PLoS One. 2025 Aug 7;20(8):e0329492. doi: 10.1371/journal.pone.0329492. eCollection 2025.

ABSTRACT

BACKGROUND: The rapid development of urban systems and rising requirements for sustainable development lift resource management issues in smart communities. A fundamental problem for contemporary communities involves effectively using energy and water resources and waste management systems under environmental limitations. Artificial intelligence (AI) techniques at an advanced level deliver new methods that optimize resource management systems.

OBJECTIVE: The research builds and examines a deep-learning framework that optimizes the management of smart community resources. The framework leverages long short-term memory (LSTM) networks for temporal data, convolutional neural networks (CNNs) for spatial analysis, and autoencoders for anomaly detection. The system focuses on two main objectives, which include better forecasting precision, optimum resource distribution, and efficient detection of operational problems.

METHODS: Research validation employed data from the Amsterdam Open Data Platform and Singapore Government Open Data Portal joined by crowdsourced platforms FixMyStreet and OneService. The preprocessing phase involved three stages, i.e., cleaning and normalization and feature engineering steps, before model training and testing phases. Predictive models received assessment based on Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R². A comparison with traditional methods revealed the proposed approach delivered superior performance results.

RESULTS: The deep learning framework demonstrated superior performance, achieving an average reduction of 18.7% in resource consumption and a 16.2% reduction in operational costs. The models outperformed baseline methods, with LSTMs achieving an MAE of 1.8 for water demand prediction and autoencoders detecting anomalies with an F1-score of 95.5%.

CONCLUSION: Due to its effective capabilities, the proposed framework solves challenges in resource management for smart communities while showing the potential of AI-driven solutions for sustainable urban development. Research results demonstrate that integrating sophisticated deep-learning methods yields more significant potential for optimizing resource utilization while improving operational effectiveness.

PMID:40773471 | DOI:10.1371/journal.pone.0329492

Discov Oncol. 2025 Aug 7;16(1):1492. doi: 10.1007/s12672-025-03303-7.

ABSTRACT

INTRODUCTION: Glioblastoma (GBM) is the most malignant primary brain cancer, associated with a median overall survival of 15 months. Traditional diagnosis and prognosis heavily rely on clinical examination and histological investigation, both of which are subjective and time-consuming. advances in machine learning (ML) and deep learning (DL) have largely accelerated the research of GBMs by enhancing tumour segmentation, molecular characterization and survival prediction.

METHODOLOGY: We refer to the PRISMA guidelines to report this systematic review and meta-analysis. A total of 44 studies published from 2021 to 2025 were analyzed. We thoroughly searched the following sources: PubMed, Scopus and Web of Science. Review-specific inclusion criteria included studies reporting on diagnostic, prognostic, or response-prediction tasks in GBM that used ML/DL models and reports on quantitative performance metrics. The independent random-effects model estimated the performance of each clinical task, and subgroup analysis determined the variables influencing model accuracy.

RESULTS: The performance of the machine and deep learning models was strong across different clinical tasks. For overall survival prognosis, the pooled C-index was 0.78 (95%CI 0.74-0.82, I2 = 68.5%). The tumor segmentation models had a high average Dice Similarity Coefficient value (0.91, 95% CI 0.87-0.94, I2 = 45.2%). Molecular tests were highly accurate for the prediction of IDH1 mutation (pooled accuracy = 90.5%, 95% CI 88.1% to 92.8%) and MGMT methylation status (pooled accuracy = 97.8%, 95% CI 96.4% to 99.1%). Transformer models excelled over CNN in segmentation, and radionics-based ML could improve non-invasive molecular assessment.

CONCLUSION: Although AI techniques have demonstrated encouraging results in GBM studies for various clinical tasks, substantial challenges still preclude efficient clinical applicability. These developments can potentially improve medical practice with improved diagnosis, personalized treatment and fewer invasive procedures. Nevertheless, variation in data, weak external validation, and missing prospective clinical studies warrant careful interpretation of these results.

PMID:40773129 | DOI:10.1007/s12672-025-03303-7

Curr Med Sci. 2025 Aug 7. doi: 10.1007/s11596-025-00103-6. Online ahead of print.

