Radiography (Lond). 2025 Jan 3;31(1):372-378. doi: 10.1016/j.radi.2024.12.018. Online ahead of print.

ABSTRACT

BACKGROUND: Facial recognition technology in medical imaging, particularly with head scans, poses privacy risks due to identifiable facial features. This study evaluates the use of facial recognition software in identifying facial features from head CT scans and explores a defacing pipeline using TotalSegmentator to reduce re-identification risks while preserving data integrity for research.

METHODS: 1404 high-quality renderings from the UCLH EIT Stroke dataset, both with and without defacing were analysed. The performance of defacing with the face mask created by TotalSegmentator was compared to a state-of-the-art CT defacing algorithm. Face detection was performed using deep learning models. The cosine similarity between facial embeddings for intra- and inter-patient images was compared. A Support Vector Machine was trained on cosine similarity values to assess defacing performance, determining if two renderings came from the same patient. This analysis was conducted on defaced and non-defaced images using 5-fold cross-validation.

RESULTS: Faces were detected in 76.5 % of non-defaced images. Intra-patient images exhibited a median cosine similarity of 0.65 (IQR: 0.47-0.80), compared to 0.50 (IQR: 0.39-0.62) for inter-patient images. A binary classifier performed moderately on non-defaced images, achieving a ROC-AUC of 0.69 (SD = 0.01) and an accuracy of 0.65 (SD = 0.01) in distinguishing whether a scan belonged to the same or a different individual. Following defacing, performance declined markedly. Defacing with the TotalSegmentator decreased the ROC-AUC to 0.55 (SD = 0.02) and the accuracy to 0.56 (SD = 0.01), whereas the CTA-DEFACE algorithm brought the performance down to a ROC-AUC of 0.60 (SD = 0.02) and an accuracy of 0.59 (SD = 0.01). These results demonstrate the effectiveness of defacing algorithms in mitigating re-identification risks, with the TotalSegmentator providing slightly superior privacy protection.

CONCLUSION: Facial recognition software can identify facial features from partial and complete head CT scan renderings. However, using the TotalSegmentator to deface images reduces re-identification risks to a near-chance level. We offer code to implement this privacy-preserving pipeline.

IMPLICATIONS FOR PRACTICE: Utilizing the TotalSegmentator framework, the proposed pipeline efficiently removes facial features from CT images, making it ideal for multi-site research and data sharing. It is a useful tool for radiographers and radiologists who must comply with medico-legal requirements necessitating the removal of facial features.

PMID:39754865 | DOI:10.1016/j.radi.2024.12.018

Sci Rep. 2025 Jan 4;15(1):818. doi: 10.1038/s41598-024-84421-0.

ABSTRACT

Mobile Ad Hoc Networks (MANETs) are increasingly replacing conventional communication systems due to their decentralized and dynamic nature. However, their wireless architecture makes them highly vulnerable to flooding attacks, which can disrupt communication, deplete energy resources, and degrade network performance. This study presents a novel hybrid deep learning approach integrating Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures to effectively detect and mitigate flooding attacks in MANETs. To optimize the model's efficiency, a unique DECEHGS algorithm combining Differential Evolution and Evolutionary Population Dynamics techniques is employed, enhancing both convergence and performance. The proposed model demonstrates significant improvements over existing methods, achieving an accuracy of 95%, a 12% increase in packet delivery ratio, and a 20% reduction in routing overhead compared to traditional techniques. These advancements underline the model's superiority in detecting malicious nodes, conserving energy, and ensuring reliable network performance. The comprehensive evaluation using MATLAB R2023a validates the proposed approach as an effective and energy-efficient solution for enhancing MANET security.

PMID:39755804 | DOI:10.1038/s41598-024-84421-0

NPJ Digit Med. 2025 Jan 4;8(1):7. doi: 10.1038/s41746-024-01364-6.

ABSTRACT

Adaptive deep brain stimulation (DBS) provides individualized therapy for people with Parkinson's disease (PWP) by adjusting the stimulation in real-time using neural signals that reflect their motor state. Current algorithms, however, utilize condensed and manually selected neural features which may result in a less robust and biased therapy. In this study, we propose Neural-to-Gait Neural network (N2GNet), a novel deep learning-based regression model capable of tracking real-time gait performance from subthalamic nucleus local field potentials (STN LFPs). The LFP data were acquired when eighteen PWP performed stepping in place, and the ground reaction forces were measured to track their weight shifts representing gait performance. By exhibiting a stronger correlation with weight shifts compared to the higher-correlation beta power from the two leads and outperforming other evaluated model designs, N2GNet effectively leverages a comprehensive frequency band, not limited to the beta range, to track gait performance solely from STN LFPs.

