Proc SPIE Int Soc Opt Eng. 2025 Feb;13406:134062B. doi: 10.1117/12.3047257. Epub 2025 Apr 11.

ABSTRACT

Radiomics models have been extensively used to predict clinical outcomes across various applications. However, their generalizability is often limited by undesirable feature values variability due to diverse imaging conditions. To address this issue, we previously developed a dual-domain deep learning approach to recover ground truth feature values in the presence of known blur and noise. The model consists of a differentiable approximation for radiomics calculation and a dual-domain loss function. We demonstrated model performance for gray-level co-occurrence matrix (GLCM) and histogram-based features. In this work, we extend the method to gray-level run length matrix (GLRLM) feature recovery. We introduce a novel algorithm for the differentiable approximation of GLRLMs. We assessed the performance of the GLRLM feature restoration network using lung CT image patches, with a focus on the accuracy of recovered feature values and classification performance between normal and COVID-positive lungs. The proposed network outperformed the baselines, achieving the lowest MSE in GLRLM feature recovery. Furthermore, a classification model based on the recovered GLRLM features reached an accuracy of 86.65%, closely aligning with the 88.85% accuracy of models using ground truth features, compared to 82.00% accuracy from degraded features. These results demonstrate the potential of our method as a robust tool for radiomics standardization.

PMID:40727416 | PMC:PMC12291091 | DOI:10.1117/12.3047257

R Soc Open Sci. 2025 Jul 16;12(7):242240. doi: 10.1098/rsos.242240. eCollection 2025 Jul.

ABSTRACT

Tipping points occur in many real-world systems, at which the system shifts suddenly from one state to another. The ability to predict the occurrence of tipping points from time series data remains an outstanding challenge and a major interest in a broad range of research fields. Particularly, the widely used methods based on bifurcation theory are neither reliable in prediction accuracy nor applicable for irregularly sampled time series which are commonly observed from real-world systems. Here, we address this challenge by developing a deep learning algorithm for predicting the occurrence of tipping points in untrained systems, by exploiting information about normal forms. Our algorithm not only outperforms traditional methods for regularly sampled model time series but also achieves accurate predictions for irregularly sampled model time series and empirical time series. Our ability to predict tipping points for complex systems paves the way for mitigation risks, prevention of catastrophic failures and restoration of degraded systems, with broad applications in social science, engineering and biology.

PMID:40727414 | PMC:PMC12303098 | DOI:10.1098/rsos.242240

Nat Commun. 2025 Jul 28;16(1):6925. doi: 10.1038/s41467-025-61833-8.

ABSTRACT

Protein-ligand interactions play central roles in myriad biological processes and are of key importance in drug design. Deep learning approaches are becoming cost-effective alternatives to high-throughput experimental methods for ligand identification. Here, to predict the binding affinity between proteins and small molecules, we introduce Ligand-Transformer, a deep learning method based on the transformer architecture. Ligand-Transformer implements a sequence-based approach, where the inputs are the amino acid sequence of the target protein and the topology of the small molecule to enable the prediction of the conformational space explored by the complex between the two. We apply Ligand-Transformer to screen and validate experimentally inhibitors targeting the mutant EGFRLTC kinase, identifying compounds with low nanomolar potency. We then use this approach to predict the conformational population shifts induced by known ABL kinase inhibitors, showing that sequence-based predictions enable the characterisation of the population shift upon binding. Overall, our results illustrate the potential of Ligand-Transformer to accurately predict the interactions of small molecules with proteins, including the binding affinity and the changes in the free energy landscapes upon binding, thus uncovering molecular mechanisms and facilitating the initial steps in drug design.

PMID:40721411 | DOI:10.1038/s41467-025-61833-8

AJNR Am J Neuroradiol. 2025 Jul 28:ajnr.A8943. doi: 10.3174/ajnr.A8943. Online ahead of print.