ABSTRACT

Artificial intelligence (AI) serves as a key technology in global industrial transformation and technological restructuring and as the core driver of the fourth industrial revolution. Currently, deep learning techniques, such as convolutional neural networks, enable intelligent information collection in fields such as tongue and pulse diagnosis owing to their robust feature-processing capabilities. Natural language processing models, including long short-term memory and transformers, have been applied to traditional Chinese medicine (TCM) for diagnosis, syndrome differentiation, and prescription generation. Traditional machine learning algorithms, such as neural networks, support vector machines, and random forests, are also widely used in TCM diagnosis and treatment because of their strong regression and classification performance on small structured datasets. Future research on AI in TCM diagnosis and treatment may emphasize building large-scale, high-quality TCM datasets with unified criteria based on syndrome elements; identifying algorithms suited to TCM theoretical data distributions; and leveraging AI multimodal fusion and ensemble learning techniques for diverse raw features, such as images, text, and manually processed structured data, to increase the clinical efficacy of TCM diagnosis and treatment.

PMID:40773005 | DOI:10.1007/s11596-025-00103-6

Radiol Med. 2025 Aug 7. doi: 10.1007/s11547-025-02051-6. Online ahead of print.

ABSTRACT

Fractures and their complications are recognized as major public health problems. Especially for occult fractures that are difficult to judge radiologically, timely and accurate diagnosis is particularly important for the treatment and prognosis of patients. In recent years, the successful application of radiomics and deep learning in medical diagnosis has shown great potential for providing more timely and accurate diagnostic methods for occult fractures. This review provides an introduction to radiomics and deep learning, summarizes their respective characteristics in detecting occult fractures, and subsequently conducts a detailed analysis on the potential value and future prospects of integrating these two techniques to develop an enhanced approach for prompt and precise detection of occult fractures.

PMID:40772999 | DOI:10.1007/s11547-025-02051-6

Neuroinformatics. 2025 Aug 7;23(3):41. doi: 10.1007/s12021-025-09743-4.

ABSTRACT

The amygdala plays a central role in emotion, memory, and decision-making and comprises approximately 13 distinct nuclei with connectivity. Despite its functional importance, high-resolution subnuclear mapping is challenging. This study aimed to construct a 3D model of the anatomical location of the amygdala in the brain and a functional dynamic model of the amygdala, integrating deep learning and elastic shape metrics. We used multimodal datasets from the Julich-Brain Atlas, BigBrain Project, and FreeSurfer, which were aligned with the Montreal Neurological Institute (MNI) and Colin 27 spaces. Subnuclei segmentation was performed using a Bayesian Fully Convolutional Network (FCN), and geometric morphometrics were analyzed using elastic shape analysis on the unit sphere. Functional dynamics were simulated using a MATLAB-based model of the amygdala incorporating theta (4-8 Hz) and gamma (30-40 Hz) oscillations with spike-timing-dependent plasticity (STDP). The mean MNI coordinates of the left and right amygdalae were (-20, -4, -15) and (22, -2, -15), respectively, with an inter-amygdalar distance of 42.48 mm. The Dice Similarity Coefficients (DSCs) for FCN-based subnuclear segmentation were as follows: basolateral amygdala (BLA) nucleus = 0.89 ± 0.03, centromedial nucleus = 0.83 ± 0.04, and cortical nucleus = 0.81 ± 0.05. Principal component analysis of elastic shape metrics revealed post-traumatic stress disorder (PTSD)-related morphological deviations, with the first principal component (PC1) accounting for 38% of the variance (p < 0.01). Oscillatory simulations captured the BLA rhythm dynamics and STDP-induced synaptic changes. This study presents a comprehensive 3D model of the human amygdala that bridges anatomical accuracy with computational modeling. Unlike prior models that focus solely on structural or functional domains, our approach integrates subnuclear segmentation, morphometrics, and real-time functional simulation. This study introduces a fully integrated anatomical-functional 3D model of the human amygdala, providing a translational platform for neuromodulation targeting, psychiatric diagnostics, and computational neuroengineering applications.

PMID:40772991 | DOI:10.1007/s12021-025-09743-4

Rev Sci Instrum. 2025 Aug 1;96(8):084701. doi: 10.1063/5.0279493.