PMID:39755754 | DOI:10.1038/s41746-024-01364-6

Sci Rep. 2025 Jan 4;15(1):845. doi: 10.1038/s41598-024-81248-7.

ABSTRACT

This study presents a novel approach to identifying meters and their pointers in modern industrial scenarios using deep learning. We developed a neural network model that can detect gauges and one or more of their pointers on low-quality images. We use an encoder network, jump connections, and a modified Convolutional Block Attention Module (CBAM) to detect gauge panels and pointer keypoints in images. We also combine the output of the decoder network and the output of the improved CBAM as inputs to the Object Heatmap-Scalarmap Module to find pointer tip heat map peaks and predict pointer pointing. The method proposed in this paper is compared with several deep learning networks. The experimental results show that the model in this paper has the highest recognition correctness, with an average precision of 0.95 and 0.763 for Object Keypoint Similarity and Vector Direction Similarity, and an average recall of 0.951 and 0.856 in the test set, respectively, and achieves the best results in terms of efficiency and accuracy achieve the best trade-off, and performs well in recognizing multiple pointer targets. This demonstrates its robustness in real scenarios and provides a new idea for recognizing pointers in low-quality images more efficiently and accurately in complex industrial scenarios.

PMID:39755689 | DOI:10.1038/s41598-024-81248-7

Cancer Lett. 2025 Jan 2:217438. doi: 10.1016/j.canlet.2025.217438. Online ahead of print.

ABSTRACT

Preoperative detection of muscle-invasive bladder cancer (MIBC) remains a great challenge in practice. We aimed to develop and validate a deep Vesical Imaging Network (ViNet) model for the detection of MIBC using high-resolution T2-weighted MR imaging (hrT2WI) in a multicenter cohort. ViNet was designed using a modified 3D ResNet, in which, the encoder layers were pretrained using a self-supervised foundation model on over 40,000 cross-modal imaging datasets for transfer learning, and the classification modules were weakly supervised by an experiential knowledge-domain mask indicated by a nnUNet segmentation model. Optimal ViNet model was trained in derivation data (cohort 1, n = 312) and validated in multicenter data (cohort 2, n = 79; cohort 3, n = 44; cohort 4, n = 56) across a multi-ablation-test for model selection. In internal validation, ViNet using hrT2WI outperformed all ablation-test models (odds ratio [OR], 7.41 versus 1.85 - 2.70; all P < 0.05). In external validation, the performance of ViNet using hrT2WI versus ablation-test models was heterogeneous (OR, 1.31 - 3.89 versus 0.89 - 9.75; P = 0.03 - 0.15). In addition, clinical benefit of ViNet was evaluated between six readers using the Vesical Imaging-Reporting and Data System (VI-RADS) versus ViNet-adjusted VI-RADS. As a result, ViNet-adjusted VI-RADS upgraded 62.9% (17/27) of MIBC missed in VI-RADS score 2, while downgraded 84.1% (69/84), 62.5% (35/56) and 67.9% (19/28) of non-muscle-invasive bladder cancer (NMIBC) overestimated in VI-RADS score 3-5. We concluded that ViNet presents a promising alternative for diagnosing MIBC using hrT2WI instead of conventional multiparametric MRI.

PMID:39755362 | DOI:10.1016/j.canlet.2025.217438

Methods. 2025 Jan 2:S1046-2023(24)00283-4. doi: 10.1016/j.ymeth.2024.11.016. Online ahead of print.

