Artif Intell Med. 2025 Mar 5;163:103093. doi: 10.1016/j.artmed.2025.103093. Online ahead of print.

ABSTRACT

Electrocardiographic imaging (ECGI) aims to noninvasively estimate heart surface potentials starting from body surface potentials. This is classically based on geometric information on the torso and the heart from imaging, which complicates clinical application. In this study, we aim to develop a deep learning framework to estimate heart surface potentials solely from body surface potentials, enabling wider clinical use. The framework introduces two main components: the transformation of 3D torso and heart geometries into standard 2D representations, and the development of a customized deep learning network model. The 2D torso and heart representations maintain a consistent layout across different subjects, making the proposed framework applicable to different torso-heart geometries. With spatial information incorporated in the 2D representations, the torso-heart physiological relationship can be learnt by the network. The deep learning model is based on a Pix2Pix network, adapted to work with 2.5D data in our task, i.e., 2D body surface potential maps (BSPMs) and 2D heart surface potential maps (HSPMs) with time sequential information. We propose a new loss function tailored to this specific task, which uses a cosine similarity and different weights for different inputs. BSPMs and HSPMs from 11 healthy subjects (8 females and 3 males) and 29 idiopathic ventricular fibrillation (IVF) patients (11 females and 18 males) were used in this study. Performance was assessed on a test set by measuring the similarity and error between the output of the proposed model and the solution provided by mainstream ECGI, by comparing HSPMs, the concatenated electrograms (EGMs), and the estimated activation time (AT) and recovery time (RT). The mean of the mean absolute error (MAE) for the HSPMs was 0.012 ± 0.011, and the mean of the corresponding structural similarity index measure (SSIM) was 0.984 ± 0.026. The mean of the MAE for the EGMs was 0.004 ± 0.004, and the mean of the corresponding Pearson correlation coefficient (PCC) was 0.643 ± 0.352. Results suggest that the model is able to precisely capture the structural and temporal characteristics of the HSPMs. The mean of the absolute time differences between estimated and reference activation times was 6.048 ± 5.188 ms, and the mean of the absolute differences for recovery times was 18.768 ± 17.299 ms. Overall, results show similar performance between the proposed model and standard ECGI, exhibiting low error and consistent clinical patterns, without the need for CT/MRI. The model shows to be effective across diverse torso-heart geometries, and it successfully integrates temporal information in the input. This in turn suggests the possible use of this model in cost effective clinical scenarios like patient screening or post-operative follow-up.

PMID:40073713 | DOI:10.1016/j.artmed.2025.103093

Comput Biol Chem. 2025 Mar 6;117:108416. doi: 10.1016/j.compbiolchem.2025.108416. Online ahead of print.

ABSTRACT

Drug-target affinity prediction is a fundamental task in the field of drug discovery. Extracting and integrating structural information from proteins effectively is crucial to enhance the accuracy and generalization of prediction, which remains a substantial challenge. This paper proposes a pocket-based multimodal deep learning model named PocketDTA for drug-target affinity prediction, based on the principle of "structure determines function". PocketDTA introduces the pocket graph structure that encodes protein residue features pretrained using a biological language model as nodes, while edges represent different protein sequences and spatial distances. This approach overcomes the limitations of lack of spatial information in traditional prediction models with only protein sequence input. Furthermore, PocketDTA employs relational graph convolutional networks at both atomic and residue levels to extract structural features from drugs and proteins. By integrating multimodal information through deep neural networks, PocketDTA combines sequence and structural data to improve affinity prediction accuracy. Experimental results demonstrate that PocketDTA outperforms state-of-the-art prediction models across multiple benchmark datasets by showing strong generalization under more realistic data splits and confirming the effectiveness of pocket-based methods for affinity prediction.

PMID:40073710 | DOI:10.1016/j.compbiolchem.2025.108416

Spectrochim Acta A Mol Biomol Spectrosc. 2025 Mar 6;335:125997. doi: 10.1016/j.saa.2025.125997. Online ahead of print.

