Abdom Radiol (NY). 2025 Jul 30. doi: 10.1007/s00261-025-05123-3. Online ahead of print.

ABSTRACT

OBJECTIVES: Delineating the Couinaud segments is a critical component of liver surgery and monitoring that has traditionally relied on labor-intensive methods that are prone to variability. While fully or semi-automatic methods exist, they generally lack accuracy or require extensive post-processing or corrections to the outputs.

METHODS: We present a framework that integrates deep learning-based segmentation with auxiliary landmark identification to create a personalized pictorial model on which to base precise Couinaud landmark localization. Data from 225 non-contrast T1-weighted MRIs from 4 different studies were used to evaluate the performance against benchmark techniques and human-defined ground truth.

RESULTS: The personalized model outperformed the benchmark method in every landmark placement and Couinaud segment volume estimation, being significantly better in 5/8 landmarks and 7/8 segments.

CONCLUSION: The proposed system is explainable, agnostic to imaging modality and is able to incorporate new data without retraining, enhancing its robustness and scalability across diverse clinical contexts. These findings underscore the potential of our framework to substantially improve Couinaud accuracy and streamline clinical workflows, optimizing liver surgery planning and monitoring.

PMID:40736570 | DOI:10.1007/s00261-025-05123-3

Korean J Radiol. 2025 Aug;26(8):771-781. doi: 10.3348/kjr.2025.0208.

ABSTRACT

OBJECTIVE: To evaluate the effect of deep learning (DL)-based artificial intelligence (AI) software on the diagnostic performance of radiologists with different experience levels in detecting nigrosome 1 (N1) abnormalities on susceptibility map-weighted imaging (SMwI).

MATERIALS AND METHODS: This retrospective diagnostic case-control study analyzed 139 SMwI scans of 59 patients with Parkinson's disease (PD) and 80 healthy participants. Participants were imaged using 3T MRI, and AI-generated assessments for N1 abnormalities were obtained using an AI model (version 1.0.1.0; Heuron Corporation, Seoul, Korea), which utilized YOLOX-based object detection and SparseInst segmentation models. Four radiologists (two experienced neuroradiologists and two less experienced residents) evaluated N1 abnormalities with and without AI in a crossover study design. Diagnostic performance metrics, inter-reader agreements, and reader responses to AI-generated assessments were evaluated.

RESULTS: Use of AI significantly improved diagnostic performance compared with interpretation without it across three readers, with significant increases in specificity (0.86 vs. 0.94, P = 0.004; 0.91 vs. 0.97, P = 0.024; and 0.90 vs. 0.97, P = 0.012). Inter-reader agreement also improved with AI, as Fleiss's kappa increased from 0.73 (95% confidence interval [CI]: 0.61-0.84) to 0.87 (95% CI: 0.76-0.99). The net reclassification index (NRI) demonstrated significant improvement in three of the four readers. When grouped by experience level, less experienced readers showed greater improvement (NRI = 12.8%, 95% CI: 0.067-0.190) than experienced readers (NRI = 0.8%, 95% CI: -0.037-0.051). In the less experienced group, reader-AI disagreement was significantly higher in the PD group than in the normal group (8.1% vs. 3.8%, P = 0.029).

CONCLUSION: DL-based AI enhances the diagnostic performance in detecting N1 abnormalities on SMwI, particularly benefiting less experienced radiologists. These findings underscore the potential for improving diagnostic workflows for PD.

PMID:40736409 | DOI:10.3348/kjr.2025.0208

Sci Prog. 2025 Jul-Sep;108(3):368504251352707. doi: 10.1177/00368504251352707. Epub 2025 Jul 30.

ABSTRACT

In response to the challenges of complex background interference, inadequate feature utilization, and model redundancy in multispectral crown extraction, this paper proposes a dual-channel crown detection and segmentation approach based on an improved YOLOv7 architecture, named Dual-YOLOv7. First, a dual-branch feature extraction network is designed, integrating visible light and infrared spectral information and dynamically weights key features through an attention mechanism. Second, the D-SimSPPF module is introduced, which employs depthwise separable convolution to optimize spatial pyramid pooling, thereby enhancing the capability to capture fine details while reducing the number of parameters. Furthermore, the CIoU-C loss function is developed, incorporating a shape penalty factor to improve the accuracy of bounding box regression. Experimental results demonstrate that the improved model achieves detection and segmentation mAP50 scores of 91.6% and 90.1%, respectively, representing increases of 7.7 and 7.6 percentage points over YOLOv7-seg. After channel pruning, the model parameter count is reduced by 14.2%, offering a lightweight solution suitable for unmanned aerial vehicle platforms.