ABSTRACT

Spatial transcriptomics has significantly advanced the measurement of spatial gene expression in the field of biology. However, the high cost of ST limits its application in large-scale studies. Using deep learning to predict spatial gene expression from H&E-stained histology images offers a more cost-effective alternative, but existing methods fail to fully leverage the multimodal information provided by Spatial transcriptomics and pathology images. In response, this paper proposes STMCL, a novel multimodal contrastive learning framework. STMCL integrates multimodal information, including histology images, gene expression features of spots, and their locations, to accurately infer spatial gene expression profiles. We tested four different types of multi-slice spatial transcriptomics datasets generated by the 10X Genomics platform. The results indicate that STMCL has advantages over baseline methods in predicting spatial gene expression profiles. Furthermore, STMCL is capable of capturing cancer-specific highly expressed genes and preserving gene expression patterns while maintaining the original spatial structure of gene expression. Our code is available at https://github.com/wenwenmin/STMCL.

PMID:39755346 | DOI:10.1016/j.ymeth.2024.11.016

J Neurosci Methods. 2025 Jan 2:110359. doi: 10.1016/j.jneumeth.2024.110359. Online ahead of print.

ABSTRACT

BACKGROUND: The hippocampus plays a crucial role in memory and is one of the first structures affected by Alzheimer's disease. Postmortem MRI offers a way to quantify the alterations by measuring the atrophy of the inner structures of the hippocampus. Unfortunately, the manual segmentation of hippocampal subregions required to carry out these measures is very time-consuming.

NEW METHOD: In this study, we explore the use of fully automated methods relying on state-of-the-art Deep Learning approaches to produce these annotations. More specifically, we propose a new segmentation framework made of a set of encoder-decoder blocks embedding self-attention mechanisms and atrous spatial pyramidal pooling to produce better maps of the hippocampus and identify four hippocampal regions: the dentate gyrus, the hippocampal head, the hippocampal body, and the hippocampal tail.

RESULTS: Trained using slices extracted from 15 postmortem T1-weighted, T2-weighted, and susceptibility-weighted MRI scans, our new approach produces hippocampus parcellations that are better aligned with the manually delineated parcellations provided by neuroradiologists.

COMPARISON WITH EXISTING METHODS: Four standard deep learning segmentation architectures: UNet, Double UNet, Attention UNet, and Multi-resolution UNet have been utilized for the qualitative and quantitative comparison of the proposed hippocampal region segmentation model.

CONCLUSIONS: Postmortem MRI serves as a highly valuable neuroimaging technique for examining the effects of neurodegenerative diseases on the intricate structures within the hippocampus. This study opens the way to large sample-size postmortem studies of the hippocampal substructures.

PMID:39755177 | DOI:10.1016/j.jneumeth.2024.110359

Ageing Res Rev. 2025 Jan 2:102651. doi: 10.1016/j.arr.2024.102651. Online ahead of print.

ABSTRACT

Parkinson's disease (PD) is one of the most incapacitating neurodegenerative diseases (NDDs). PD is the second most common NDD worldwide which affects approximately 1 to 2 percent of people over 65 years. It is an attractive pursuit for artificial intelligence (AI) to contribute to and evolve PD treatments through drug repositioning by repurposing existing drugs, shelved drugs, or even candidates that do not meet the criteria for clinical trials. A search was conducted in three databases Web of Science, Scopus, and PubMed. We reviewed the data related to the last years (1975-present) to identify those drugs currently being proposed for repositioning in PD. Moreover, we reviewed the present status of the computational approach, including AI/Machine Learning (AI/ML)-powered pharmaceutical discovery efforts and their implementation in PD treatment. It was found that the number of drug repositioning studies for PD has increased recently. Repositioning of drugs in PD is taking off, and scientific communities are increasingly interested in communicating its results and finding effective treatment alternatives for PD. A better chance of success in PD drug discovery has been made possible due to AI/ML algorithm advancements. In addition to the experimentation stage of drug discovery, it is also important to leverage AI in the planning stage of clinical trials to make them more effective. New AI-based models or solutions that increase the success rate of drug development are greatly needed.

PMID:39755176 | DOI:10.1016/j.arr.2024.102651

Med Phys. 2025 Jan 4. doi: 10.1002/mp.17609. Online ahead of print.

ABSTRACT

BACKGROUND: Respiratory motion during radiotherapy (RT) may reduce the therapeutic effect and increase the dose received by organs at risk. This can be addressed by real-time tracking, where respiration motion prediction is currently required to compensate for system latency in RT systems. Notably, for the prediction of future images in image-guided adaptive RT systems, the use of deep learning has been considered.

PURPOSE: This study proposed a modified generative adversarial network (GAN) for predicting cine-MR images in real time.