ABSTRACT

Lung cancer is a malignant tumor that poses a serious threat to human health. Existing lung cancer diagnostic techniques face the challenges of high cost and slow diagnosis. Early and rapid diagnosis and treatment are essential to improve the outcome of lung cancer. In this study, a deep learning-based multi-modal spectral information fusion (MSIF) network is proposed for lung adenocarcinoma cell detection. First, multi-modal data of Fourier transform infrared spectra, UV-vis absorbance spectra, and fluorescence spectra of normal and patient cells were collected. Subsequently, the spectral text data were efficiently processed by one-dimensional convolutional neural network. The global and local features of the spectral images are deeply mined by the hybrid model of ResNet and Transformer. An adaptive depth-wise convolution (ADConv) is introduced to be applied to feature extraction, overcoming the shortcomings of conventional convolution. In order to achieve feature learning between multi-modalities, a cross-modal interaction fusion (CMIF) module is designed. This module fuses the extracted spectral image and text features in a multi-faceted interaction, enabling full utilization of multi-modal features through feature sharing. The method demonstrated excellent performance on the test sets of Fourier transform infrared spectra, UV-vis absorbance spectra and fluorescence spectra, achieving 95.83 %, 97.92 % and 100 % accuracy, respectively. In addition, experiments validate the superiority of multi-modal spectral data and the robustness of the model generalization capability. This study not only provides strong technical support for the early diagnosis of lung cancer, but also opens a new chapter for the application of multi-modal data fusion in spectroscopy.

PMID:40073660 | DOI:10.1016/j.saa.2025.125997

Food Chem. 2025 Mar 7;478:143671. doi: 10.1016/j.foodchem.2025.143671. Online ahead of print.

ABSTRACT

Online detection of internal quality of strawberries presents challenges particularly concerning fruit damage, detection accuracy, and processing efficiency. This study explores the feasibility of using Vis/NIRS for online detection of SSC in strawberries during hanging transportation. After analyzing SSC distribution in strawberries, an optical sensing system was developed, and optimal configurations were identified using PLSR models. When employing a horizontal optical beam through the strawberry center, the PLSR model combined with SNV preprocessing and CARS feature selection achieved the best conventional chemometric results (RPD of 4.793). Additionally, three 1D-CNN approaches were investigated, with the 1D-CNN-LSTM method exhibiting superior performance (Rp2 of 0.963, RMSEP of 0.209°Brix, RPD of 5.332). These findings demonstrate the excellent capability of our developed system, enhanced by deep learning methods, for online detection of SSC in strawberries. This work may open new avenues for the online assessment of internal quality in small and delicate fruits.

PMID:40073605 | DOI:10.1016/j.foodchem.2025.143671

Med Image Anal. 2025 Mar 6;102:103529. doi: 10.1016/j.media.2025.103529. Online ahead of print.

ABSTRACT

This study addresses the challenges of confounding effects and interpretability in artificial-intelligence-based medical image analysis. Whereas existing literature often resolves confounding by removing confounder-related information from latent representations, this strategy risks affecting image reconstruction quality in generative models, thus limiting their applicability in feature visualization. To tackle this, we propose a different strategy that retains confounder-related information in latent representations while finding an alternative confounder-free representation of the image data. Our approach views the latent space of an autoencoder as a vector space, where imaging-related variables, such as the learning target (t) and confounder (c), have a vector capturing their variability. The confounding problem is addressed by searching a confounder-free vector which is orthogonal to the confounder-related vector but maximally collinear to the target-related vector. To achieve this, we introduce a novel correlation-based loss that not only performs vector searching in the latent space, but also encourages the encoder to generate latent representations linearly correlated with the variables. Subsequently, we interpret the confounder-free representation by sampling and reconstructing images along the confounder-free vector. The efficacy and flexibility of our proposed method are demonstrated across three applications, accommodating multiple confounders and utilizing diverse image modalities. Results affirm the method's effectiveness in reducing confounder influences, preventing wrong or misleading associations, and offering a unique visual interpretation for in-depth investigations by clinical and epidemiological researchers. The code is released in the following GitLab repository: https://gitlab.com/radiology/compopbio/ai_based_association_analysis.

PMID:40073582 | DOI:10.1016/j.media.2025.103529

Comput Biol Med. 2025 Mar 11;189:109956. doi: 10.1016/j.compbiomed.2025.109956. Online ahead of print.