PMID:40736346 | DOI:10.1177/00368504251352707

Am J Clin Pathol. 2025 Jul 30:aqaf067. doi: 10.1093/ajcp/aqaf067. Online ahead of print.

ABSTRACT

OBJECTIVE: Cytologic examination, which confirms the presence or absence of malignant cells, detects malignant cells from various organs, with adenocarcinoma as the most common histologic type. We developed a deep learning model to detect malignant cells in images obtained following effusion cytology.

METHODS: The deep learning model was created using the YOLOv8 object detection algorithm (Roboflow, Inc) and 275 cases of adenocarcinoma comprising 12 182 images and 29 245 labels as well as 188 cases negative for malignancy comprising 1980 images.

RESULTS: The adenocarcinoma test dataset exhibited Precision, Recall, F1, and mean average Precision scores of 0.909, 0.911, 0.910, and 0.955, respectively. The number of adenocarcinoma test images in which 1 or more malignant cells were detected was 2710 of 2731. The sensitivity in the nonadenocarcinoma dataset was 97.1%, and the false-positive rate in the negative-for-malignancy dataset was 7.3%. The accuracy, sensitivity, and specificity of the model using all the test datasets were 96.3%, 98.5%, and 92.7%, respectively.

CONCLUSIONS: Although some issues regarding cell annotation when creating an object detection model remain, the accuracy is sufficient to assist cancer screening in effusion cytology. It is vital to reliably detect malignant cells in effusion cytology, and the further development of automated systems to reduce false-negative results is expected.

PMID:40736208 | DOI:10.1093/ajcp/aqaf067

J Chem Inf Model. 2025 Jul 30. doi: 10.1021/acs.jcim.5c01017. Online ahead of print.

ABSTRACT

Lead optimization in drug discovery faces the dual challenge of maintaining structural diversity while preserving core molecular features and optimizing the balance between biological activity and drug-like properties. To address these challenges, we introduce the Deep Genetic Molecule Modification (DGMM) algorithm, a novel computational framework that synergistically integrates deep learning architectures with genetic algorithms for efficient molecular optimization. DGMM leverages a variational autoencoder (VAE) with an enhanced representation learning strategy that incorporates scaffold constraints during training, significantly improving the latent space organization to balance structural variation with scaffold retention. A multiobjective optimization strategy, combining Monte Carlo search and Markov processes, enables systematic exploration of the trade-offs between drug likeness and target activity. Evaluation results indicate that DGMM achieves state-of-the-art performance in activity optimization, generating structurally diverse, yet pharmacologically relevant compounds. To rigorously establish its utility, we first demonstrated its generalizability through extensive retrospective validation on three diverse targets (CHK1, CDK2, and HDAC8), reproducing their known optimization pathways. Building on this validated generalizability, we deployed DGMM in a prospective campaign, which culminated in the wet-lab discovery of novel ROCK2 inhibitors with a notable 100-fold increase in biological activity. This success establishes DGMM as an effective tool for structural optimization of drug molecules.

PMID:40736165 | DOI:10.1021/acs.jcim.5c01017

Anal Chem. 2025 Jul 30. doi: 10.1021/acs.analchem.5c02934. Online ahead of print.

ABSTRACT

Raman imaging utilizes molecular fingerprint information to visualize the spatial distribution of a substance within the scanned area. Subject to its scanning mechanism, it usually costs a prolonged data acquisition duration for achieving high-resolution Raman images. In this study, we propose a generative adversarial network (GANs) based algorithm to significantly enhance both the Raman spectral imaging speed and spatial resolution. The proposed method was trained and evaluated on 186 hyperspectral Raman datasets acquired from unlabeled cells, and its reconstruction performance was quantitatively evaluated by the parameters of peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and root-mean-square error (RMSE). Univariate imaging and K-means clustering analysis (KCA) were both adopted to evaluate the preservation of biochemical information after image reconstructing. The results demonstrated that the proposed method effectively enhances spatial resolution by a factor of 2-4 while accelerating imaging speed by a factor of 4-16. Furthermore, transfer learning was utilized to adapt the pretrained model to different objects, validating its generalization capabilities and extending its universalities. This study highlighted the potential of deep learning for super-resolution Raman imaging, providing a promising pathway for high-throughput and real-time biochemical analysis.