ABSTRACT

The accurate detection and evaluation of metal material defects is of great significance to the current production and life. When the metal material is damaged, its internal magnetic permeability will change locally. The electromagnetic tomography (EMT) technique can be used to reconstruct the combined permeability and conductivity distribution of metal materials. However, the ill-posed and ill-conditioned nature of the EMT inverse problem, coupled with the high permeability and conductivity of ferromagnetic materials, poses significant challenges for defect detection. To address this, we propose an improved deep learning model, P-LeNet, based on a convolutional neural network for EMT defect detection and image reconstruction. By establishing a nonlinear mapping between induced voltage measurements and the combined permeability and conductivity distribution, the model extracts multi-scale features to enhance reconstruction accuracy and robustness. The correlation coefficient and image error are used as indicators to evaluate the quality of image reconstruction. In order to visually demonstrate the imaging effect of the proposed model, numerical simulations are performed. The imaging results show that the proposed P-LeNet model is superior to traditional algorithms in imaging accuracy, artifact suppression, and overall performance. At the same time, Gaussian white noise is introduced to evaluate the anti-noise ability of the model, and the random sample is used to test the generalization ability of the model to fully demonstrate the superiority and application potential of the method. Furthermore, experiments with a nine-coil planar EMT sensor are conducted to verify the effectiveness and superiority of the proposed model.

PMID:40772849 | DOI:10.1063/5.0279493

Expert Rev Proteomics. 2025 Aug 7. doi: 10.1080/14789450.2025.2545828. Online ahead of print.

ABSTRACT

INTRODUCTION: Machine learning holds significant promise for accelerating biomarker discovery in clinical proteomics, yet its real-world impact remains limited by widespread methodological pitfalls and unrealistic expectations.

AREAS COVERED: In this perspective, we critically examine the application of machine learning for biomarker discovery in clinical proteomics, emphasizing that algorithmic novelty alone cannot compensate for issues such as small sample sizes, batch effects, overfitting, data leakage, and poor model generalization.

EXPERT OPINION: We caution against the uncritical application of complex models, such as deep learning architectures, that often exacerbate these problems, offering limited interpretability and negligible performance gains in typical clinical proteomics datasets. Instead, we advocate for the realistic and responsible use of machine learning, grounded in rigorous study design, appropriate validation strategies, and transparent, reproducible modeling practices. Emphasizing simplicity, interpretability, and domain awareness over hype-driven complexity is essential if machine learning is to fulfill its translational potential in the clinic.

PMID:40772544 | DOI:10.1080/14789450.2025.2545828

Ann Med. 2025 Dec;57(1):2540596. doi: 10.1080/07853890.2025.2540596. Epub 2025 Aug 7.

ABSTRACT

BACKGROUND: The incidence of adrenal incidentalomas (AIs) is increasing yearly. The early discovery of AIs is helpful to better manage adrenal diseases, especially subclinical primary aldosteronism, Cushing's syndrome and pheochromocytoma.

METHODS: In this multicenter retrospective study, a total of 778 patients from three different medical centers were assessed. The two-stage cascade network consisted of a 3D Res-Unet network for adrenal gland segmentation and a classifier for determining the presence of AIs. The segmentation network was mainly evaluated by the Dice similarity coefficient (DSC), and the classifier was evaluated by the area under the receiver operator characteristic curve (AUC), accuracy, sensitivity, and specificity. The Delong test was used to compare the classification performance between the cascade network and manual segmentation.

RESULTS: A total of 443 patients were randomly assigned in a 7:3 ratio, stratified sampling, to train and valid sets of the model development cohort, and 335 patients from the three centers were included in the test cohort. In the validation set, the AUC of the model for identifying left AI was 88.15%, and the AUC of the model for identifying right AI was 87.90%. There was no significant difference between model performance and manual segmentation of AIs (p > 0.05). In the test cohort, the cascade network achieved AUC of more than 80% and accuracy of more than 75% for both left and right adrenal glands.

CONCLUSIONS: The two-stage cascade network based on a deep learning algorithm can be used for automatic recognition of AIs in nonenhanced CT from different centers.

PMID:40772430 | DOI:10.1080/07853890.2025.2540596

Front Genet. 2025 Jul 23;16:1614222. doi: 10.3389/fgene.2025.1614222. eCollection 2025.

ABSTRACT

INTRODUCTION: Enhancer-promoter interactions (EPIs) play a vital role in the regulation of gene expression. Although traditional wet-lab methods provide valuable insights into EPIs, they are often constrained by high costs and limited scalability. As a result, the development of efficient computational models has become essential. However, many current deep learning and machine learning approaches utilize simplistic feature fusion strategies, such as direct averaging or concatenation, which fail to effectively model complex relationships and dynamic importance across features. This often results in suboptimal performance in challenging biological contexts.

METHODS: To address these limitations, we propose a deep learning model named EPI-DynFusion. This model begins by encoding DNA sequences using pre-trained DNA embeddings and extracting local features through convolutional neural networks (CNNs). It then integrates a Transformer and Bidirectional Gated Recurrent Unit (BiGRU) architecture with a Dynamic Feature Fusion mechanism to adaptively learn deep dependencies among features. Furthermore, we incorporate the Convolutional Block Attention Module (CBAM) to enhance the model's ability to focus on informative regions. Based on this core architecture, we develop two variants: EPI-DynFusion-gen, a general model, and EPI-DynFusion-best, a fine-tuned version for cell line-specific data.