ABSTRACT

Among various post-translational modifications (PTMs), predicting C-linked and S-linked glycosites is an essential task, yet experimental techniques such as Capillary Electrophoresis (CE), Enzymatic Deglycosylation, and Mass Spectrometry (MS) are expensive. Therefore, computational techniques are required to predict these glycosites. Here, different language model embeddings and sequential features were explored. Two separate feature selection methods: Recursive Feature Elimination (RFE) and Particle Swarm Optimization (PSO) were employed and utilized for identifying the optimal feature set. Cross-validation results were generated for choosing the final models. Three sampling strategies to handle imbalanced datasets were examined: Random undersampling, Synthetic Minority Over-sampling Technique (SMOTE) and Adaptive Synthetic Sampling Approach for Imbalanced Learning (ADASYN). In this study, two models: DeepCSEmbed-C and DeepCSEmbed-S are proposed for C-linked and S-linked glycosylation prediction respectively. DeepCSEmbed-C is a dual-branch deep learning model comprising a Feedforward Neural Network (FNN) branch and an Inception branch, coupled with a Random undersampling strategy. DeepCSEmbed-S is a Categorical Boosting (CAT) model with the SMOTE oversampling strategy. DeepCSEmbed-C outperformed available state-of-the-art (SOTA) methods, achieving 92.9% sensitivity, 95.1% F1-score and 90.6% MCC on the Independent dataset. Datasets and python scripts for training and testing the models are provided and made freely accessible at https://github.com/nafcoder/DeepCSEmbed.

PMID:40073495 | DOI:10.1016/j.compbiomed.2025.109956

Comput Biol Med. 2025 Mar 11;189:109896. doi: 10.1016/j.compbiomed.2025.109896. Online ahead of print.

ABSTRACT

With the ongoing advancement of digital technology, oral medicine transitions from traditional diagnostics to computer-assisted diagnosis and treatment. Identifying dental implants in patients without records is complex and time-consuming. Accurate identification of dental implants is crucial for ensuring the sustainability and reliability of implant treatment, particularly in cases where patients lack available medical records. In this paper, we propose a multi-task fine-grained CBCT dental implant classification and segmentation method using deep learning, called MFPT-Net.This method, based on progressive training with multiscale feature extraction and enhancement, can differentiate minor implant features and similar features that are easily confused, such as implant threads. It addresses the problem of large intra-class differences and small inter-class differences of implants, achieving automatic, synchronized classification and segmentation of implant systems in CBCT images. In this paper, 437 CBCT sequences with 723 dental implants, acquired from three different centers, are included in our dataset. This dataset is the first instance of utilizing such a comprehensive collection of data for CBCT analysis. Our method achieved a satisfying classification result with accuracy of 92.98%, average precision of 93.15%, average recall of 93.31%, and average F1 score of 93.18%, which exceeded the second-best model by nearly 10%. Moreover, our segmentation Dice similarity coefficient reached 98.04%, which is significantly better than the current state-of-the-art method. External clinical validation with 252 implants proved our model's clinical feasibility. The result demonstrates that our proposed method could assist dentists with dental implant classification and segmentation in CBCT images, enhancing efficiency and accuracy in clinical practice.

PMID:40073494 | DOI:10.1016/j.compbiomed.2025.109896

J Neural Eng. 2025 Mar 12. doi: 10.1088/1741-2552/adbfc1. Online ahead of print.

ABSTRACT

OBJECTIVE: In the field of brain-computer interface (BCI), achieving high information transfer rates (ITR) with a large number of targets remains a challenge. This study aims to address this issue by developing a novel code-modulated visual evoked potential (c-VEP) BCI system capable of handling an extensive instruction set while maintaining high performance.

METHOD: We propose a c-VEP BCI system that employs narrow-band random sequences as visual stimuli and utilizes a convolutional neural network (CNN)-based EEG2Code decoding algorithm. This algorithm predicts corresponding stimulus sequences from EEG data and achieves efficient and accurate classification.

MAIN RESULTS: Offline experiments which conducted in a sequential paradigm, resulted in an average accuracy of 87.66% and a simulated ITR of 260.14 bits/min. In online experiments, the system demonstrated an accuracy of 76.27% and an ITR of 213.80 bits/min in a cued spelling task.

SIGNIFICANCE: This work represents an advancement in c-VEP BCI systems, offering one of the largest known instruction set in VEP-based BCIs and demonstrating robust performance metrics. The proposed system is potential for more practical and efficient BCI applications.