METHODS: Sagittal cine magnetic resonance (cine-MR) images of 15 patients with liver cancer who received RT were collected. The image series length of each patient was 300, and each series was divided into training, validation, and test sets. The datasets were further divided using a sliding window size of 10 and a stride of 1. A pix2pix GAN with the generator replaced by convolutional long short-term memory (ConvLSTM) was proposed herein. A five-frame cine-MR image series was inputted into the network, which predicted the next five frames. The proposed network was compared with three advanced networks: ConvLSTM, Eidetic 3D LSTM (E3D-LSTM), and SwinLSTM. Personalized models were trained for each patient. The peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), visual information fidelity (VIF), Pearson correlation coefficient (Pearson corr), and respiratory motion accuracy of the predicted images were used to evaluate the methods.

RESULTS: The proposed network demonstrated optimal performance in the four networks across various indicators. The proposed method provided better SSIM values than ConvLSTM at time steps 1, 2, 3, and 4, and outperformed E3DLSTM at all time steps. In terms of the VIF, the proposed method outperformed E3D-LSTM at all time steps and SwinLSTM at time steps 2, 3, 4, and 5. The proposed method was not significantly different from other methods in terms of Pearson correlation values except that it outperformed E3DLSTM at time step 1. In terms of the Pearson corr, the proposed method consistently achieves better values, especially in the high-frequency components. Low average landmark tracking errors were provided by the proposed method at time steps 4 and 5 (2.42 ± 0.91 and 2.44 ± 0.96 mm, respectively).

CONCLUSIONS: The GAN-ConvLSTM network can generate high-acutance real-time cine-MR images and predict respiratory motion with better accuracy.

PMID:39755123 | DOI:10.1002/mp.17609

Eur J Nucl Med Mol Imaging. 2025 Jan 4. doi: 10.1007/s00259-024-07043-8. Online ahead of print.

ABSTRACT

OBJECTIVE: This study aims to conduct a bibliometric analysis to explore research trends, collaboration patterns, and emerging themes in the PET/MR field based on published literature from 2010 to 2024.

METHODS: A detailed literature search was performed using the Web of Science Core Collection (WoSCC) database with keywords related to PET/MR. A total of 4,349 publications were retrieved and analyzed using various bibliometric tools, including VOSviewer and CiteSpace.

RESULTS: The analysis revealed an initial increase in PET/MR publications, peaking at 495 in 2021, followed by a slight decline. The USA, Germany, and China were the most prolific countries, with the USA demonstrating strong collaborative networks. Key institutions included the Stanford University, Technical University of Munich and University of Duisburg-Essen. Prominent authors were primarily from Germany, with significant contributions from University Hospital Essen. Major journals in the field included the European Journal of Nuclear Medicine, Journal of Nuclear Medicine, and Physics in Medicine and Biology. Emerging research areas focused on oncology, neurological disorders, and cardiovascular diseases, with keywords such as "prostate cancer," "Alzheimer's disease," and "breast cancer" showing high research activity. Recent trends also highlight the growing integration of AI, particularly deep learning, to improve imaging reconstruction and diagnostic accuracy.

CONCLUSION: The findings emphasize the need for continuous investment, strategic planning, and technological innovations to expand PET/MR's clinical applications. Future research should focus on optimizing imaging techniques, fostering international collaborations, and integrating emerging technologies like artificial intelligence to enhance PET/MR's diagnostic and therapeutic potential in precision medicine.

PMID:39754665 | DOI:10.1007/s00259-024-07043-8

Nano Lett. 2025 Jan 4. doi: 10.1021/acs.nanolett.4c05120. Online ahead of print.

ABSTRACT

Topological plasmonics combines principles of topology and plasmonics to provide new methods for controlling light, analogous to topological edge states in photonics. However, designing such topological states remains challenging due to the complexity of the high-dimensional design space. We present a novel method that uses supervised, physics-informed deep learning and surrogate modeling to design topological devices for desired wavelengths. By embedding physical constraints in the neural network's training, our model efficiently explores the design space, significantly reducing simulation time. Additionally, we use nonplanar wavefront excitations to probe topologically protected plasmonic modes, making the design and training process nonlinear. Using this approach, we design a topological device with unidirectional edge modes in a ring resonator at specific operational frequencies. Our method reduces computational cost and time while maintaining high accuracy, highlighting the potential of combining machine learning and advanced techniques for photonic device innovation.