ABSTRACT

Magnetic resonance imaging (MRI) is a cornerstone of neuroimaging, providing unparalleled soft-tissue contrast. However, its clinical utility is often limited by long acquisition times, which contribute to motion artifacts, patient discomfort, and increased costs. Although traditional acceleration techniques, such as parallel imaging and compressed sensing help reduce scan times, they may reduce signal-to-noise ratio (SNR) and introduce artifacts. The advent of deep learning-based image reconstruction (DLBIR) may help in several ways to reduce scan times while preserving or improving image quality. Various DLBIR techniques are currently available through different vendors, with claimed reductions in gradient times up to 85% while maintaining or enhancing lesion conspicuity, improved noise suppression and diagnostic accuracy. The evolution of DLBIR from 2D to 3D acquisitions, coupled with advancements in self-supervised learning, further expands its capabilities and clinical applicability. Despite these advancements, challenges persist in generalizability across scanners and imaging conditions, susceptibility to artifacts and potential alterations in pathology representation. Additionally, limited data on training, underlying algorithms and clinical validation of these vendor-specific closed-source algorithms pose barriers to end-user trust and widespread adoption. This review explores the current applications of DLBIR in neuroimaging, vendor-driven implementations, and emerging trends that may impact accelerated MRI acquisitions.ABBREVIATIONS: PI＝ parallel imaging; CS＝ compressed sensing; DLBIR = deep learning-based image reconstruction; AI= artificial intelligence; DR =. Deep resolve; ACS = Artificial-intelligence-assisted compressed sensing.

PMID:40721279 | DOI:10.3174/ajnr.A8943

Neuroimage. 2025 Jul 26:121401. doi: 10.1016/j.neuroimage.2025.121401. Online ahead of print.

ABSTRACT

Diffusion MRI (dMRI) is an advanced imaging technique that enables in-vivo tracking of white matter fiber tracts and estimates the underlying cellular microstructure of brain tissues. Groupwise registration of dMRI data from multiple individuals is an important task for brain template creation and investigation of inter-subject brain variability. However, groupwise registration is a challenging task due to the uniqueness of dMRI data that include multi-dimensional, orientation-dependent signals that describe not only the strength but also the orientation of water diffusion in brain tissues. Deep learning approaches have shown successful performance in standard subject-to-subject dMRI registration. However, no deep learning methods have yet been proposed for groupwise dMRI registration. . In this work, we propose Deep Diffusion MRI Template (DDTemplate), which is a novel deep-learning-based method building upon the popular VoxelMorph framework to take into account dMRI fiber tract information. DDTemplate enables joint usage of whole-brain tissue microstructure and tract-specific fiber orientation information to ensure alignment of white matter fiber tracts and whole brain anatomical structures. We propose a novel deep learning framework that simultaneously trains a groupwise dMRI registration network and generates a population brain template. During inference, the trained model can be applied to register unseen subjects to the learned template. We compare DDTemplate with several state-of-the-art registration methods and demonstrate superior performance on dMRI data from multiple cohorts (adolescents, young adults, and elderly adults) acquired from different scanners. Furthermore, as a testbed task, we perform a between-population analysis to investigate sex differences in the brain, using the popular Tract-Based Spatial Statistics (TBSS) method that relies on groupwise dMRI registration. We find that using DDTemplate can increase the sensitivity in population difference detection, showing the potential of our method's utility in real neuroscientific applications.

PMID:40721052 | DOI:10.1016/j.neuroimage.2025.121401

J Adv Res. 2025 Jul 26:S2090-1232(25)00572-7. doi: 10.1016/j.jare.2025.07.042. Online ahead of print.

ABSTRACT

INTRODUCTION: Precise diagnosis and treatment of diseases necessitate quantitative visualization and modulation of subcellular structures. The endoplasmic reticulum (ER), as one of the most essential organelles, presents a complex target due to its intricate morphology and diverse cellular roles. Regulating ER stress offers a promising strategy for treating diseases such as tumors. However, achieving accurate targeting and therapeutic intervention at the subcellular level remains a significant challenge. Thus, there is an urgent need for theranostic agents that can precisely target and modulate ER stress.

OBJECTIVES: This study proposes a novel AI-driven dual-targeting strategy combining "passive + active" mechanisms to efficiently design molecules that resolve the balance between passive ER enrichment and precise modulation. We aim to design multifunctional theranostic molecules that precisely target Grp78, a key biomarker of ER stress, at the atomic level, enabling concurrent imaging and modulation of ER stress.

METHODS: A machine learning (ML)-based molecular fingerprints transfer method was developed for passive targeting based on identified subcellular targeting substructures. Meanwhile, a deep learning (DL)-based 3D molecular generation model, PM-1, was designed for active targeting through specific receptor interactions. By transferring key fingerprints and fluorescent motifs into PM-1-generated molecules, desired theranostic agents were produced. Their key properties were validated via dynamic simulations and quantitative calculations, followed by wet experiments.