PMID:40073451 | DOI:10.1088/1741-2552/adbfc1

Br J Radiol. 2025 Mar 12:tqaf047. doi: 10.1093/bjr/tqaf047. Online ahead of print.

ABSTRACT

OBJECTIVES: To develop a deep learning (DL) model based on ultrasound (US) images of lymph nodes for predicting cervical lymph node metastasis (CLNM) in postoperative patients with differentiated thyroid carcinoma (DTC).

METHODS: Retrospective collection of 352 lymph nodes from 330 patients with cytopathology findings between June 2021 and December 2023 at our institution. The database was randomly divided into the training and test cohort at an 8:2 ratio. The DL basic model of longitudinal and cross-sectional of lymph nodes was constructed based on ResNet50 respectively, and the results of the two basic models were fused (1:1) to construct a longitudinal + cross-sectional DL model. Univariate and multivariate analysis were used to assess US features and construct a conventional US model. Subsequently, a combined model was constructed by integrating DL and US.

RESULTS: The diagnostic accuracy of the longitudinal + cross-sectional DL model was higher than that of longitudinal or cross-sectional alone. The AUC of the combined model (US＋DL) was 0.855 (95%CI: 0.767-0.942), and the accuracy, sensitivity and specificity were 0.786 (95%CI: 0.671-0.875), 0.972 (95%CI: 0.855-0.999) and 0.588 (95%CI: 0.407-0.754), respectively. Compared with US and DL models, the IDI and NRI of the combined model are both positive.

CONCLUSIONS: This study preliminary shows that the DL model based on US images of lymph nodes has a high diagnostic efficacy for predicting CLNM in postoperative patients with DTC, and the combined model of US+DL is superior to single conventional US and DL for predicting CLNM in this population.

ADVANCES IN KNOWLEDGE: We innovatively used DL of lymph node US images to predict the status of cervical lymph nodes in postoperative patients with DTC.

PMID:40073229 | DOI:10.1093/bjr/tqaf047

Sci Adv. 2025 Mar 14;11(11):eadr3757. doi: 10.1126/sciadv.adr3757. Epub 2025 Mar 12.

ABSTRACT

Brain age gap (BAG), the deviation between estimated brain age and chronological age, is a promising marker of brain health. However, the genetic architecture and reliable targets for brain aging remains poorly understood. In this study, we estimate magnetic resonance imaging (MRI)-based brain age using deep learning models trained on the UK Biobank and validated with three external datasets. A genome-wide association study for BAG identified two unreported loci and seven previously reported loci. By integrating Mendelian Randomization (MR) and colocalization analysis on eQTL and pQTL data, we prioritized seven genetically supported druggable genes, including MAPT, TNFSF12, GZMB, SIRPB1, GNLY, NMB, and C1RL, as promising targets for brain aging. We rediscovered 13 potential drugs with evidence from clinical trials of aging and prioritized several drugs with strong genetic support. Our study provides insights into the genetic basis of brain aging, potentially facilitating drug development for brain aging to extend the health span.

PMID:40073132 | DOI:10.1126/sciadv.adr3757

Sci Adv. 2025 Mar 14;11(11):eadq8131. doi: 10.1126/sciadv.adq8131. Epub 2025 Mar 12.

ABSTRACT

There is great interest in using genetically tractable organisms such as Drosophila to gain insights into the regulation and function of sleep. However, sleep phenotyping in Drosophila has largely relied on simple measures of locomotor inactivity. Here, we present FlyVISTA, a machine learning platform to perform deep phenotyping of sleep in flies. This platform comprises a high-resolution closed-loop video imaging system, coupled with a deep learning network to annotate 35 body parts, and a computational pipeline to extract behaviors from high-dimensional data. FlyVISTA reveals the distinct spatiotemporal dynamics of sleep and wake-associated microbehaviors at baseline, following administration of the sleep-inducing drug gaboxadol, and with dorsal fan-shaped body drivers. We identify a microbehavior ("haltere switch") exclusively seen during quiescence that indicates a deeper sleep stage. These results enable the rigorous analysis of sleep in Drosophila and set the stage for computational analyses of microbehaviors in quiescent animals.

