predictions. Recurrent Neural Networks (RNNs) with continuous-time hidden states are well-suited for modeling irregularly sampled time series but struggle with long-term dependencies due to gradient vanishing or exploding, as revealed by the ordinary differential equation (ODE) based hidden state dynamics-regardless of the ODE solver used. Continuous-time neural processes, defined by differential equations, are limited by numerical solvers, restricting scalability and hindering the modeling of complex phenomena like neural dynamics-ideally addressed via closed-form solutions. In contrast to ODE-based continuous models, closed-form networks demonstrate superior scalability over traditional deep-learning approaches. As continuous-time neural networks, Neural ODEs excel in modeling the intricate dynamics of time-series data, presenting a robust alternative to traditional LSTM models. We propose two ODE-based models: a transformer-based ODE model and a closed-form ODE model. Empirical evaluations show these models significantly enhance prediction accuracy, with improvements ranging from 2.91 to 14.15% for 1-hour to 8-hour predictions when compared to LSTM-based models. Moreover, after conducting the paired t-test, the RMSE values of the proposed model (CCCFC) were found to be significantly different from those of BILSTM, LSTM, GRU, ODE-LSTM, and PCNN,CNN-LSSTM. This implies that CCCFC demonstrates a distinct performance advantage, reinforcing its effectiveness in hourly PM2.5 forecasting.

PMID:40640232 | DOI:10.1038/s41598-025-05958-2

Sci Rep. 2025 Jul 10;15(1):24940. doi: 10.1038/s41598-025-08249-y.

ABSTRACT

Gastric Tract Disease (GTD) constitutes a medical emergency, emphasizing the critical importance of early diagnosis and intervention to lessen its severity. Clinical practices often utilize endoscopy-supported examinations for GTD screening. The images obtained during this procedure are examined to identify the presence of the disease and investigate its severity. Autoimmune Gastritis (AIG) is a chronic inflammatory GTD and timely detection and treatment is crucial to reduce its harshness. This research aims to develop a deep-learning (DL) tool to detect the AIG from clinical-grade endoscopic images. Various stages in the DL tool comprise; (i) Image collection and resizing, (ii) image pre-processing using Entropy-function and Moth-Flame (MF) Algorithm, (iii) deep-features extraction using a chosen DL-model, (iv) feature optimization using MF algorithm and serial features concatenation, and (iv) classification and performance confirmation using five-fold cross-validation. This study aims to develop a DL tool to assist clinicians during the AIG examination and hence better detection accuracy is preferred. The merit of the DL model is demonstrated in the individual deep-features and serially concatenated-features and the experimental outcome of this study provides a detection accuracy of 99.33% when the detection is performed with fused-features and K-Nearest Neighbor classifier. This authenticates that this tool offers a clinically important outcome on the endoscopy database.

PMID:40640222 | DOI:10.1038/s41598-025-08249-y

Sci Data. 2025 Jul 10;12(1):1184. doi: 10.1038/s41597-025-05346-5.

ABSTRACT

Multiple Sclerosis (MS) is a chronic autoimmune disease that primarily affects the central nervous system and is predominantly diagnosed in adults, making pediatric cases rare and underrepresented in medical research. This paper introduces the first publicly available MRI dataset specifically dedicated to pediatric multiple sclerosis lesion segmentation. The dataset comprises longitudinal MRI scans from 9 pediatric patients, each with between one and six timepoints, with a total of 28 MRI scans. It includes T1-weighted (MPRAGE), T2-weighted, and FLAIR sequences. Additionally, it provides clinical data and initial symptoms for each patient, offering valuable insights into disease progression. Lesion segmentation was performed by senior experts, ensuring high-quality annotations. To demonstrate the dataset's reliability and utility, we evaluated two deep learning models, achieving competitive segmentation performance. This dataset aims to advance research in pediatric MS, improve lesion segmentation models, and contribute to federated learning approaches.

PMID:40640191 | DOI:10.1038/s41597-025-05346-5

BMC Pulm Med. 2025 Jul 10;25(1):336. doi: 10.1186/s12890-025-03805-8.

ABSTRACT

OBJECTIVES: To investigate the value of multi-model based on preoperative CT scans in predicting EGFR/TP53 co-mutation status.

METHODS: We retrospectively included 2171 patients with non-small cell lung cancer (NSCLC) with pre-treatment computed tomography (CT) scans and predicting epidermal growth factor receptor (EGFR) gene sequencing from West China Hospital between January 2013 and April 2024. The deep-learning model was built for predicting EGFR / tumor protein 53 (TP53) co-occurrence status. The model performance was evaluated by area under the curve (AUC) and Kaplan-Meier analysis. We further compared multi-dimension model with three one-dimension models separately, and we explored the value of combining clinical factors with machine-learning factors. Additionally, we investigated 546 patients with 56-panel next-generation sequencing and low-dose computed tomography (LDCT) to explore the biological mechanisms of radiomics.