RESULTS: We evaluated the performance of our models across six benchmark cell lines. The average area under the receiver operating characteristic curve (AUROC) scores achieved by the specific, generic, and best models were 94.8%, 95.0%, and 96.2%, respectively. The average area under the precision-recall curve (AUPR) scores were 81.2%, 71.1%, and 83.3%, respectively, demonstrating the superior performance of the fine-tuned model in the precision-recall space. These results confirm that the proposed fusion strategies and attention mechanisms contribute to significant improvements in performance.

DISCUSSION: In conclusion, EPI-DynFusion presents a robust and scalable framework for predicting enhancer-promoter interactions solely based on DNA sequence information. By addressing the limitations of conventional fusion techniques and incorporating attention mechanisms alongside sequence modeling, our method achieves state-of-the-art performance while enhancing the interpretability and generalizability of enhancer-promoter interaction prediction tasks.

PMID:40772277 | PMC:PMC12325019 | DOI:10.3389/fgene.2025.1614222

Front Neurosci. 2025 Jul 23;19:1592070. doi: 10.3389/fnins.2025.1592070. eCollection 2025.

ABSTRACT

Emotion recognition based on electroencephalogram (EEG) faces substantial challenges. The variability of neural signals among different subjects and the scarcity of labeled data pose obstacles to the generalization ability of traditional domain adaptation (DA) methods. Existing approaches, especially those relying on the maximum mean discrepancy (MMD) technique, are often highly sensitive to domain mean shifts induced by noise. To overcome these limitations, a novel framework named Domain Adaptive Deep Possibilistic clustering (DADPc) is proposed. This framework integrates deep domain-invariant feature learning with possibilistic clustering, reformulating Maximum Mean Discrepancy (MMD) as a one-centroid clustering task under a fuzzy entropy-regularized framework. Moreover, the DADPc incorporates adaptive weighted loss and memory bank strategies to enhance the reliability of pseudo-labels and cross-domain alignment. The proposed framework effectively mitigates noise-induced domain shifts while maintaining feature discriminability, offering a robust solution for EEG-based emotion recognition in practical applications. Extensive experiments conducted on three benchmark datasets (SEED, SEED-IV, and DEAP) demonstrate the superior performance of DADPc in emotion recognition tasks. The results show significant improvements in recognition accuracy and generalization capability across different experimental protocols, including cross-subject and cross-session scenarios. This research contributes to the field by providing a comprehensive approach that combines deep learning with possibilistic clustering, advancing the state-of-the-art in cross-domain EEG analysis.

PMID:40772260 | PMC:PMC12326482 | DOI:10.3389/fnins.2025.1592070

Cureus. 2025 Jul 7;17(7):e87461. doi: 10.7759/cureus.87461. eCollection 2025 Jul.

ABSTRACT

Idiopathic pulmonary fibrosis (IPF) is a devastating interstitial lung disease (ILD) characterized by progressive fibrosis and poor survival outcomes. Accurate diagnosis and prognosis remain challenging due to overlapping features with other ILDs and variability in imaging interpretation. This systematic review evaluates the current evidence on artificial intelligence (AI) and machine learning (ML) applications for the diagnosis and prognosis of IPF using computed tomography (CT) imaging. Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, eight studies published between 2017 and 2024 were included, demonstrating promising results across various methodologies, including deep learning (DL) models, support vector machines (SVMs), and ensemble approaches. AI-derived parameters, particularly measures of fibrotic burden and pulmonary vascular volume, consistently outperformed conventional visual CT scores for prognostication. Strong correlations between AI-quantified CT features and pulmonary function (PF) tests suggest potential surrogate markers for physiological parameters. Novel prognostic biomarkers identified through AI analysis expand understanding beyond traditional parenchymal assessment. Despite these advances, limitations include retrospective designs, sample size constraints, male-predominant cohorts, and limited external validation. Future research should prioritize large, prospective, multi-center studies with diverse populations, standardized protocols, explainable AI (XAI) techniques, and integration into clinical workflows to realize the transformative potential of AI for improving IPF management.

PMID:40772136 | PMC:PMC12327841 | DOI:10.7759/cureus.87461

Nat Mach Intell. 2024 Sep;6(9):1106-1118. doi: 10.1038/s42256-024-00892-w. Epub 2024 Sep 19.