PMID:40073129 | DOI:10.1126/sciadv.adq8131

Biomacromolecules. 2025 Mar 12. doi: 10.1021/acs.biomac.4c01124. Online ahead of print.

ABSTRACT

We use a combination of Brownian dynamics (BD) simulation results and deep learning (DL) strategies for the rapid identification of large structural changes caused by missense mutations in intrinsically disordered proteins (IDPs). We used ∼6500 IDP sequences from MobiDB database of length 20-300 to obtain gyration radii from BD simulation on a coarse-grained single-bead amino acid model (HPS2 model) used by us and others [Dignon, G. L. PLoS Comput. Biol. 2018, 14, e1005941,Tesei, G. Proc. Natl. Acad. Sci. U.S.A. 2021, 118, e2111696118,Seth, S. J. Chem. Phys. 2024, 160, 014902] to generate the training sets for the DL algorithm. Using the gyration radii ⟨Rg⟩ of the simulated IDPs as the training set, we develop a multilayer perceptron neural net (NN) architecture that predicts the gyration radii of 33 IDPs previously studied by using BD simulation with 97% accuracy from the sequence and the corresponding parameters from the HPS model. We now utilize this NN to predict gyration radii of every permutation of missense mutations in IDPs. Our approach successfully identifies mutation-prone regions that induce significant alterations in the radius of gyration when compared to the wild-type IDP sequence. We further validate the prediction by running BD simulations on the subset of identified mutants. The neural network yields a (104-106)-fold faster computation in the search space for potentially harmful mutations. Our findings have substantial implications for rapid identification and understanding of diseases related to missense mutations in IDPs and for the development of potential therapeutic interventions. The method can be extended to accurate predictions of other mutation effects in disordered proteins.

PMID:40072940 | DOI:10.1021/acs.biomac.4c01124

Gigascience. 2025 Jan 6;14:giae098. doi: 10.1093/gigascience/giae098.

ABSTRACT

Magnetic resonance imaging (MRI) preprocessing is a critical step for neuroimaging analysis. However, the computational cost of MRI preprocessing pipelines is a major bottleneck for large cohort studies and some clinical applications. While high-performance computing and, more recently, deep learning have been adopted to accelerate the computations, these techniques require costly hardware and are not accessible to all researchers. Therefore, it is important to understand the performance bottlenecks of MRI preprocessing pipelines to improve their performance. Using the Intel VTune profiler, we characterized the bottlenecks of several commonly used MRI preprocessing pipelines from the Advanced Normalization Tools (ANTs), FMRIB Software Library, and FreeSurfer toolboxes. We found few functions contributed to most of the CPU time and that linear interpolation was the largest contributor. Data access was also a substantial bottleneck. We identified a bug in the Insight Segmentation and Registration Toolkit library that impacts the performance of the ANTs pipeline in single precision and a potential issue with the OpenMP scaling in FreeSurfer recon-all. Our results provide a reference for future efforts to optimize MRI preprocessing pipelines.

PMID:40072903 | DOI:10.1093/gigascience/giae098

Phys Chem Chem Phys. 2025 Mar 12. doi: 10.1039/d4cp04579c. Online ahead of print.

ABSTRACT

The phosphorylation of residue T177 produces a significant effect on the conformational dynamics of CDK6. Gaussian accelerated molecular dynamics (GaMD) simulations followed by deep learning (DL) are applied to explore the molecular mechanism of the phosphorylation-mediated effect on the conformational dynamics of CDK6 bound by three inhibitors 6ZV, 6ZZ and 0RS, in which 6ZV and 6ZZ have been used to test clinical performance. The DL finds that the β-sheets, αC helix as well as the T-loop are involved in obvious differences of conformation contacts and suggests that the T-loop plays a key role in the function of CDK6. The analyses of free energy landscapes (FELs) reveal that the phosphorylation of T177 leads to alterations of the T-loop conformation and the results from principal component analysis (PCA) indicate that the phosphorylation affects the fluctuation behavior of the β-sheets and the T-loop in CDK6. Interaction networks of inhibitors with CDK6 were analyzed and the information reveals that 6ZV contributes more hydrogen binding interactions (HBIs) and hot interaction spots with CDK6. Our MM-GBSA calculations suggest that the binding ability of 6ZV to CDK6 is stronger than 6ZZ and 0RS. We anticipate that this work could provide useful information for further understanding of CDK6 function and developing new promising inhibitors targeting CDK6.