RESULTS: Guided by these strategies, we identified unreported ER-targeting rules by discovering key passive-targeting fingerprints derived from ML models, and generated diverse new structures with high affinity binding to Grp78. We successfully synthesized ABT-CN2, a multidimensional fluorescent agent that demonstrates cost-effective chemical structure (molecular weight <400), robust targeting capability (Pearson's correlation coefficient = 0.93), and potential antitumor activity (IC50 = 53.21 μM).

CONCLUSION: This work presents a new paradigm for the intelligent design of fluorescent molecular probes with precise organelle-targeting capabilities for integrated diagnosis and therapy.

PMID:40721023 | DOI:10.1016/j.jare.2025.07.042

Int J Radiat Oncol Biol Phys. 2025 Jul 26:S0360-3016(25)06029-8. doi: 10.1016/j.ijrobp.2025.07.1431. Online ahead of print.

ABSTRACT

PURPOSE: Comprehensive geriatric assessment can identify older adult oncology patients at high-risk for adverse outcomes, but is variably feasible. Therefore, we assessed whether an abridged geriatric vulnerability model incorporating abstracted G8 score (G8), Charlson Comorbidity Index (CCI), and FaceAge (an AI-based aging measure) was associated with all-cause mortality or falls risk in patients undergoing stereotactic body radiation therapy (SBRT) for early-stage NSCLC.

METHODS: We reviewed the records of 708 patients aged ≥65 with stage I-II NSCLC treated with SBRT 06/01/09-03/31/23. We abstracted demographics, functional status (Eastern Cooperative Oncology Group (ECOG) score), oncologic history, G8, CCI, falls risk (Morse Fall Scale or Strategies to Reduce Injuries and Develop Confidence in Elders (STRIDE) screening tool), and time-to-death. FaceAge was calculated using AI-based facial analysis of SBRT simulation photographs. We examined associations between a vulnerability model incorporating G8, CCI, and FaceAge and 1) all-cause mortality and 2) falls risk, using Cox and regression models adjusted for age, sex, stage, ECOG, and significant covariates (p<0.10).

RESULTS: Patients (median age 76.2 years, 60.7% female, median stage IA), commonly had functional limitations (median ECOG 1, IQR 1-2), multimorbidity (median CCI 7, IQR 6-8), poor G8 scores (median 12.5, IQR 11.5-13.5), and elevated biological FaceAge (median 2.6 years above chronological age). In an adjusted Cox regression model, worse performance on all three geriatric vulnerability measures was independently associated with higher all-cause mortality (HRFaceAge 1.04, 95% confidence interval (CI)FaceAge 1.02-1.06, p=0.002; HRG8=1.15, 95%CIG8=1.08-1.22, p<0.001; HRCCI = 1.14, 95%CICCI=1.06-1.22, p<0.001). However, only worse G8 was associated with falls risk (HR=1.19, 95%CI 1.05-1.35,p=0.006).

CONCLUSIONS: Among older adults with early-stage NSCLC, a multimodal vulnerability measure leveraging routinely collected data was associated with all-cause mortality, identifying patients who might benefit from additional services.

PMID:40720998 | DOI:10.1016/j.ijrobp.2025.07.1431

Environ Monit Assess. 2025 Jul 28;197(8):960. doi: 10.1007/s10661-025-14386-8.

ABSTRACT

This study aims to evaluate the applicability of existing machine learning and deep learning techniques for the rapid prediction of hydrocarbon contamination in soils using hyperspectral data. Soil samples of three types, i.e., clayey, silty, and sandy, were synthetically contaminated with crude oil, diesel, and gasoline, creating a contamination range of 0 to 10,000 mg/kg. Hyperspectral imaging was employed to capture the spectral signatures of these samples, which were then analyzed using established models, including an XGB regressor and neural networks. Gas chromatography-mass spectrometry (GC-MS) was used to obtain reference contamination values. The models were trained and tested to predict hydrocarbon levels, with performance evaluated using R-squared and RMSE metrics. The models demonstrated strong predictive ability, achieving an R-squared value of 0.96 and an RMSE of 600 mg/kg on the testing set. Performance varied depending on the petroleum type and soil matrix. Gasoline models showed lower accuracy due to less distinguishable spectral features, while diesel and crude oil models performed better. Incorporating selected spectral bands as model inputs further improved performance by reducing overfitting. Among the evaluated models, the XGB regressor consistently provided a good balance between accuracy and robustness. This study highlights the effectiveness of applying hyperspectral spectral analysis with machine learning and deep learning models for soil contamination assessment. The findings support the use of ensemble-based models like XGB for practical spectral applications in environmental monitoring.