RESULTS: In our cohort of 2171 patients (1,153 males, 1,018 females; median age 60 years), single-dimensional models were developed using data from 1,055 eligible patients. The multi-dimensional model utilizing a Random Forest classifier achieved superior performance, yielding the highest AUC of 0.843 for predicting EGFR/TP53 co-mutations in the test set.

CONCLUSION: The multi-dimensional model demonstrates promising potential for non-invasive prediction of EGFR and TP53 co-mutations, facilitating early and informed clinical decision-making in NSCLC patients at risk of treatment resistance.

PMID:40640822 | DOI:10.1186/s12890-025-03805-8

J Transl Med. 2025 Jul 10;23(1):774. doi: 10.1186/s12967-025-06617-w.

ABSTRACT

BACKGROUND: Accurate prediction of pathological complete response (pCR) to neoadjuvant chemotherapy has significant clinical utility in the management of breast cancer treatment. Although multimodal deep learning models have shown promise for predicting pCR from medical imaging and other clinical data, their adoption has been limited due to challenges with interpretability and generalizability across institutions.

METHODS: We developed a multimodal deep learning model combining post contrast-enhanced whole-breast MRI at pre- and post-treatment timepoints with non-imaging clinical features. The model integrates 3D convolutional neural networks and self-attention to capture spatial and cross-modal interactions. We utilized two public multi-institutional datasets to perform internal and external validation of the model. For model training and validation, we used data from the I-SPY 2 trial (N = 660). For external validation, we used the I-SPY 1 dataset (N = 114).

RESULTS: Of the 660 patients in I-SPY 2, 217 patients achieved pCR (32.88%). Of the 114 patients in I-SPY 1, 29 achieved pCR (25.44%). The attention-based multimodal model yielded the best predictive performance with an AUC of 0.73 ± 0.04 on the internal data and an AUC of 0.71 ± 0.02 on the external dataset. The MRI-only model (internal AUC = 0.68 ± 0.03, external AUC = 0.70 ± 0.04) and the non-MRI clinical features-only model (internal AUC = 0.66 ± 0.08, external AUC = 0.71 ± 0.03) trailed in performance, indicating the combination of both modalities is most effective.

CONCLUSION: We present a robust and interpretable deep learning framework for pCR prediction in breast cancer patients undergoing NAC. By combining imaging and clinical data with attention-based fusion, the model achieves strong predictive performance and generalizes across institutions.

PMID:40640789 | DOI:10.1186/s12967-025-06617-w

BMC Bioinformatics. 2025 Jul 10;26(1):171. doi: 10.1186/s12859-025-06213-1.

ABSTRACT

BACKGROUND: Predicting protein structural similarity and detecting homologous sequences remain fundamental and challenging tasks in computational biology. Accurate identification of structural homologs enables function inference for newly discovered or unannotated proteins. Traditional approaches often require full 3D structural data, which is unavailable for most proteins. Thus, there is a need for sequence-based methods capable of inferring structural similarity efficiently and at scale.

RESULT: We present Rprot-Vec (Rapid Protein Vector), a deep learning model that predicts protein structural similarity and performs homology detection using only primary sequence data. The model integrates bidirectional GRU and multi-scale CNN layers with ProtT5-based encoding, enabling accurate and fast similarity estimation. Rprot-Vec achieves a 65.3% accurate similarity prediction rate in the homologous region (TM-score > 0.8), with an average prediction error of 0.0561 across all TM-score intervals. Despite having only 41% of the parameters of TM-vec, Rprot-Vec outperforms both public and locally trained TM-vec baselines in all tested settings. Additionally, we constructed and released three curated training datasets (CATH_TM_score_S/M/L), supporting further research in this area.

CONCLUSION: Rprot-Vec offers a fast and lightweight solution for sequence-based structural similarity prediction. It can be applied in protein homology detection, structure-function inference, drug repurposing, and other downstream biological tasks. Its open-source availability and released datasets facilitate broader adoption and further development by the research community.

PMID:40640710 | DOI:10.1186/s12859-025-06213-1

BMC Med Imaging. 2025 Jul 10;25(1):278. doi: 10.1186/s12880-025-01805-y.

ABSTRACT

BACKGROUND: Non-alcoholic fatty liver disease (NAFLD) is a significant risk factor for liver cancer and cardiovascular diseases, imposing substantial social and economic burdens. Computed tomography (CT) scans are crucial for diagnosing NAFLD and assessing its severity. However, current manual measurement techniques require considerable human effort and resources from radiologists, and there is a lack of standardized methods for classifying the severity of NAFLD in existing research.