PMID:40072875 | DOI:10.1039/d4cp04579c

IEEE Trans Biomed Eng. 2025 Mar 12;PP. doi: 10.1109/TBME.2025.3541643. Online ahead of print.

ABSTRACT

OBJECTIVE: To enable fast and stable neonatal brain MR imaging by integrating learned neonate-specific subspace model and model-driven deep learning.

METHODS: Fast data acquisition is critical for neonatal brain MRI, and deep learning has emerged as an effective tool to accelerate existing fast MRI methods by leveraging prior image information. However, deep learning often requires large amounts of training data to ensure stable image reconstruction, which is not currently available for neonatal MRI applications. In this work, we addressed this problem by utilizing a subspace model-assisted deep learning approach. Specifically, we used a subspace model to capture the spatial features of neonatal brain images. The learned neonate-specific subspace was then integrated with a deep network to reconstruct high-quality neonatal brain images from very sparse k-space data.

RESULTS: The effectiveness and robustness of the proposed method were validated using both the dHCP dataset and testing data from four independent medical centers, yielding very encouraging results. The stability of the proposed method has been confirmed with different perturbations, all showing remarkably stable reconstruction performance. The flexibility of the learned subspace was also shown when combined with other deep neural networks, yielding improved image reconstruction performance.

CONCLUSION: Fast and stable neonatal brain MR imaging can be achieved using subspace-assisted deep learning with sparse sampling. With further development, the proposed method may improve the practical utility of MRI in neonatal imaging applications.

PMID:40072865 | DOI:10.1109/TBME.2025.3541643

IEEE Trans Vis Comput Graph. 2025 Mar 12;PP. doi: 10.1109/TVCG.2025.3549147. Online ahead of print.

ABSTRACT

In Augmented Reality (AR), virtual content enhances user experience by providing additional information. However, improperly positioned or designed virtual content can be detrimental to task performance, as it can impair users' ability to accurately interpret real-world information. In this paper we examine two types of task-detrimental virtual content: obstruction attacks, in which virtual content prevents users from seeing real-world objects, and information manipulation attacks, in which virtual content interferes with users' ability to accurately interpret real-world information. We provide a mathematical framework to characterize these attacks and create a custom open-source dataset for attack evaluation. To address these attacks, we introduce ViDDAR (Vision language model-based Task-Detrimental content Detector for Augmented Reality), a comprehensive full-reference system that leverages Vision Language Models (VLMs) and advanced deep learning techniques to monitor and evaluate virtual content in AR environments, employing a user-edge-cloud architecture to balance performance with low latency. To the best of our knowledge, ViDDAR is the first system to employ VLMs for detecting task-detrimental content in AR settings. Our evaluation results demonstrate that ViDDAR effectively understands complex scenes and detects task-detrimental content, achieving up to 92.15% obstruction detection accuracy with a detection latency of 533 ms, and an 82.46% information manipulation content detection accuracy with a latency of 9.62 s.

PMID:40072851 | DOI:10.1109/TVCG.2025.3549147

Jpn J Radiol. 2025 Mar 12. doi: 10.1007/s11604-025-01760-2. Online ahead of print.

ABSTRACT

PURPOSE: Myocardial computed tomography (CT) late enhancement (LE) allows assessment of myocardial scarring. Super-resolution deep learning image reconstruction (SR-DLR) trained on data acquired from ultra-high-resolution CT may improve image quality for CT-LE. Therefore, this study investigated image noise and image quality with SR-DLR compared with conventional DLR (C-DLR) and hybrid iterative reconstruction (hybrid IR).

METHODS AND METHODS: We retrospectively analyzed 30 patients who underwent CT-LE using 320-row CT. The CT protocol comprised stress dynamic CT perfusion, coronary CT angiography, and CT-LE. CT-LE images were reconstructed using three different algorithms: SR-DLR, C-DLR, and hybrid IR. Image noise, signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and qualitative image quality scores are in terms of noise reduction, sharpness, visibility of scar and myocardial boarder, and overall image quality. Inter-observer differences in myocardial scar sizing in CT-LE by the three algorithms were also compared.