ABSTRACT

Two-photon calcium imaging provides large-scale recordings of neuronal activities at cellular resolution. A robust, automated and high-speed pipeline to simultaneously segment the spatial footprints of neurons and extract their temporal activity traces while decontaminating them from background, noise and overlapping neurons is highly desirable to analyze calcium imaging data. In this paper, we demonstrate DeepCaImX, an end-to-end deep learning method based on an iterative shrinkage-thresholding algorithm and a long-short-term-memory neural network to achieve the above goals altogether at a very high speed and without any manually tuned hyper-parameter. DeepCaImX is a multi-task, multi-class and multi-label segmentation method composed of a compressed-sensing-inspired neural network with a recurrent layer and fully connected layers. It represents the first neural network that can simultaneously generate accurate neuronal footprints and extract clean neuronal activity traces from calcium imaging data. We trained the neural network with simulated datasets and benchmarked it against existing state-of-the-art methods with in vivo experimental data. DeepCaImX outperforms existing methods in the quality of segmentation and temporal trace extraction as well as processing speed. DeepCaImX is highly scalable and will benefit the analysis of mesoscale calcium imaging.

PMID:40771998 | PMC:PMC12327232 | DOI:10.1038/s42256-024-00892-w

Front Neurol. 2025 Jul 16;16:1607924. doi: 10.3389/fneur.2025.1607924. eCollection 2025.

ABSTRACT

This bibliometric review examines the evolving landscape of artificial intelligence (AI) in neurodegenerative diseases research from 2000 to March 16, 2025, utilizing data from 1,402 publications (1,159 articles, 243 reviews) indexed in the Web of Science Core Collection. Through advanced tools - VOSviewer, CiteSpace, and Bibliometrix R - the study maps collaboration networks, keyword trends, and knowledge trajectories. Results reveal exponential growth post-2017, driven by advancements in deep learning and multimodal data integration. The United States (25.96%) and China (24.11%) dominate publication volume, while the UK exhibits the highest collaboration centrality (0.24) and average citations per publication (31.68). Core journals like Scientific Reports and Frontiers in Aging Neuroscience published the most articles in this field. Highly cited publications and burst references highlight important milestones in the development history. High-frequency keywords include "alzheimer's disease," "parkinson's disease," "magnetic resonance imaging," "convolutional neural network," "biomarkers," "dementia," "classification," "mild cognitive impairment," "neuroimaging," and "feature extraction." Key hotspots include intelligent neuroimaging analysis, machine learning methodological iterations, molecular mechanisms and drug discovery, and clinical decision support systems for early diagnosis. Future priorities encompass advanced deep learning architectures, multi-omics integration, explainable AI systems, digital biomarker-based early detection, and transformative technologies including transformers and telemedicine. This analysis delineates AI's transformative role in optimizing diagnostics and accelerating therapeutic innovation, while advocating for enhanced interdisciplinary collaboration to bridge computational advances with clinical translation.

PMID:40771972 | PMC:PMC12327369 | DOI:10.3389/fneur.2025.1607924

Front Artif Intell. 2025 Jul 23;8:1589358. doi: 10.3389/frai.2025.1589358. eCollection 2025.

ABSTRACT

INTRODUCTION: Knee injuries, especially Anterior Cruciate Ligament (ACL) tears and meniscus tears, are becoming increasingly common and can severely restrict mobility and quality of life. Early diagnosis is essential for effective treatment and for preventing long-term complications such as knee osteoarthritis. While deep learning approaches have shown promise in identifying knee injuries from MRI scans, they often require large amounts of labeled data, which can be both scarce and privacy-sensitive.

METHODS: This paper analyses a hybrid methodology that integrates few-shot learning with federated learning for the diagnosis of knee injuries using MRI scans. The proposed model used a 3DResNet50 architecture as the backbone to enhance both feature extraction and embedding representation. A combined Centralized and Federated Few-Shot Learning Framework is analysed to leverage episodic-intermittent training strategy based on Prototypical Networks. The model is trained incorporating Stochastic Gradient Descent (SGD), Cross-Entropy Loss, and a MultiStep Learning Rate scheduler to enhance few-shot classification. This model also addressed the challenge of limited annotated data ensuring patient data privacy through distributed learning across multiple regions.

RESULTS: The models performance was evaluated on the MRNet dataset for multi-label classification. In the centralized setting, the model achieved accuracies of 85.3% on axial views, 82.1% on sagittal views, and 71% on coronal views. The propose work attained accuracies as 83% (axial), 83.9% (sagittal), and 65% (coronal), demonstrating the framework's effectiveness across different learning configurations.