PMID:40735851 | DOI:10.1021/acs.analchem.5c02934

Front Physiol. 2025 Jul 15;16:1629637. doi: 10.3389/fphys.2025.1629637. eCollection 2025.

ABSTRACT

BACKGROUND: With global aging and lifestyle changes, carotid atherosclerotic plaques are a major cause of cerebrovascular disease and ischemic stroke. However, ultrasound images suffer from high noise, low contrast, and blurred edges, making it difficult for traditional image processing methods to accurately extract plaque information.

OBJECTIVE: To establish a deep learning-based DualPlaqueNet model for semantic segmentation and size prediction of plaques in carotid ultrasound images, thereby providing comprehensive and accurate auxiliary information for clinical risk assessment and personalized diagnosis and treatment.

METHODS: DualPlaqueNet uses a dual-branch architecture combined with attention mechanisms and joint loss functions to optimize segmentation and regression. Notably, a multi-layer one-dimensional convolutional structure is introduced within the Efficient Channel Attention (ECA) module. The original dataset contained 287 carotid ultrasound images from patients at Zhengzhou First People's Hospital, which were divided into training, validation, and test sets. Model training, validation, and testing were performed after preprocessing and data augmentation of the training set. Its performance was compared with three other models.

RESULTS: In the plaque semantic segmentation task, DualPlaqueNet outperformed the other three models across all metrics, achieving MIoU of 88.91 ± 1.027 (%), IoU (excluding background) of 88.22 ± 1.065 (%), DSC of 89.95 ± 1.102 (%), and Accuracy of 95.98 ± 0.073 (%). For plaque size prediction, this model demonstrated lower MSE and MAE, along with a higher coefficient of determination R 2, proving its ability to accurately extract plaque size information from ultrasound images.

CONCLUSION: The dual-branch design and attention mechanisms of DualPlaqueNet effectively address the challenges of ultrasound images, achieving precise segmentation and size prediction, demonstrating its potential as an auxiliary tool for future clinical applications.

PMID:40735675 | PMC:PMC12303906 | DOI:10.3389/fphys.2025.1629637

Front Physiol. 2025 Jul 15;16:1611267. doi: 10.3389/fphys.2025.1611267. eCollection 2025.

ABSTRACT

INTRODUCTION: Anterior cruciate ligament (ACL) injuries hold significant clinical importance, making the development of accurate and efficient diagnostic tools essential. Deep learning has emerged as an effective method for detecting ACL tears. However, current models often struggle with multiscale and boundary-sensitive tear patterns and tend to be computationally intensive.

METHODS: We present LRU-Net, a lightweight residual U-Net designed for ACL tear segmentation. LRU-Net integrates an advanced attention mechanism that emphasizes gradients and leverages the anatomical position of the ACL, thereby improving boundary sensitivity. Furthermore, it employs a dynamic feature extraction module for adaptive multiscale feature extraction. A dense decoder featuring dense connections enhances feature reuse.

RESULTS: In experimental evaluations, LRU-Net achieves a Dice Coefficient Score of 97.93% and an Intersection over Union (IoU) of 96.40%.

DISCUSSION: It surpasses benchmark models such as Attention-Unet, Attention-ResUnet, InceptionV3-Unet, Swin-UNet, Trans-UNet and Rethinking ResNets. With a reduced computational footprint, LRU-Net provides a practical and highly accurate solution for the clinical analysis of ACL tears.

PMID:40735674 | PMC:PMC12303908 | DOI:10.3389/fphys.2025.1611267

Quant Plant Biol. 2025 Jul 25;6:e23. doi: 10.1017/qpb.2025.10018. eCollection 2025.

ABSTRACT

Recent advancements in data science and artificial intelligence have significantly transformed plant sciences, particularly through the integration of image recognition and deep learning technologies. These innovations have profoundly impacted various aspects of plant research, including species identification, disease detection, cellular signaling analysis, and growth monitoring. This review summarizes the latest computational tools and methodologies used in these areas. We emphasize the importance of data acquisition and preprocessing, discussing techniques such as high-resolution imaging and unmanned aerial vehicle (UAV) photography, along with image enhancement methods like cropping and scaling. Additionally, we review feature extraction techniques like colour histograms and texture analysis, which are essential for plant identification and health assessment. Finally, we discuss emerging trends, challenges, and future directions, offering insights into the applications of these technologies in advancing plant science research and practical implementations.