PMID:39754588 | DOI:10.1021/acs.nanolett.4c05120

Microb Biotechnol. 2025 Jan;18(1):e70072. doi: 10.1111/1751-7915.70072.

ABSTRACT

Antimicrobial peptides (AMPs) are promising candidates to combat multidrug-resistant pathogens. However, the high cost of extensive wet-lab screening has made AI methods for identifying and designing AMPs increasingly important, with machine learning (ML) techniques playing a crucial role. AI approaches have recently revolutionised this field by accelerating the discovery of new peptides with anti-infective activity, particularly in preclinical mouse models. Initially, classical ML approaches dominated the field, but recently there has been a shift towards deep learning (DL) models. Despite significant contributions, existing reviews have not thoroughly explored the potential of large language models (LLMs), graph neural networks (GNNs) and structure-guided AMP discovery and design. This review aims to fill that gap by providing a comprehensive overview of the latest advancements, challenges and opportunities in using AI methods, with a particular emphasis on LLMs, GNNs and structure-guided design. We discuss the limitations of current approaches and highlight the most relevant topics to address in the coming years for AMP discovery and design.

PMID:39754551 | DOI:10.1111/1751-7915.70072

Anal Chem. 2025 Jan 4. doi: 10.1021/acs.analchem.4c04448. Online ahead of print.

ABSTRACT

Cyclic voltammetry (CV) has been a powerful technique to provide impactful insights for electrochemical systems, including reaction mechanism, kinetics, diffusion coefficients, etc., in various fields of study, notably energy storage and energy conversion. However, the separation between the faradaic current component of CV and the nonfaradaic current contribution to extract useful information remains a major issue for researchers. Herein, we report a deep learning algorithm inspired by speech denoising that utilizes the theoretical faradaic current as a study target and predicts it from the overall current response from cyclic voltammograms. This deep neural network (DNN) is constructed from a series of fully connected layers, which apply a weight matrix to the inputs and transform it using an activation function to obtain the desired regression. Our model performed well with overall mean absolute percentage errors (MAPEs) of 6.36% between theoretical faradaic currents and the predicted responses from the total currents, with a peak position difference of 2.56 mV for anodic peaks and 2.44 mV for cathodic ones. Furthermore, the algorithm is also capable of extracting peak current values from experimental data with 3.37% MAPE and minimal peak position error (less than 0.75 mV). This innovative approach may be used as a tool to assist researchers in studying electrochemical systems using CV.

PMID:39754543 | DOI:10.1021/acs.analchem.4c04448

Biomed Tech (Berl). 2025 Jan 6. doi: 10.1515/bmt-2024-0299. Online ahead of print.

ABSTRACT

OBJECTIVES: Diabetic retinopathy (DR) is associated with long-term diabetes and is a leading cause of blindness if it is not diagnosed early. The rapid growth of deep learning eases the clinicians' DR diagnosing procedure. It automatically extracts the features and performs the grading. However, training the image toward the majority of background pixels can impact the accuracy and efficiency of grading tasks. This paper proposes an auto-thresholding algorithm that reduces the negative impact of considering the background pixels for feature extraction which highly affects the grading process.

METHODS: The PSO-based thresholding algorithm for retinal segmentation is proposed in this paper, and its efficacy is evaluated against the Otsu, histogram-based sigma, and entropy algorithms. In addition, the importance of retinal segmentation is analyzed using Explainable AI (XAI) to understand how each feature impacts the model's performance. For evaluating the accuracy of the grading, ResNet50 was employed.

RESULTS: The experiments were conducted using the IDRiD fundus dataset. Despite the limited data, the retinal segmentation approach provides significant accuracy than the non-segmented approach, with a substantial accuracy of 83.70 % on unseen data.

CONCLUSIONS: The result shows that the proposed PSO-based approach helps automatically determine the threshold value and improves the model's accuracy.

PMID:39754503 | DOI:10.1515/bmt-2024-0299

Orthod Craniofac Res. 2025 Jan 4. doi: 10.1111/ocr.12895. Online ahead of print.