DISCUSSION: The proposed method outperforms in diagnostic accuracy, generalization across MRI planes, and patient privacy via federated learning. However, it faces limitations, including lower coronal view performance and high computational demands due to its complex architecture.

PMID:40771942 | PMC:PMC12326743 | DOI:10.3389/frai.2025.1589358

Front Artif Intell. 2025 Jul 23;8:1582960. doi: 10.3389/frai.2025.1582960. eCollection 2025.

ABSTRACT

BACKGROUND: Osteoporosis, a systemic skeletal disorder characterized by deteriorated bone microarchitecture and low bone mass, poses substantial fracture risks to aging populations globally. Early detection of reduced bone mineral density (BMD) through opportunistic screening is critical for preventing fragility fractures. Although dual-energy X-ray absorptiometry (DXA) is the gold standard for diagnosing osteoporosis, many patients have not undergone screening with this technique. Therefore, developing an automated tool that can diagnose bone density through routine chest and abdominal CT examinations is highly important. With advancements in technology and the accumulation of clinical data, the role of bone density artificial intelligence (AI) in the diagnosis and management of osteoporosis is becoming increasingly significant.

OBJECTIVE: First to validate the diagnostic equivalence of AI-based BMD prediction against quantitative CT (QCT) reference standards, second to assess inter-device measurement consistency across multi-vendor CT systems (Siemens, GE, Philips). Ultimately, the objective is to determine the clinical utility of AI-derived BMD for osteoporosis classification.

METHODS: In this retrospective multicenter study, paired CT/QCT datasets from 702 patients (2019-2022) were analyzed. The accuracy, sensitivity, and specificity of an Bone Density AI model were evaluated by comparing the predicted bone mineral density values from bone density AI with the measured values from QCT. Moreover, the consistency of lumbar spine BMD measurements between QCT and Bone Density AI on different devices was compared.

RESULTS: The AUC of Bone Density AI model in diagnosing osteoporosis was 0.822 (95% CI: 0.787-0.867, p < 0.001), with an accuracy of 0.9456, sensitivity of 0.9601, and specificity of 0.9270, indicating good performance in predicting bone density. The consistency study between Bone Density AI and QCT for the vertebral BMD measurements revealed no statistically significant difference in R 2 values, suggesting no significant difference in performance between the two methods in measuring BMD. The linear regression fit between the R 2 values of QCT and Bone Density AI for measuring lumbar spine BMD with different equipment ranged from 0.88 to 0.96, indicating a high degree of consistency between the two measurement methods across devices.

CONCLUSION: This multicenter study pioneers a dual-validation framework to establish the clinical validity of deep learning-based BMD prediction algorithms using routine thoracic/abdominal CT scans. Our data suggest that AI-driven BMD quantification demonstrates non-inferior diagnostic accuracy to QCT while overcoming DXA's accessibility limitations. This technology enables cost-effective, radiation-free osteoporosis screening through routine CT repurposing, particularly beneficial for resource-constrained settings.

PMID:40771941 | PMC:PMC12325224 | DOI:10.3389/frai.2025.1582960

Front Artif Intell. 2025 Jul 23;8:1618607. doi: 10.3389/frai.2025.1618607. eCollection 2025.

ABSTRACT

BACKGROUND: Abdominal ultrasonography is a primary diagnostic tool for evaluating medical conditions within the abdominal cavity. Accurate determination of the relative locations of intra-abdominal organs and lesions based on anatomical features in ultrasound images is essential in diagnostic sonography. Recognizing and extracting anatomical landmarks facilitates lesion evaluation and enhances diagnostic interpretation. Recent artificial intelligence (AI) segmentation methods employing deep neural networks (DNNs) and transformers encounter computational efficiency challenges to balance the preservation of feature dependencies information with model efficiency, limiting their clinical applicability.

METHODS: The anatomical structure recognition framework, MaskHybrid, was developed using a private dataset comprising 34,711 abdominal ultrasound images of 2,063 patients from CSMUH. The dataset included abdominal organs and vascular structures (hepatic vein, inferior vena cava, portal vein, gallbladder, kidney, pancreas, spleen) and liver lesions (hepatic cyst, tumor). MaskHybrid adopted a mamba-transformer hybrid architecture consisting of an evolved backbone network, encoder, and corresponding decoder to capture long-range spatial dependencies and contextual information effectively, demonstrating improved image segmentation capabilities in visual tasks while mitigating the computational burden associated with the transformer-based attention mechanism.