PMID:40735612 | PMC:PMC12304785 | DOI:10.1017/qpb.2025.10018

Digit Health. 2025 Jul 28;11:20552076251362508. doi: 10.1177/20552076251362508. eCollection 2025 Jan-Dec.

ABSTRACT

Lymphoma is a prevalent malignant tumor within the hematological system, posing significant challenges to clinical practice due to its diverse subtypes, intricate radiological and metabolic manifestations. Lymphoma segmentation studies based on positron emission tomography/computed tomography (PET/CT), CT, and magnetic resonance imaging represent key strategies for addressing these challenges. This article reviews the advancements in lymphoma segmentation research utilizing deep learning methods, offering a comparative analysis with traditional approaches, and conducting an in-depth examination and summary of aspects such as dataset characteristics, backbone networks of models, adjustments to network structures based on research objectives, and model performance. The article also explores the potential and challenges of translating deep learning-based lymphoma segmentation research into clinical scenarios, with a focus on practical clinical applications. The future research priorities in lymphoma segmentation are identified as enhancing the models' clinical generalizability, integrating into clinical workflows, reducing computational demands, and expanding high-quality datasets. These efforts aim to facilitate the broad application of deep learning in the diagnosis and treatment monitoring of lymphoma.

PMID:40735544 | PMC:PMC12304644 | DOI:10.1177/20552076251362508

MethodsX. 2025 Jul 15;15:103498. doi: 10.1016/j.mex.2025.103498. eCollection 2025 Dec.

ABSTRACT

Wildfires present a growing threat to ecosystems, human settlements, and climate stability, necessitating accurate and interpreted prediction systems. Existing AI-based models often prioritize performance over explainability, limiting their utility in real-time decision-making contexts. Current wildfire forecasting models struggle to incorporate uncertainty and offer transparent response strategies. Moreover, many models fail to integrate domain knowledge in a way that supports actionable interventions. This study utilizes the Canadian Fire Spread Dataset, augmented with Sentinel, ERA5, and SRTM data, encompassing vegetation, meteorological, and topographic variables. The suggested system uses a Transformer-based model to predict fires over time and space, along with a Fuzzy Rule-Based System (FRBS) to create rules for responding to those predictions. This integration allows for both high accuracy and interpretability in decision-making under uncertain environmental conditions. The novelty lies in the use of symbolic fuzzy reasoning layered onto a deep attention-based architecture. Performance was evaluated using metrics such as accuracy, precision, recall, F1-score, and AUC. The model achieved an F1-score of 92.9 % and accuracy of 94.8 %, significantly outperforming baseline and deep learning alternatives. • Integrates deep learning with fuzzy logic for both accurate forecasting and interpretable response planning. • Enables uncertainty-aware reasoning by translating predictions into actionable fire management rules. • Demonstrates superior performance across diverse environmental datasets using multi-source satellite and climate inputs.

PMID:40735517 | PMC:PMC12304742 | DOI:10.1016/j.mex.2025.103498

Eur J Radiol Open. 2025 Jul 22;15:100671. doi: 10.1016/j.ejro.2025.100671. eCollection 2025 Dec.

ABSTRACT

PURPOSE: To evaluate and compare the image quality and lesion conspicuity of prostate T2-weighted imaging (T2WI) using four reconstruction methods: conventional Sensitivity Encoding (SENSE), compressed sensing (CS), model-based deep learning reconstruction (DL), and deep learning super-resolution reconstruction (SR).

METHODS: This retrospective study included 49 patients who underwent multiparametric MRI (mpMRI) or biparametric MRI (bpMRI) for suspected prostate cancer. Axial T2WI was acquired using two protocols: conventional SENSE and CS-based acquisition. From the CS-based data, three reconstruction methods (CS, DL, and SR) were applied to generate additional images. Two board-certified radiologists independently assessed overall image quality and sharpness using a 4-point Likert scale (1 = poor, 4 = excellent). Quantitative analysis included signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and sharpness index. PI-RADS T2WI scoring and lesion conspicuity were preliminarily evaluated in 18 individuals with pathologically confirmed prostate cancer. Statistical comparisons were conducted using the Wilcoxon signed-rank test.

