J Pharm Anal. 2025 Jun;15(6):101194. doi: 10.1016/j.jpha.2025.101194. Epub 2025 Jan 8.

ABSTRACT

Identifying drug-drug interactions (DDIs) is essential to prevent adverse effects from polypharmacy. Although deep learning has advanced DDI identification, the gap between powerful models and their lack of clinical application and evaluation has hindered clinical benefits. Here, we developed a Multi-Dimensional Feature Fusion model named MDFF, which integrates one-dimensional simplified molecular input line entry system sequence features, two-dimensional molecular graph features, and three-dimensional geometric features to enhance drug representations for predicting DDIs. MDFF was trained and validated on two DDI datasets, evaluated across three distinct scenarios, and compared with advanced DDI prediction models using accuracy, precision, recall, area under the curve, and F1 score metrics. MDFF achieved state-of-the-art performance across all metrics. Ablation experiments showed that integrating multi-dimensional drug features yielded the best results. More importantly, we obtained adverse drug reaction reports uploaded by Xiangya Hospital of Central South University from 2021 to 2023 and used MDFF to identify potential adverse DDIs. Among 12 real-world adverse drug reaction reports, the predictions of 9 reports were supported by relevant evidence. Additionally, MDFF demonstrated the ability to explain adverse DDI mechanisms, providing insights into the mechanisms behind one specific report and highlighting its potential to assist practitioners in improving medical practice.

PMID:40678481 | PMC:PMC12268060 | DOI:10.1016/j.jpha.2025.101194

J Pharm Anal. 2025 Jun;15(6):101295. doi: 10.1016/j.jpha.2025.101295. Epub 2025 Apr 9.

ABSTRACT

Combined with elastic network model (ENM), the perturbation response scanning (PRS) has emerged as a robust technique for pinpointing allosteric interactions within proteins. Here, we proposed the PRS analysis of drug-target networks (DTNs), which could provide a promising avenue in network medicine. We demonstrated the utility of the method by introducing a deep learning and network perturbation-based framework, for drug repurposing of multiple sclerosis (MS). First, the MS comorbidity network was constructed by performing a random walk with restart algorithm based on shared genes between MS and other diseases as seed nodes. Then, based on topological analysis and functional annotation, the neurotransmission module was identified as the "therapeutic module" of MS. Further, perturbation scores of drugs on the module were calculated by constructing the DTN and introducing the PRS analysis, giving a list of repurposable drugs for MS. Mechanism of action analysis both at pathway and structural levels screened dihydroergocristine as a candidate drug of MS by targeting a serotonin receptor of serotonin 2B receptor (HTR2B). Finally, we established a cuprizone-induced chronic mouse model to evaluate the alteration of HTR2B in mouse brain regions and observed that HTR2B was significantly reduced in the cuprizone-induced mouse cortex. These findings proved that the network perturbation modeling is a promising avenue for drug repurposing of MS. As a useful systematic method, our approach can also be used to discover the new molecular mechanism and provide effective candidate drugs for other complex diseases.

PMID:40678478 | PMC:PMC12268079 | DOI:10.1016/j.jpha.2025.101295

J Pharm Anal. 2025 Jun;15(6):101174. doi: 10.1016/j.jpha.2024.101174. Epub 2024 Dec 28.

ABSTRACT

Negative logarithm of the acid dissociation constant (pK a) significantly influences the absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties of molecules and is a crucial indicator in drug research. Given the rapid and accurate characteristics of computational methods, their role in predicting drug properties is increasingly important. Although many pK a prediction models currently exist, they often focus on enhancing model precision while neglecting interpretability. In this study, we present GraFpK a, a pK a prediction model using graph neural networks (GNNs) and molecular fingerprints. The results show that our acidic and basic models achieved mean absolute errors (MAEs) of 0.621 and 0.402, respectively, on the test set, demonstrating good predictive performance. Notably, to improve interpretability, GraFpK a also incorporates Integrated Gradients (IGs), providing a clearer visual description of the atoms significantly affecting the pK a values. The high reliability and interpretability of GraFpK a ensure accurate pK a predictions while also facilitating a deeper understanding of the relationship between molecular structure and pK a values, making it a valuable tool in the field of pK a prediction.