RESULTS: Experiments on the retrospective dataset achieved a mean average precision (mAP) score of 74.13% for anatomical landmarks segmentation in abdominal ultrasound images. Our proposed framework outperformed baselines across most organ and lesion types and effectively segmented challenging anatomical structures. Moreover, MaskHybrid exhibited a significantly shorter inference time (0.120 ± 0.013 s), achieving 2.5 times faster than large-sized AI models of similar size. Combining Mamba and transformer architectures, this hybrid design was well-suited for the timely analysis of complex anatomical structures segmentation in abdominal ultrasonography, where accuracy and efficiency are critical in clinical practice.

CONCLUSION: The proposed mamba-transformer hybrid recognition framework simultaneously detects and segments multiple abdominal organs and lesions in ultrasound images, achieving superior segmentation accuracy, visualization effect, and inference efficiency, thereby facilitating improved medical image interpretation and near real-time diagnostic sonography that meets clinical needs.

PMID:40771938 | PMC:PMC12325247 | DOI:10.3389/frai.2025.1618607

Front Pharmacol. 2025 Jul 23;16:1581820. doi: 10.3389/fphar.2025.1581820. eCollection 2025.

ABSTRACT

BACKGROUND: Sorafenib, a multi-kinase inhibitor, is a key therapeutic agent in the treatment of advanced hepatocellular carcinoma (HCC), metastatic renal cell carcinoma (RCC), and radioactive iodine-refractory differentiated thyroid cancer (DTC). However, its clinical efficacy is frequently hampered by the rising prevalence of sorafenib resistance, particularly in HCC. This reality underscores the urgent need for a comprehensive pan-cancer investigation to elucidate the underlying mechanisms of resistance.

METHODS: We employed a systematic bibliometric approach utilizing the Web of Science Core Collection to conduct a structured literature search. Performance analysis and visualization were conducted using VOSviewer and CiteSpace. A triphasic screening protocol was implemented to identify publications focused on sorafenib resistance, covering a period from 2006 to 2025.

RESULTS: Our analysis identified 1,484 eligible publications, with a peak of 194 articles published in 2022. The majority of research (79.48%) centered on HCC, with significant contributions from institutions in China and the United States. Co-authorship and keyword network analyses revealed a robust interdisciplinary collaboration landscape. Key themes emerged, including dysregulation of drug transporters and clearance mechanisms, metabolic reprogramming, programmed cell death, interactions within the tumor microenvironment, and epigenetic regulatory mechanisms, highlighting critical areas contributing to resistance.

CONCLUSION: This study highlights the current research landscape concerning sorafenib resistance, facilitating the identification of emerging trends and research gaps. Future research priorities should include biomarker-driven clinical trials, the development of nanoparticle delivery systems, and the clinical translation of combination therapy strategies. Additionally, the integration of deep learning algorithms in the context of big data has the potential to enhance our understanding of resistance mechanisms in silico, ultimately aiming to overcome resistance challenges and improve patient survival outcomes.

PMID:40771926 | PMC:PMC12325432 | DOI:10.3389/fphar.2025.1581820

Front Pharmacol. 2025 Jul 23;16:1608832. doi: 10.3389/fphar.2025.1608832. eCollection 2025.

ABSTRACT

INTRODUCTION: Drug combination therapy represents a promising strategy for addressing complex diseases, offering the potential for improved efficacy while mitigating safety concerns. However, conventional wet-lab experimentation for identifying optimal drug combinations is resource-intensive due to the vast combinatorial search space. To address this challenge, computational methods leveraging machine learning and deep learning have emerged to effectively navigate this space.

METHODS: In this study, we introduce a Calibrated Deep Feature Aggregation (CDFA) framework for screening synergistic drug combinations. Concretely, CDFA utilizes a novel cell line representation based on the protein information and gene expression capturing complementary biological determinants of drug response. Besides, a novel feature aggregation network is proposed based on the Transformer to model the intricate interactions between drug pairs and cell lines through multi-head attention mechanisms, enabling discovery of non-linear synergy patterns. Furthermore, a method is introduced to quantify and calibrate the uncertainties associated with CDFA's predictions, enhancing the reliability of the identified synergistic drug combinations.

RESULTS: Experiments results have demonstrated that CDFA outperforms existing state-of-the-art deep learning models.

DISCUSSION: The superior performance of CDFA stems from its biologically informed cell line representation, its ability to capture complex non-linear drug-cell interactions via attention mechanisms, and its enhanced reliability through uncertainty calibration. This framework provides a robust computational tool for efficient and reliable drug combination screening.