PMID:40721876 | DOI:10.1007/s10661-025-14386-8

Sci Rep. 2025 Jul 28;15(1):27506. doi: 10.1038/s41598-025-13275-x.

ABSTRACT

Skin diseases are one of the most common health problems that affect people of all ages around the world and significantly reduce the quality of life of individuals. Diseases of eczema, seborrheic dermatitis and skin cancer need to be diagnosed and correctly classified promptly. This issue, which is of great importance in terms of control and practical and effective treatment, is the study's starting point. The study included 693 individuals with eczema, 750 with skin cancer and 770 with seborrheic dermatitis. In the study, which focused on the classification of 3 different skin diseases, the Relief algorithm was used to increase the classification success and to ensure the selection of more meaningful qualities. With AlexNet with cross-validation, the accuracy rate was 89.39% for 80% training and 20% test rates. When SVM classification with the Relief algorithm was used for the same rates, the accuracy rate was 92.10%. In the analysis performed on the ISIC 2017 dataset, the accuracy rate is 89.16% for 80% training and 20% test rate. When the training and test rate was changed to 70% training and 30% test rate, the accuracy rate was 91.11%. It was observed that SVM classification with Relief's algorithm offers higher accuracy rates than other methods. The proposed model provides an original contribution to the literature, particularly through its integration of feature selection and a simplified architecture. This high success rate reveals that deep learning is an effective method in classifying skin diseases and the transfer learning process and will reduce the mortality rates due to cancer diseases with early and effective treatment while enabling skin diseases to be easily distinguished.

PMID:40721845 | DOI:10.1038/s41598-025-13275-x

Sci Rep. 2025 Jul 28;15(1):27436. doi: 10.1038/s41598-025-09846-7.

ABSTRACT

Monitoring species' presence in an ecosystem is crucial for conservation and understanding habitat diversity, but can be expensive and time consuming. As a result, ecologists have begun using the DNA that animals naturally leave behind in water or soil (called environmental DNA, or eDNA) to identify the species present in an environment. Recent work has shown that when used to identify species, convolutional neural networks (CNNs) can be as much as 150 times faster than ObiTools, a traditional method that does not use deep learning. However, CNNs are black boxes, meaning it is impossible to "fact check" why they predict that a given sequence belongs to a particular species. In this work, we introduce an interpretable, prototype-based CNN using the ProtoPNet framework that surpasses previous accuracy on a challenging eDNA dataset. The network is able to visualize the sequences of bases that are most distinctive for each species in the dataset, and introduces a novel skip connection that improves the interpretability of the original ProtoPNet. Our results show that reducing reliance on the convolutional output increases both interpretability and accuracy.

PMID:40721613 | DOI:10.1038/s41598-025-09846-7

Sci Rep. 2025 Jul 28;15(1):27476. doi: 10.1038/s41598-025-10794-5.

ABSTRACT

Intelligent Transportation Systems (ITS) necessitate scalable, real-time, and adaptive traffic flow prediction models to enhance urban mobility and alleviate congestion. Conventional Graph Neural Network methodologies encounter difficulties in managing extensive road networks, long-range temporal relationships, and computing efficiency for real-time applications. An innovative deep learning framework named Meta Temporal Hyperbolic Quantum Graph Neural Networks that integrates hyperbolic embeddings, meta learning, quantum graph, Neural Ordinary Differential Equation (NODEs) to improve the ITS Performance. Across many cities, meta learning facilitates swift adaptation with minimum retraining whereas hyperbolic graph embeddings efficiently depict hierarchical route configurations The usage of Quantum Graph Neural Networks (QGNNs) enhances graph-based scheming, enabling real-time traffic flow to forecast for extensive networks. Also, NODEs summarize ongoing traffic progress, enhancing precision under dynamic sceneries. Datasets like Los-loop and SZ-taxi datasets are validated by experiments which highlights the impact of the proposed MTH-QGNN model, acquiringamean value RMSE of 4.5 and MAE of 3.5, ensuring minimal prediction error. MTH-QGNN model constantly sustained accuracy above 80% and R2 values exceeding 83%, representing robust predictive trustworthiness. MTH-QGNN effectively captures complex spatiotemporal traffic patterns with a variance score above threshold value.