RESULTS: SR consistently achieved the highest scores in both qualitative (overall image quality, image sharpness) and quantitative (SNR, CNR, sharpness index) assessments, compared with SENSE, CS, and DL (all pairwise comparisons, Bonferroni-corrected p < 0.0001). In lesion-based analysis, SR showed a trend toward improved lesion conspicuity, although PI-RADS T2WI scores were similar across reconstruction.

CONCLUSION: SR reconstruction demonstrated superior image quality in both qualitative and quantitative assessments and showed potential benefits for lesion visualization. These findings, although based on a small sample, suggest that SR may be a promising approach for prostate MRI and warrants further investigation in larger populations.

PMID:40735490 | PMC:PMC12305308 | DOI:10.1016/j.ejro.2025.100671

IEEE Access. 2024;12:98096-98106. doi: 10.1109/access.2024.3427631. Epub 2024 Jul 15.

ABSTRACT

This study developed an automated matchmaking system using deep neural networks to enhance the efficiency of pairing researcher biosketches with funders' requests for proposals (RFPs). In thus U.S., with over 900 federal grant programs and 86,000+ foundations, researchers often spend up to 200 hours on each application due to low success rates, forcing them to apply multiple times a year. Our approach improves on existing systems by fixing issues like unreliable keyword searches, and one-size-fits-all recommendations. We analyzed 12,991 biosketches from a research institution and 2,234 RFPs from the National Institutes of Health, spanning 2014 to 2019. Employing four advanced deep-learning models, utilizing cross and Siamese encoding strategies, we benchmarked their performance against conventional predictive models such as logistic regression and support vector machines. The most effective model integrated BERT with cross-encoding, a post-BERT BiLSTM layer, and back translation (BC2BT), achieving an F1-score of 71.15%. These results demonstrate the potential of sophisticated natural language processing techniques to automate complex matchmaking tasks in the research funding sector. This approach not only improves the precision of matching researchers to suitable funding opportunities but also sets a promising foundation for future advancements in automated funding mechanisms.

PMID:40735468 | PMC:PMC12306506 | DOI:10.1109/access.2024.3427631

Front Med (Lausanne). 2025 Jul 15;12:1590201. doi: 10.3389/fmed.2025.1590201. eCollection 2025.

ABSTRACT

The early and accurate diagnosis of Alzheimer's Disease and Frontotemporal Dementia remains a critical challenge, particularly with traditional machine learning models which often fail to provide transparency in their predictions, reducing user confidence and treatment effectiveness. To address these limitations, this paper introduces an explainable and lightweight deep learning framework comprising temporal convolutional networks and long short-term memory networks that efficiently classifies Frontotemporal dementia (FTD), Alzheimer's Disease (AD), and healthy controls using electroencephalogram (EEG) data. Feature engineering has been conducted using modified Relative Band Power (RBP) analysis, leveraging six EEG frequency bands extracted through power spectrum density (PSD) calculations. The model achieves high classification accuracies of 99.7% for binary tasks and 80.34% for multi-class classification. Furthermore, to enhance the transparency and interpretability of the framework, SHAP (SHapley Additive exPlanations) has been utilized as an explainable artificial intelligence technique that provides insights into feature contributions.

PMID:40735445 | PMC:PMC12303882 | DOI:10.3389/fmed.2025.1590201

Comput Struct Biotechnol J. 2025 Jul 15;27:3182-3190. doi: 10.1016/j.csbj.2025.07.026. eCollection 2025.

ABSTRACT

PURPOSE: This study aimed to predict the cell types that infiltrate the tumor microenvironment using hematoxylin and eosin-stained images from colon cancer and breast cancer samples.

METHODS: Two datasets, one focused on colon cancer and the other on breast cancer, were used to develop deep learning models. Cell segmentation was performed using Stardist, followed by the K-Nearest Neighbor method to construct a neighborhood-enhanced cellular extraction matrix for model training. Transductive semi-supervised learning was applied to the breast cancer dataset, where the Base-4 model was trained on S1 and S2 samples and subsequently used to generate assigned labels for the S3, S4, and S5 sets, on which the Base-4+ model was trained.