METHODS: To address these challenges, we propose a novel method for NAFLD segmentation and automated severity scoring. The method consists of three key modules: (1) The Semi-automatization nnU-Net Module (SNM) constructs a high-quality dataset by combining manual annotations with semi-automated refinement; (2) The Focal Feature Fusion Swin-Unet Module (FSM) enhances liver and spleen segmentation through multi-scale feature fusion and Swin Transformer-based architectures; (3) The Automated Severity Scoring Module (ASSM) integrates segmentation results with radiological features to classify NAFLD severity. These modules are embedded in a Flask-RESTful API-based system, enabling users to upload abdominal CT data for automated preprocessing, segmentation, and scoring.

RESULTS: The Focal Feature Fusion Swin-Unet (FF Swin-Unet) method significantly improves segmentation accuracy, achieving a Dice similarity coefficient (DSC) of 95.64% and a 95th percentile Hausdorff distance (HD95) of 15.94. The accuracy of the automated severity scoring is 90%. With model compression and ONNX deployment, the evaluation speed for each case is approximately 5 seconds. Compared to manual diagnosis, the system can process a large volume of data simultaneously, rapidly, and efficiently while maintaining the same level of diagnostic accuracy, significantly reducing the workload of medical professionals.

CONCLUSIONS: Our research demonstrates that the proposed system has high accuracy in processing large volumes of CT data and providing automated NAFLD severity scores quickly and efficiently. This method has the potential to significantly reduce the workload of medical professionals and holds immense clinical application potential.

PMID:40640704 | DOI:10.1186/s12880-025-01805-y

Surg Endosc. 2025 Jul 10. doi: 10.1007/s00464-025-11967-z. Online ahead of print.

ABSTRACT

BACKGROUND: Surgical proficiency influences surgical quality and patient outcomes in robot-assisted radical prostatectomy (RARP). Manual video evaluations are labor-intensive and lack standardized objective metrics. Herein, we aimed to develop an artificial intelligence (AI) deep-learning model that can identify the surgical phases in RARP videos and create a parameter-based scoring system to distinguish experts from novice surgeons based on the results of the AI model.

METHODS: A dataset of 410 RARP videos from 18 Japanese medical institutions was analyzed. The videos were annotated into 11 phases and divided into training and testing sets. Surgeons were categorized as experts or novices based on their RARP experience. We developed a deep-learning-based surgical phase classification model and compared the phase duration, number of transitions between phases, and AI confidence scores (AICS) between the groups based on the model's output. Key parameters were standardized and identified using stepwise multivariate logistic regression. A surgical skill scoring system was constructed based on the receiver operating characteristic curve cut-off values.

RESULTS: Of the 213 videos, 99 were used for training, 20 for validation, and 94 for testing (61 experts and 33 novices). The model achieved an accuracy of 0.89 in identifying surgical phases. The experts had significantly shorter durations in phases 2-8 and higher AICS than the novices. Stepwise analysis identified phases 2 (Retzius space expansion), 7 (dorsal venous complex incision, apex treatment, hemostasis), and 8 (urethrovesical anastomosis) and the AICS as key predictors of expertise. The scoring system developed from these variables effectively distinguished experts from novices with an accuracy of 86.2%.

CONCLUSIONS: The developed AI model revealed that the duration of several surgical phases and AICS are key parameters in assessing surgical skill proficiency in RARP. The new scoring system established based on these indicators reliably differentiates expert from novice surgeons.

PMID:40640619 | DOI:10.1007/s00464-025-11967-z

NPJ Digit Med. 2025 Jul 10;8(1):425. doi: 10.1038/s41746-025-01831-8.

ABSTRACT

With the shift toward de-escalating surgery in breast cancer, prediction models incorporating imaging can reassess the need for surgical axillary staging. This study employed advancements in deep learning to comprehensively evaluate routine mammograms for preoperative lymph node metastasis prediction. Mammograms and clinicopathological data from 1265 cN0 T1-T2 breast cancer patients (primary surgery, no neoadjuvant therapy) were retrospectively collected from three Swedish institutions. Compared to models using only clinical variables, incorporating full-breast mammograms with preoperative clinical variables improved the ROC AUC from 0.690 to 0.774 (improvement: 0.001-0.154) in the independent test set. The combined model showed good calibration and, at sensitivity ≥90%, achieved a significantly better net benefit, and a sentinel lymph node biopsy reduction rate of 41.7% (13.0-62.6%). Our findings suggest that routine mammograms, particularly full-breast images, can enhance preoperative nodal status prediction. They may substitute key predictors such as pathological tumor size and multifocality, aiding patient stratification before surgery.

PMID:40640522 | DOI:10.1038/s41746-025-01831-8

Sci Rep. 2025 Jul 10;15(1):24925. doi: 10.1038/s41598-025-10497-x.