PMID:40771923 | PMC:PMC12325400 | DOI:10.3389/fphar.2025.1608832

Lancet Digit Health. 2025 Aug 5:100894. doi: 10.1016/j.landig.2025.100894. Online ahead of print.

ABSTRACT

BACKGROUND: Sexually transmitted infections (STIs) are a substantial public health concern. We aimed to evaluate the accuracy and applicability of deep learning algorithms in the early detection of STIs from skin lesions.

METHODS: In this systematic review and meta-analysis, we searched PubMed, Institute of Electrical and Electronics Engineers Xplore, Web of Science, Scopus for studies employing deep learning for classifying clinical skin lesion images of STIs published between Jan 1, 2010, and Dec 31, 2023. Studies that did not include clinical images were excluded. The primary outcome was diagnostic performance, assessed by pooled sensitivity and specificity. We conducted a meta-analysis of the studies providing contingency tables using a unified hierarchical model. We additionally assessed the quality of the studies using modified QUADAS-2 and CheckList for Evaluation of image-based AI Reports in Dermatology (CLEAR Derm) criteria. This study was registered with PROSPERO, CRD42024496966.

FINDINGS: Among the 1946 studies identified, we included 101 in our review. The majority of the included studies focused on mpox (91 [88%] of 101 studies), followed by scabies (eight [8%] studies), herpes (four [4%] studies), syphilis (one [1%] study), and molluscum (one [1%] study). A meta-analysis of 55 studies showed that deep learning algorithms had a pooled sensitivity of 0·97 (95% CI 0·95-0·98) and a specificity of 0·99 (0·98-0·99) for mpox, and a sensitivity of 0·95 (0·90-0·98) and specificity of 0·97 (0·86-0·99) for scabies. The majority of studies (86 [85%] of 101 studies) utilised public datasets; traditional convolutional neural networks with backbone architectures such as ResNet and VGGNet were used in all studies. However, notable quality issues related to the data, technical descriptions of labelling methods and diagnostic label references, technical assessment for public evaluation of algorithms, benchmarking and bias assessments, application descriptions of use cases, and target conditions and potential impacts were identified in CLEAR Derm. Potential biases in performance evaluation metrics and applicability concerns in the data, deep learning algorithms, and performance evaluation metrics might impede the generalisability of these models to real-world clinical practice and STI screening across diverse populations.

INTERPRETATION: Although deep learning shows potential for early detection of STIs, there are challenges to ensuring the generalisability of such algorithms due to limited heterogeneous data. Standardised, diverse skin lesion image datasets are crucial to ensure fair comparisons and reliable performance.

FUNDING: City University of Hong Kong.

PMID:40769792 | DOI:10.1016/j.landig.2025.100894

Int J Pharm. 2025 Aug 4:126024. doi: 10.1016/j.ijpharm.2025.126024. Online ahead of print.

ABSTRACT

Prostate cancer (PCa) represents one of the most prevalent cancers among men, with substantial challenges in timely and accurate diagnosis and subsequent treatment. Traditional diagnosis and treatment methods for PCa, such as prostate-specific antigen (PSA) biomarker detection, digital rectal examination, imaging (CT/MRI) analysis, and biopsy histopathological examination, suffer from limitations such as a lack of specificity, generation of false positives or negatives, and difficulty in handling large data, leading to overdiagnosis and overtreatment. The integration of artificial intelligence (AI) in PCa diagnosis and treatment is revolutionizing traditional approaches by offering advanced tools for early detection, personalized treatment planning, and patient management. AI technologies, especially machine learning and deep learning, improve diagnostic accuracy and treatment planning. The AI algorithms analyze imaging data, like MRI and ultrasound, to identify cancerous lesions effectively with great precision. In addition, AI algorithms enhance risk assessment and prognosis by combining clinical, genomic, and imaging data. This leads to more tailored treatment strategies, enabling informed decisions about active surveillance, surgery, or new therapies, thereby improving quality of life while reducing unnecessary diagnoses and treatments. This review examines current AI applications in PCa care, focusing on their transformative impact on diagnosis and treatment planning while recognizing potential challenges. It also outlines expected improvements in diagnosis through AI-integrated systems and decision support tools for healthcare teams. The findings highlight AI's potential to enhance clinical outcomes, operational efficiency, and patient-centred care in managing PCa.

PMID:40769449 | DOI:10.1016/j.ijpharm.2025.126024