RESULTS: The Base-7 model trained on colon cancer cell images achieved accuracy of 0.85 on the hold-out test set and 0.74- on the independent test set, with six neighboring cells identified as the optimal condition for prediction. In addition, the Base-4 model achieved a prediction accuracy of 0.69 with four neighboring cells as the optimal condition in the breast cancer dataset, while the Base-4+ model reached an accuracy of up to 0.93 on the validation set. The model also captured invasive and ductal carcinoma cells with overall agreement relative to spot-based cell types (0.63).

CONCLUSIONS: Deep learning models accurately predicted cell types in breast and colon cancer datasets using only cell morphology and neighborhood embedding.

PMID:40735431 | PMC:PMC12305602 | DOI:10.1016/j.csbj.2025.07.026

Sci Rep. 2025 Jul 29;15(1):27552. doi: 10.1038/s41598-025-12264-4.

ABSTRACT

Criteria to identify patients who are ready to be liberated from mechanical ventilation (MV) are imprecise, often resulting in prolonged MV or reintubation, both of which are associated with adverse outcomes. Daily protocol-driven assessment of the need for MV leads to earlier extubation but requires dedicated personnel. We sought to determine whether machine learning (ML) applied to the electronic health record could predict next-day extubation. We examined 37 clinical features aggregated from 12AM-8AM on each patient-ICU-day from a single-center prospective cohort study of patients in our quaternary care medical ICU who received MV. We also tested our models on an external test set from a community hospital ICU in our health care system. We used three data encoding/imputation strategies and built XGBoost, LightGBM, logistic regression, LSTM, and RNN models to predict next-day extubation. We compared model predictions and actual events to examine how model-driven care might have differed from actual care. Our internal cohort included 448 patients and 3,095 ICU days, and our external test cohort had 333 patients and 2,835 ICU days. The best model (LSTM) predicted next-day extubation with an AUROC of 0.870 (95% CI 0.834-0.902) on the internal test cohort and 0.870 (95% CI 0.848-0.885) on the external test cohort. Across multiple model types, measures previously demonstrated to be important in determining readiness for extubation were found to be most informative, including plateau pressure and Richmond Agitation Sedation Scale (RASS) score. Our model often predicted patients to be stable for extubation in the days preceding their actual extubation, with 63.8% of predicted extubations occurring within three days of true extubation. Our findings suggest that an ML model may serve as a useful clinical decision support tool rather than complete replacement of clinical judgement. However, any ML-based model should be compared with protocol-based practice in a prospective, randomized controlled trial to determine improvement in outcomes while maintaining safety as well as cost effectiveness.

PMID:40731125 | DOI:10.1038/s41598-025-12264-4

Sci Rep. 2025 Jul 29;15(1):27690. doi: 10.1038/s41598-025-13079-z.

ABSTRACT

Rice diseases pose a critical threat to global crop yields, underscoring the need for rapid and accurate diagnostic tools to ensure effective crop management and productivity. Traditional diagnostic approaches often lack both precision and scalability, frequently necessitating specialized equipment and expertise. This study presents a deep learning-based automated diagnostic system for rice leaf diseases, leveraging a large-scale dataset comprising annotated images spanning six common rice diseases: bacterial stripe, false smut, leaf blast, neck blast, sheath blight, and brown spot. We evaluated seven advanced deep learning architectures-MobileNetV2, GoogLeNet, EfficientNet, ResNet-34, DenseNet-121, VGG16, and ShuffleNetV2-across a range of performance metrics including precision, recall, and overall diagnostic accuracy. Among these, GoogLeNet, DenseNet-121, ResNet-34, and VGG16 demonstrated superior performance, particularly in minimizing class confusion and enhancing diagnostic accuracy. These models were selected based on diverse architectural principles to ensure complementary feature extraction capabilities. An ensemble model, integrating these four high-performing networks via a simple average fusion strategy, was subsequently developed, significantly reducing misclassification rates and providing robust, scalable diagnostic capabilities suitable for deployment in real-world agricultural settings. The model's performance was further validated on independent test data collected under varying environmental conditions.

PMID:40730662 | DOI:10.1038/s41598-025-13079-z

Sci Rep. 2025 Jul 29;15(1):27607. doi: 10.1038/s41598-025-13337-0.