RESULTS: SR-DLR significantly decreased image noise by 35% compared to C-DLR (median 6.2 HU, interquartile range [IQR] 5.6-7.2 HU vs 9.6 HU, IQR 8.4-10.7 HU; p < 0.001) and by 37% compared to hybrid IR (9.8 HU, IQR 8.5-12.0 HU; p < 0.001). SNR and CNR of CT-LE reconstructed using SR-DLR were significantly higher than with C-DLR (both p < 0.001) and hybrid IR (both p < 0.05). All qualitative image quality scores were higher with SR-DLR than those with C-DLR and hybrid IR (all p < 0.001). The inter-observer differences in scar sizing were reduced with SR-DLR and C-DLR compared with hybrid IR (both p = 0.02).

CONCLUSION: SR-DLR reduces image noise and improves image quality of myocardial CT-LE compared with C-DLR and hybrid IR techniques and improves inter-observer reproducibility of scar sizing compared to hybrid IR. The SR-DLR approach has the potential to improve the assessment of myocardial scar by CT late enhancement.

PMID:40072715 | DOI:10.1007/s11604-025-01760-2

J Med Internet Res. 2025 Mar 12;27:e68442. doi: 10.2196/68442.

ABSTRACT

BACKGROUND: Hypertension is a major global health issue and a significant modifiable risk factor for cardiovascular diseases, contributing to a substantial socioeconomic burden due to its high prevalence. In China, particularly among populations living near desert regions, hypertension is even more prevalent due to unique environmental and lifestyle conditions, exacerbating the disease burden in these areas, underscoring the urgent need for effective early detection and intervention strategies.

OBJECTIVE: This study aims to develop, calibrate, and prospectively validate a 2-year hypertension risk prediction model by using large-scale health examination data collected from populations residing in 4 regions surrounding the Taklamakan Desert of northwest China.

METHODS: We retrospectively analyzed the health examination data of 1,038,170 adults (2019-2021) and prospectively validated our findings in a separate cohort of 961,519 adults (2021-2023). Data included demographics, lifestyle factors, physical examinations, and laboratory measurements. Feature selection was performed using light gradient-boosting machine-based recursive feature elimination with cross-validation and Least Absolute Shrinkage and Selection Operator, yielding 24 key predictors. Multiple machine learning (logistic regression, random forest, extreme gradient boosting, light gradient-boosting machine) and deep learning (Feature Tokenizer + Transformer, SAINT) models were trained with Bayesian hyperparameter optimization.

RESULTS: Over a 2-year follow-up, 15.20% (157,766/1,038,170) of the participants in the retrospective cohort and 10.50% (101,077/961,519) in the prospective cohort developed hypertension. Among the models developed, the CatBoost model demonstrated the best performance, achieving area under the curve (AUC) values of 0.888 (95% CI 0.886-0.889) in the retrospective cohort and 0.803 (95% CI 0.801-0.804) in the prospective cohort. Calibration via isotonic regression improved the model's probability estimates, with Brier scores of 0.090 (95% CI 0.089-0.091) and 0.102 (95% CI 0.101-0.103) in the internal validation and prospective cohorts, respectively. Participants were ranked by the positive predictive value calculated using the calibrated model and stratified into 4 risk categories (low, medium, high, and very high), with the very high group exhibiting a 41.08% (5741/13,975) hypertension incidence over 2 years. Age, BMI, and socioeconomic factors were identified as significant predictors of hypertension.

CONCLUSIONS: Our machine learning model effectively predicted the 2-year risk of hypertension, making it particularly suitable for preventive health care management in high-risk populations residing in the desert regions of China. Our model exhibited excellent predictive performance and has potential for clinical application. A web-based application was developed based on our predictive model, which further enhanced the accessibility for clinical and public health use, aiding in reducing the burden of hypertension through timely prevention strategies.

PMID:40072485 | DOI:10.2196/68442

J Biomol Struct Dyn. 2025 Mar 12:1-13. doi: 10.1080/07391102.2025.2475229. Online ahead of print.