PMID:40721612 | DOI:10.1038/s41598-025-10794-5

Sci Rep. 2025 Jul 28;15(1):27388. doi: 10.1038/s41598-025-11632-4.

ABSTRACT

Deep learning techniques have become pivotal in medical image segmentation, but their success often relies on large, manually annotated datasets, which are expensive and labor-intensive to obtain. Additionally, different segmentation tasks frequently require retraining models from scratch, resulting in substantial computational costs. To address these limitations, we propose PDoRA, an innovative parameter-efficient fine-tuning method that leverages knowledge transfer from a pre-trained SwinUNETR model for a wide range of brain image segmentation tasks. PDoRA minimizes the reliance on extensive data annotation and computational resources by decomposing model weights into principal and residual weights. The principal weights are further divided into magnitude and direction, enabling independent fine-tuning to enhance the model's ability to capture task-specific features. The residual weights remain fixed and are later fused with the updated principal weights, ensuring model stability while enhancing performance. We evaluated PDoRA on three diverse medical image datasets for brain structure and metastasis segmentation. The results demonstrate that PDoRA consistently outperforms existing parameter-efficient fine-tuning methods, achieving superior segmentation accuracy and efficiency. Our code is available at https://github.com/Perfect199001/PDoRA/tree/main .

PMID:40721601 | DOI:10.1038/s41598-025-11632-4

NPJ Digit Med. 2025 Jul 28;8(1):483. doi: 10.1038/s41746-025-01866-x.

ABSTRACT

Three-dimensional magnetic resonance vessel wall imaging (3D MR-VWI) is critical for characterizing cerebrovascular pathologies, yet its clinical adoption is hindered by labor-intensive postprocessing. We developed VWI Assistant, a multi-sequence integrated deep learning platform trained on multicenter data (study cohorts 1981 patients and imaging datasets) to automate artery segmentation and reconstruction. The framework demonstrated robust performance across diverse patient populations, imaging protocols, and scanner manufacturers, achieving 92.9% qualified rate comparable to expert manual delineation. VWI Assistant reduced processing time by over 90% (10-12 min per case) compared to manual methods (p < 0.001) and improved inter-/intra-reader agreement. Real-world deployment (n = 1099 patients) demonstrated rapid clinical adoption, with utilization rates increasing from 10.8% to 100.0% within 12 months. By streamlining 3D MR-VWI workflows, VWI Assistant addresses scalability challenges in vascular imaging, offering a practical tool for routine use and large-scale research, significantly improving workflow efficiency while reducing labor and time costs.

PMID:40721485 | DOI:10.1038/s41746-025-01866-x

Sci Rep. 2025 Jul 28;15(1):27444. doi: 10.1038/s41598-025-12541-2.

ABSTRACT

Global population aging highlights the need to understand how the elderly perceive safety in urban public spaces. This study used image semantic segmentation to identify key visual elements from panoramic images. A dataset was created by combining manual scoring with deep learning to explore how pocket park environments impact older adults' safety perceptions. Analyzing 497 images from 29 pocket parks in Xiamen Island with LightGBM and SHAP tools, researchers identified visual elements that significantly affect seniors' safety perceptions. The findings indicate: (1) Elderly environmental safety perceptions in the 29 surveyed parks on Xiamen Island were generally positive, yet safety scores varied markedly across parks. (2) Pedestrian area, car, wall, person, billboard, parterre, and vegetation were identified as the seven visual elements most impactful on elderly environmental safety perceptions. (3) Interactions among visual elements were observed, with vegetation exerting a notably regulatory effect on environmental safety perceptions, significantly enhancing the elderly's perception of security. This study's empirical analysis elucidates the influence of visual elements in pocket parks on elderly environmental safety perceptions, offering practical guidance for park planners to design more inclusive and secure green spaces for the elderly, with broad application potential.

PMID:40721444 | DOI:10.1038/s41598-025-12541-2

Sci Rep. 2025 Jul 28;15(1):27423. doi: 10.1038/s41598-025-10758-9.