ABSTRACT

OBJECTIVE: This retrospective study aimed to evaluate morphometric changes in mandibular condyles of patients with skeletal Class III malocclusion following two-jaw orthognathic surgery planned using virtual surgical planning (VSP) and analysed with automated three-dimensional (3D) image analysis based on deep-learning techniques.

MATERIALS AND METHODS: Pre-operative (T1) and 12-18 months post-operative (T2) Cone-Beam Computed Tomography (CBCT) scans of 17 patients (mean age: 24.8 ± 3.5 years) were analysed using 3DSlicer software. Deep-learning algorithms automated CBCT orientation, registration, bone segmentation, and landmark identification. By utilising voxel-based superimposition of pre- and post-operative CBCT scans and shape correspondence, the overall changes in condylar morphology were assessed, with a focus on bone resorption and apposition at specific regions (superior, lateral and medial poles). The correlation between these modifications and the extent of actual condylar movements post-surgery was investigated. Statistical analysis was conducted with a significance level of α = 0.05.

RESULTS: Overall condylar remodelling was minimal, with mean changes of < 1 mm. Small but statistically significant bone resorption occurred at the condylar superior articular surface, while bone apposition was primarily observed at the lateral pole. The bone apposition at the lateral pole and resorption at the superior articular surface were significantly correlated with medial condylar displacement (p < 0.05).

CONCLUSION: The automated 3D analysis revealed distinct patterns of condylar remodelling following orthognathic surgery in skeletal Class III patients, with minimal overall changes but significant regional variations. The correlation between condylar displacements and remodelling patterns highlights the need for precise pre-operative planning to optimise condylar positioning, potentially minimising harmful remodelling and enhancing stability.

PMID:39754473 | DOI:10.1111/ocr.12895

NPJ Precis Oncol. 2025 Jan 3;9(1):2. doi: 10.1038/s41698-024-00789-2.

ABSTRACT

Recent progress in deep learning (DL) is producing a new generation of tools across numerous clinical applications. Within the analysis of brain tumors in magnetic resonance imaging, DL finds applications in tumor segmentation, quantification, and classification. It facilitates objective and reproducible measurements crucial for diagnosis, treatment planning, and disease monitoring. Furthermore, it holds the potential to pave the way for personalized medicine through the prediction of tumor type, grade, genetic mutations, and patient survival outcomes. In this review, we explore the transformative potential of DL for brain tumor care and discuss existing applications, limitations, and future directions and opportunities.

PMID:39753730 | DOI:10.1038/s41698-024-00789-2

Sci Rep. 2025 Jan 4;15(1):742. doi: 10.1038/s41598-024-84992-y.

ABSTRACT

Event-based surveillance is crucial for the early detection and rapid response to potential public health risks. In recent years, social networking services (SNS) have been recognized for their potential role in this domain. Previous studies have demonstrated the capacity of SNS posts for the early detection of health crises and affected individuals, including those related to infectious diseases. However, the reliability of such posts, being subjective and not clinically diagnosed, remains a challenge. In this study, we address this issue by assessing the classification performance of transformer-based pretrained language models to accurately classify Japanese tweets related to heat stroke, a significant health effect of climate change, as true or false. We also evaluated the efficacy of combining SNS and artificial intelligence for event-based public health surveillance by visualizing the data on correctly classified tweets and heat stroke emergency medical evacuees in time-space and animated video, respectively. The transformer-based pretrained language models exhibited good performance in classifying the tweets. Spatiotemporal and animated video visualizations revealed a reasonable correlation. This study demonstrates the potential of using Japanese tweets and deep learning algorithms based on transformer networks for event-based surveillance at high spatiotemporal levels to enable early detection of heat stroke risks.

PMID:39753702 | DOI:10.1038/s41598-024-84992-y

Sci Rep. 2025 Jan 3;15(1):675. doi: 10.1038/s41598-024-83944-w.