ABSTRACT

Prediction of protein-ligand interactions is critical for drug discovery and repositioning. Traditional prediction methods are computationally intensive and limited in modeling structural changes. In contrast, data-driven deep learning methods significantly reduce computational costs and offer a more efficient approach for drug discovery. However, existing models often fail to fully exploit metadata and low-frequency features, leading to suboptimal performance on sparse, imbalanced datasets. To address these challenges, this paper proposes a novel interaction prediction model based on heterogeneous graphs and data enhancement, named Heterogeneous Graph Enhanced Fusion Network (HGEF-Net). The model utilizes a heterogeneous information learning module, which deeply analyzes molecular subgraphs and substructures, fully leveraging metadata features to better capture the biological interactions between ligands and proteins. Additionally, to address the issue of low-frequency category features, a data enhancement strategy based on multi-level contrastive learning is proposed. Furthermore, a heterogeneous attention integration framework is presented, which uses multi-level attention to assign different weights to various features. This approach efficiently fuses both intramolecular and intermolecular features, enhancing the model's ability to capture key information and improving its performance on sparse, imbalanced datasets. Experimental results show that HGEF-Net outperforms other state-of-the-art models. On the BindingDB dataset (1:100 positive-to-negative ratio), HGEF-Net achieves an AUC of 0.826, AUPRC of 0.811, Precision of 0.715, and Recall of 0.709. On the Davis dataset (1:10 ratio), the data enhancement module improves AUC, AUPRC, Precision, and Recall by 11.7%, 9.7%, 10.5%, and 16.3%, respectively, validating the model's effectiveness.

PMID:40072484 | DOI:10.1080/07391102.2025.2475229

Am J Respir Crit Care Med. 2025 Mar 12. doi: 10.1164/rccm.202409-1847OC. Online ahead of print.

ABSTRACT

RATIONALE: Quantifying functional small airways disease (fSAD) requires additional expiratory computed tomography (CT) scan, limiting clinical applicability. Artificial intelligence (AI) could enable fSAD quantification from chest CT scan at total lung capacity (TLC) alone (fSADTLC).

OBJECTIVES: To evaluate an AI model for estimating fSADTLC, compare it with dual-volume parametric response mapping fSAD (fSADPRM), and assess its clinical associations and repeatability in chronic obstructive pulmonary disease (COPD).

METHODS: We analyzed 2513 participants from the SubPopulations and InteRmediate Outcome Measures in COPD Study (SPIROMICS). Using a randomly sampled subset (n = 1055), we developed a generative model to produce virtual expiratory CTs for estimating fSADTLC in the remaining 1458 SPIROMICS participants. We compared fSADTLC with dual volume, parametric response mapping fSADPRM. We investigated univariate and multivariable associations of fSADTLC with FEV1, FEV1/FVC, six-minute walk distance (6MWD), St. George's Respiratory Questionnaire (SGRQ), and FEV1 decline. The results were validated in a subset (n = 458) from COPDGene study. Multivariable models were adjusted for age, race, sex, BMI, baseline FEV1, smoking pack years, smoking status, and percent emphysema.

MEASUREMENTS AND MAIN RESULTS: Inspiratory fSADTLC showed a strong correlation with fSADPRM in both SPIROMICS (Pearson's R = 0.895) and COPDGene (R = 0.897) cohorts. Higher fSADTLC levels were significantly associated with lower lung function, including lower postbronchodilator FEV1 (L) and FEV1/FVC ratio, and poorer quality of life reflected by higher total SGRQ scores, independent of percent CT emphysema. In SPIROMICS, individuals with higher fSADTLC experienced an annual decline in FEV1 of 1.156 mL (relative decrease; 95% CI: 0.613, 1.699; P < 0.001) per year for every 1% increase in fSADTLC. The rate of decline in COPDGene was slightly lower at 0.866 mL / year (relative decrease; 95% CI: 0.345, 1.386; P < 0.001) for percent increase in fSADTLC. Inspiratory fSADTLC demonstrated greater consistency between repeated measurements with a higher intraclass correlation coefficient (ICC) of 0.99 (95% CI: 0.98, 0.99) compared to fSADPRM [ICC: 0.83 (95% CI: 0.76, 0.88)].

CONCLUSIONS: Small airways disease can be reliably assessed from a single inspiratory CT scan using generative AI, eliminating the need for an additional expiratory CT scan. fSAD estimation from inspiratory CT correlates strongly with fSADPRM, demonstrates a significant association with FEV1 decline, and offers greater repeatability.

PMID:40072247 | DOI:10.1164/rccm.202409-1847OC