ABSTRACT

With the rapid development of smart furniture, gesture recognition has gained increasing attention as a natural and intuitive interaction method. However, in practical applications, issues such as limited data resources and insufficient semantic understanding have significantly constrained the effectiveness of gesture recognition technology. To address these challenges, this study proposes HAVIT, a hybrid deep learning model based on Vision Transformer and ALBEF, aimed at enhancing the performance of gesture recognition systems under data-scarce conditions. The model achieves efficient feature extraction and accurate recognition of gesture characteristics through the organic integration of Vision Transformer's feature extraction capabilities and ALBEF's semantic understanding mechanism. Experimental results demonstrate that on a fully labeled dataset, the HAVIT model achieved a classification accuracy of 91.83% and an AUC value of 0.92; under 20% label deficiency conditions, the model maintained an accuracy of 86.89% and an AUC value of 0.88, exhibiting strong robustness. The research findings provide new solutions for the development of smart furniture interaction technology and hold significant implications for advancing practical applications in this field.

PMID:40721440 | DOI:10.1038/s41598-025-10758-9

Sci Data. 2025 Jul 28;12(1):1309. doi: 10.1038/s41597-025-05631-3.

ABSTRACT

Cocoa is a multi-billion-dollar industry but research on improving yields through pollination remains limited. New embedded hardware and AI-based data analysis is advancing information on cocoa flower visitors, their identity and implications for yields. We present the first cocoa flower visitor dataset containing 5,792 images of Ceratopogonidae, Formicidae, Aphididae, Araneae, and Encyrtidae, and 1,082 background cocoa flower images. This dataset was curated from 23 million images collected over two years by embedded cameras in cocoa plantations in Hainan province, China. We exemplify the use of the dataset with different sizes of YOLOv8 models and by progressively increasing the background image ratio in the training set to identify the best-performing model. The medium-sized YOLOv8 model achieved the best results with 8% background images (F1 Score of 0.71, mAP50 of 0.70). Overall, this dataset is useful to compare the performance of deep learning model architectures on images with low contrast images and difficult detection targets. The data can support future efforts to advance sustainable cocoa production through pollination monitoring projects.

PMID:40721425 | DOI:10.1038/s41597-025-05631-3

Sci Rep. 2025 Jul 28;15(1):27464. doi: 10.1038/s41598-025-09330-2.

ABSTRACT

Breast cancer is a relatively common carcinoma among women worldwide and remains a considerable public health concern. Consequently, the prompt identification of cancer is crucial, as research indicates that 96% of cancers are treatable if diagnosed prior to metastasis. Despite being considered the gold standard for breast cancer evaluation, conventional mammography possesses inherent drawbacks, including accessibility issues, especially in rural regions, and discomfort associated with the procedure. Therefore, there has been a surge in interest in non-invasive, radiation-free alternative diagnostic techniques, such as thermal imaging (thermography). Thermography employs infrared thermal sensors to capture and assess temperature maps of human breasts for the identification of potential tumours based on areas of thermal irregularity. This study proposes an advanced computer-aided diagnosis (CAD) system called Thermo-CAD to assess early breast cancer detection using thermal imaging, aimed at assisting radiologists. The CAD system employs a variety of deep learning techniques, specifically incorporating multiple convolutional neural networks (CNNs) to enhance diagnostic accuracy and reliability. To effectively integrate multiple deep features and diminish the dimensionality of features derived from each CNN, feature transformation and selection methods, including non-negative matrix factorization and Relief-F, are used leading to a reduction in classification complexity. The Thermo-CAD system is assessed utilising two datasets: the DMR-IR (Database for Mastology Research Infrared Images), for distinguishing between normal and abnormal breast tissues, and a novel thermography dataset to distinguish abnormal instances as benign or malignant. Thermo-CAD has proven to be an outstanding CAD system for thermographic breast cancer detection, attaining 100% accuracy on the DMR-IR dataset (normal versus abnormal breast cancer) using CSVM and MGSVM classifiers, and lower accuracy using LSVM and QSVM classifiers. However, it showed a lower ability to distinguish benign from malignant cases (second dataset), achieving an accuracy of 79.3% using CSVM. Yet, it remains a promising tool for early-stage cancer detection, especially in resource-constrained environments.

PMID:40721416 | DOI:10.1038/s41598-025-09330-2

JMIR Infodemiology. 2025 Jul 28;5:e67333. doi: 10.2196/67333.