ABSTRACT

Land Surface Temperature (LST) is widely recognized as a sensitive indicator of climate change, and it plays a significant role in ecological research. The ERA5-Land LST dataset, developed and managed by the European Centre for Medium-Range Weather Forecasts (ECMWF), is extensively used for global or regional LST studies. However, its fine-scale application is limited by its low spatial resolution. Therefore, to improve the spatial resolution of ERA5-Land LST data, this study proposes an Attention Mechanism U-Net (AMUN) method, which combines data acquisition and preprocessing on the Google Earth Engine (GEE) cloud computing platform, to downscale the hourly monthly mean reanalysis LST data of ERA5-Land across China's territory from 0.1° to 0.01°. This method comprehensively considers the relationship between the LST and surface features, organically combining multiple deep learning modules, includes the Global Multi-Factor Cross-Attention (GMFCA) module, the Feature Fusion Residual Dense Block (FFRDB) connection module, and the U-Net module. In addition, the Bayesian global optimization algorithm is used to select the optimal hyperparameters of the network in order to enhance the predictive performance of the model. Finally, the downscaling accuracy of the network was evaluated through simulated data experiments and real data experiments and compared with the Random Forest (RF) method. The results show that the network proposed in this study outperforms the RF method, with RMSE reduced by approximately 32-51%. The downscaling method proposed in this study can effectively improve the accuracy of ERA5-Land LST downscaling, providing new insights for LST downscaling research.

PMID:39753651 | DOI:10.1038/s41598-024-83944-w

Nat Commun. 2025 Jan 2;16(1):349. doi: 10.1038/s41467-024-55228-4.

ABSTRACT

Identifying transitional states is crucial for understanding protein conformational changes that underlie numerous biological processes. Markov state models (MSMs), built from Molecular Dynamics (MD) simulations, capture these dynamics through transitions among metastable conformational states, and have demonstrated success in studying protein conformational changes. However, MSMs face challenges in identifying transition states, as they partition MD conformations into discrete metastable states (or free energy minima), lacking description of transition states located at the free energy barriers. Here, we introduce Transition State identification via Dispersion and vAriational principle Regularized neural networks (TS-DAR), a deep learning framework inspired by out-of-distribution (OOD) detection in trustworthy artificial intelligence (AI). TS-DAR offers an end-to-end pipeline that can simultaneously detect all transition states between multiple free minima from MD simulations using the regularized hyperspherical embeddings in latent space. The key insight of TS-DAR lies in treating transition state structures as OOD data, recognizing that they are sparsely populated and exhibit a distributional shift from metastable states. We demonstrate the power of TS-DAR by applying it to a 2D potential, alanine dipeptide, and the translocation of a DNA motor protein on DNA, where it outperforms previous methods in identifying transition states.

PMID:39753544 | DOI:10.1038/s41467-024-55228-4

Acad Radiol. 2025 Jan 2:S1076-6332(24)00997-8. doi: 10.1016/j.acra.2024.12.029. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: Accurate preoperative pathological staging of gastric cancer is crucial for optimal treatment selection and improved patient outcomes. Traditional imaging methods such as CT and endoscopy have limitations in staging accuracy.

METHODS: This retrospective study included 691 gastric cancer patients treated from March 2017 to March 2024. Enhanced venous-phase CT and endoscopic images, along with postoperative pathological results, were collected. We developed three modeling approaches: (1) nine deep learning models applied to CT images (DeepCT), (2) 11 machine learning algorithms using handcrafted radiomic features from CT images (HandcraftedCT), and (3) ResNet-50-extracted deep features from endoscopic images followed by 11 machine learning algorithms (DeepEndo). The two top-performing models from each approach were combined into the Integrated Multi-Modal Model using a stacking ensemble method. Performance was assessed using ROC-AUC, sensitivity, and specificity.

RESULTS: The Integrated Multi-Modal Model achieved an ROC-AUC of 0.933 (95% CI, 0.887-0.979) on the test set, outperforming individual models. Sensitivity and specificity were 0.869 and 0.840, respectively. Various evaluation metrics demonstrated that the final fusion model effectively integrated the strengths of each sub-model, resulting in a balanced and robust performance with reduced false-positive and false-negative rates.

CONCLUSION: The Integrated Multi-Modal Model effectively integrates radiomic and deep learning features from CT and endoscopic images, demonstrating superior performance in preoperative pathological staging of gastric cancer. This multimodal approach enhances predictive accuracy and provides a reliable tool for clinicians to develop individualized treatment plans, thereby improving patient outcomes.

DATA AVAILABILITY: The data presented in this study are available on request from the corresponding author. The data are not publicly available due to ethical reasons. All code used in this study is based on third-party libraries and all custom code developed for this study is available upon reasonable request from the corresponding author.

PMID:39753481 | DOI:10.1016/j.acra.2024.12.029