ABSTRACT

The integration of heterogeneous multi-omics datasets at a systems level remains a central challenge for developing analytical and computational models in precision cancer diagnostics. This paper introduces Multi-Omics Graph Kolmogorov-Arnold Network (MOGKAN), a deep learning framework that utilizes messenger-RNA, micro-RNA sequences, and DNA methylation samples together with Protein-Protein Interaction (PPI) networks for cancer classification across 31 different cancer types. The proposed approach combines differential gene expression with DESeq2, Linear Models for Microarray (LIMMA), and Least Absolute Shrinkage and Selection Operator (LASSO) regression to reduce multi-omics data dimensionality while preserving relevant biological features. The model architecture is based on the Kolmogorov-Arnold theorem principle and uses trainable univariate functions to enhance interpretability and feature analysis. MOGKAN achieves classification accuracy of 96.28% and exhibits low experimental variability in comparison to related deep learning-based models. The biomarkers identified by MOGKAN were validated as cancer-related markers through Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis. By integrating multi-omics data with graph-based deep learning, our proposed approach demonstrates robust predictive performance and interpretability with potential to enhance the translation of complex multi-omics data into clinically actionable cancer diagnostics.

PMID:40730661 | DOI:10.1038/s41598-025-13337-0

NPJ Digit Med. 2025 Jul 29;8(1):485. doi: 10.1038/s41746-025-01890-x.

ABSTRACT

Existing prostate cancer monitoring methods, reliant on prostate-specific antigen (PSA) measurements in blood tests often fail to detect tumor growth. We develop a computational framework to reconstruct tumor growth from the PSA integrating physics-based modeling and machine learning in digital twins. The physics-based model considers PSA secretion and flux from tissue to blood, depending on local vascularity. This model is enhanced by deep learning, which regulates tumor growth dynamics through the patient's PSA blood tests and 3D spatial interactions of physiological variables of the digital twin. We showcase our framework by reconstructing tumor growth in real patients over 2.5 years from diagnosis, with tumor volume relative errors ranging from 0.8% to 12.28%. Additionally, our results reveal scenarios of tumor growth despite no significant rise in PSA levels. Therefore, our framework serves as a promising tool for prostate cancer monitoring, supporting the advancement of personalized monitoring protocols.

PMID:40730645 | DOI:10.1038/s41746-025-01890-x

Sci Rep. 2025 Jul 29;15(1):27668. doi: 10.1038/s41598-025-11375-2.

ABSTRACT

Traffic forecasting is considered a cornerstone of smart city development. A key challenge is capturing the long-term spatiotemporal dependencies of traffic data while improving the model's generalization ability. To address these issues, various sophisticated modules are embedded into different models. However, this approach increases the computational cost of the model. Additionally, adding or replacing datasets in a trained model requires retraining, which decreases prediction accuracy and increases time cost. To address the challenges faced by existing models in handling long-term spatiotemporal dependencies and high computational costs, this study proposes an enhanced pre-training method called the Improved Spatiotemporal Diffusion Graph (ImPreSTDG). While existing traffic prediction models, particularly those based on Graph Convolutional Networks (GCNs) and deep learning, are effective at capturing short-term spatiotemporal dependencies, they often experience accuracy degradation and increased computational demands when dealing with long-term dependencies. To overcome these limitations, we introduce a Denoised Diffusion Probability Model (DDPM) as part of the pre-training process, which enhances the model's ability to learn from long-term spatiotemporal data while significantly reducing computational costs. During the pre-training phase, ImPreSTDG employs a data masking and recovery strategy, with DDPM facilitating the reconstruction of masked data segments, thereby enabling the model to capture long-term dependencies in the traffic data. Additionally, we propose the Mamba module, which leverages the Selective State Space Model (SSM) to effectively capture long-term multivariate spatiotemporal correlations. This module enables more efficient processing of long sequences, extracting essential patterns while minimizing computational resource consumption. By improving computational efficiency, the Mamba module addresses the challenge of modeling long-term dependencies without compromising accuracy in capturing extended spatiotemporal trends. In the fine-tuning phase, the decoder is replaced with a forecasting header, and the pre-trained parameters are frozen. The forecasting header includes a meta-learning fusion module and a spatiotemporal convolutional layer, which facilitates the integration of both long-term and short-term traffic data for accurate forecasting. The model is then trained and adapted to the specific forecasting task. Experiments conducted on three real-world traffic datasets demonstrate that the proposed pre-training method significantly enhances the model's ability to handle long-term dependencies, missing data, and high computational costs, providing a more efficient solution for traffic prediction.

PMID:40730627 | DOI:10.1038/s41598-025-11375-2