ABSTRACT

BACKGROUND: User demographics are often hidden in social media data due to privacy concerns. However, demographic information on substance use (SU) can provide valuable insights, allowing public health policy makers to focus on specific cohorts and develop efficient prevention strategies, especially during global crises such as the COVID-19 pandemic.

OBJECTIVE: This study aimed to analyze SU trends at the user level across different demographic dimensions, such as age, gender, race, and ethnicity, with a focus on the COVID-19 pandemic. The study also establishes a baseline for SU trends using social media data.

METHODS: The study was conducted using large-scale English-language data from Twitter (now known as X) over a 3-year period (2019, 2020, and 2021), comprising 1.13 billion posts. Following preprocessing, the SU posts were identified using our custom-trained deep learning model (Robustly Optimized Bidirectional Encoder Representations From Transformers Pretraining Approach [RoBERTa]), which resulted in the identification of 9 million SU posts. Then, demographic attributes, such as user type, age, gender, race, and ethnicity, as well as sentiments and emotions associated with each post, were extracted via a collection of natural language processing modules. Finally, various qualitative analyses were performed to obtain insight into user behaviors based on demographics.

RESULTS: The highest level of user participation in SU discussions was observed in 2020, with a 22.18% increase compared to 2019 and a 25.24% increase compared to 2021. Throughout the study period, male users and teenagers increasingly dominated the SU discussions across all substance types. During the COVID-19 pandemic, user participation in prescription medication discussions was notably higher among female users compared to other substance types. In addition, alcohol use increased by 80% within 2 weeks after the global pandemic declaration in 2020.

CONCLUSIONS: This study presents a large-scale, fine-grained analysis of SU on social media data, examining trends by age, gender, race, and ethnicity before, during, and after the COVID-19 pandemic. Our findings, contextualized with sociocultural and pandemic-specific factors, provide actionable insights for targeted public health interventions. This study establishes social media data (powered with artificial intelligence and natural language processing tools) as a valuable platform for real-time SU surveillance and prevention during crises.

PMID:40720823 | DOI:10.2196/67333

Chaos. 2025 Jul 1;35(7):073148. doi: 10.1063/5.0273597.

ABSTRACT

We extend an advanced variation of a machine learning algorithm, next-generation reservoir computing (NGRC), to forecast synthetic data generated by the Ikeda map, which is a model of a nonlinear optical cavity with an injected laser beam. The machine learning model is created by observing time-series data generated by the Ikeda map during a training phase, and the trained model is used to forecast the behavior in a closed-loop mode where it is no longer supplied with data from the map. The Ikeda map is a particularly challenging problem to learn because of the complicated map functions. We overcome the challenge by a novel improvement of the NGRC concept by emphasizing simpler polynomial models localized to well-designed regions of phase space and then blending these models between regions, a method that we call locality blended next-generation reservoir computing. This approach allows for better performance with relatively smaller data sets in comparison to deep learning methods and gives a new level of interpretability. We achieve forecasting horizons exceeding five Lyapunov times, and we demonstrate that the model reproduces the invariant measure of the attractor beyond the short-term forecasting horizon.

PMID:40720789 | DOI:10.1063/5.0273597

PLoS One. 2025 Jul 28;20(7):e0329107. doi: 10.1371/journal.pone.0329107. eCollection 2025.

ABSTRACT

Accurately determining responsibility in traffic accidents is crucial for ensuring fairness in law enforcement and optimizing responsibility standards. Traditional methods predominantly rely on subjective judgments, such as eyewitness testimonies and police investigations, which can introduce biases and lack objectivity. To address these limitations, we propose the AMFormer(Arithmetic Feature Interaction Transformer) framework-a deep learning model designed for robust and interpretable traffic accident responsibility prediction. By capturing complex interactions among key factors through spatiotemporal feature modeling, this framework facilitates precise multi-label classification of accident responsibility. Furthermore, we employ SHAP (SHapley Additive Interpretation) analysis to improve transparency by identifying the most influential features in attribution of responsibility, and provide an in-depth analysis of key features and how they combine to significantly influence attribution of responsibility. Experiments conducted on real-world datasets demonstrate that AMFormer outperforms both other deep learning models and traditional approaches, achieving an accuracy of 93.46% and an F1-Score of 93%. This framework not only enhances the credibility of traffic accident responsibility attribution but also establishes a foundation for future research into autonomous vehicle responsibility.

PMID:40720535 | DOI:10.1371/journal.pone.0329107