ABSTRACT

This paper presents a novel approach for personalized learning path generation by integrating deep knowledge tracing and cognitive load estimation within a unified framework. We propose a dual-stream neural network architecture that simultaneously models students' knowledge states and cognitive load levels to optimize learning trajectories. The knowledge state tracking module employs a bidirectional Transformer with graph attention mechanisms to capture complex relationships between knowledge components, while the cognitive load estimation module utilizes multimodal data analysis to dynamically assess mental effort during learning activities. A dual-objective optimization algorithm balances knowledge acquisition with cognitive load management to generate paths that maintain optimal challenge levels. Experimental evaluations across multiple educational domains demonstrate that our approach outperforms existing methods in prediction accuracy (87.5%), path quality (4.4/5), and learning efficiency (24.6% improvement). The implemented system supports real-time adaptation based on performance and cognitive state, resulting in reduced frustration, higher engagement, and improved knowledge retention. This research contributes to both theoretical understanding of learning processes and practical implementation of next-generation adaptive educational technologies.

PMID:40640459 | DOI:10.1038/s41598-025-10497-x

Sci Rep. 2025 Jul 10;15(1):24942. doi: 10.1038/s41598-025-10265-x.

ABSTRACT

Lyme borreliosis and tick-borne encephalitis significantly impact public health in Europe, transmitted primarily by endemic tick species. The recent introduction of exotic tick species into northern Europe via migratory birds, imported animals, and travelers highlights the urgent need for rapid detection and accurate species identification. To address this, the Swedish Veterinary Agency launched a citizen science initiative, resulting in the submission of over 15,000 tick images spanning seven species. We developed, trained, and evaluated deep learning models incorporating image analysis, object detection, and transfer learning to support automated tick classification. The EfficientNetV2M model achieved a macro recall of 0.60 and a Matthews Correlation Coefficient (MCC) of 0.55 on out-of-distribution, citizen-submitted data. These results demonstrate the feasibility of integrating AI with citizen science for large-scale tick monitoring while also highlighting challenges related to class imbalance, species similarity, and morphological variability. Rather than robust species-level classification, our framework serves as a proof of concept for infrastructure that supports scalable and adaptive tick surveillance. This work lays the groundwork for future AI-driven systems in One Health contexts, extendable to other arthropod vectors and emerging public health threats.

PMID:40640390 | DOI:10.1038/s41598-025-10265-x

Sci Rep. 2025 Jul 10;15(1):24851. doi: 10.1038/s41598-025-09300-8.

ABSTRACT

Watershed macrotrash contamination is difficult to measure and requires tedious and labor-intensive processes. This work proposes an automated approach to waste counting, focusing on using computer vision, deep learning, and object tracking algorithms to acquire accurate counts of plastic bottles as they advect down rivers and streams. By using a combination of several publicly available labeled trash and plastic bottle image datasets, the model was trained to achieve high performance with the YOLOv8 object detection model. This was paired with the Norfair object tracking library and a novel post-processing algorithm to filter out false positives. The model performed extremely accurately over the test scenarios with just one false positive and recalls in excess of 0.947.

PMID:40640371 | DOI:10.1038/s41598-025-09300-8

Sci Rep. 2025 Jul 10;15(1):24961. doi: 10.1038/s41598-025-10092-0.

ABSTRACT

The rapid proliferation of Internet of Things (IoT) devices has introduced significant security challenges, with Distributed Denial of Service (DDoS) attacks posing a critical threat to network integrity. Traditional detection methods often rely on computationally intensive models, rendering them unsuitable for resource-constrained IoT environments. To address this limitation, this study proposes a lightweight and scalable machine learning-based DDoS detection framework specifically designed for IoT networks. Utilizing the NSL-KDD dataset, the framework employs an Extra Trees Classifier (ETC) for feature selection, reducing dimensionality while retaining critical attributes. Reduced features were selected to enhance performance and reduce processing cost. Three supervised learning models, Random Forest, Logistic Regression, and Naïve Bayes, were implemented and evaluated based on their detection accuracy, precision, recall, and F1-score. Experimental results demonstrate that the Random Forest model achieves exceptional accuracy (99.88%), precision (99.93%), recall (99.81%), and F1-score (99.87%), outperforming both Logistic Regression (91.61% accuracy) and Naïve Bayes (87.62% accuracy). Furthermore, the proposed framework significantly reduces computational overhead compared to deep learning-based approaches, making it highly suitable for IoT deployments. This research advances IoT security by providing a scalable, efficient, and accurate solution for detecting DDoS attacks, thereby bridging the gap between high-performance requirements and resource limitations in real-world IoT applications.

PMID:40640290 | DOI:10.1038/s41598-025-10092-0