PMID:40678476 | PMC:PMC12268062 | DOI:10.1016/j.jpha.2024.101174

MethodsX. 2025 Jun 20;15:103456. doi: 10.1016/j.mex.2025.103456. eCollection 2025 Dec.

ABSTRACT

Prediction of droughts has recently become imperative as frequency and intensity are increasing mostly due to climatic variation. Indeed, drought is a highly significant disaster that results in widespread damage to all kinds of ecosystems, agricultural production systems, and water resources systems. Accurate techniques of forecasting are necessary for the purpose. Conventional methods lack the intricate time-space correlation in meteorological data. The research proposes an ensemble of Extreme Gradient Boosting (XGBoost), Long Short Term Memory (LSTM), and Tabular Network (TabNet) for a higher accuracy in drought forecasting. With the large meteorological dataset that involves temperature, precipitation, humidity, and wind speed as features, the model integrates:•The tree capabilities of XGBoost perform feature selection very effectively.•Temporal Pattern Analysis using LSTM.•Insight obtained from the attention mechanism-based TabNet.Empirical results demonstrate that the proposed ensemble outperforms individual models, achieving the lowest Root Mean Square Error (RMSE: 0.6582) and Mean Absolute Error (MAE: 0.5377), and the highest Coefficient of Determination (R²: 0.5069). Furthermore, it yields the best Nash-Sutcliffe Efficiency (NSE: 0.5107) and Kling-Gupta Efficiency (KGE: 0.6039), confirming its superiority in drought severity forecasting. The ensemble outperforms traditional models, aiding early drought warnings and water conservation planning.

PMID:40678463 | PMC:PMC12268929 | DOI:10.1016/j.mex.2025.103456

MethodsX. 2025 Jun 27;15:103466. doi: 10.1016/j.mex.2025.103466. eCollection 2025 Dec.

ABSTRACT

Cardiovascular diseases (CVD) remain the leading cause of mortality worldwide, emphasizing the urgent need for accurate and efficient predictive models. This study proposes a dual-output deep learning model based on a hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model, optimized using the Evolutionary Mating Algorithm (EMA). The model predicts both a continuous risk score and a binary diagnostic outcome, supporting both quantitative assessment and early clinical decision-making. EMA was applied for hyperparameter optimization, demonstrating improved convergence and generalization over conventional methods. Performance was benchmarked against CNN-LSTM models optimized using Particle Swarm Optimization (PSO) and Barnacle Mating Optimization (BMO). The EMA-based model achieved superior results, with a Mean Absolute Error (MAE) of 0.018, Mean Squared Error (MSE) of 0.0006, Root Mean Squared Error (RMSE) of 0.024, and a coefficient of determination (R²) of 0.98 for risk prediction. For the diagnostic task, the model attained 70 % accuracy and 80 % precision. These findings validate EMA's effectiveness in tuning dual-output deep learning models and highlight its potential in enhancing cardiovascular risk stratification and early diagnosis in clinical settings.•Dual-output CNN-LSTM model optimized using EMA.•Continuous risk scores and binary diagnostic classification predictions.•EMA outperformed PSO and BMO in predictive accuracy and model robustness.

PMID:40678461 | PMC:PMC12268569 | DOI:10.1016/j.mex.2025.103466

BMC Med Res Methodol. 2025 Jul 17;25(1):175. doi: 10.1186/s12874-025-02618-x.

ABSTRACT

BACKGROUND: Individualized prediction of health outcomes supports clinical medicine and decision making. Our primary objective was to offer a comprehensive survey of methods for the dynamic prediction of Alzheimer's disease (AD), encompassing both conventional statistical methods and deep learning techniques.

METHODS: Articles were sourced from PubMed, Embase and Web of Science databases using keywords related to dynamic prediction of AD. A set of criteria was developed to identify included studies. The correlation information for the construction of models was extracted.

RESULTS: We identified four methodological frameworks for dynamic prediction from 18 studies with two-stage model (n = 3), joint model (n = 11), landmark model (n = 2) and deep learning (n = 2). We reported and summarized the specific construction of models and their applications.

CONCLUSIONS: Each framework possesses distinctive principles and attendant benefits. The dynamic prediction models excel in predicting the prognosis of individual patients in a real-time manner, surpassing the limitations of traditional baseline-only prediction models. Future work should consider various data types, complex longitudinal data, missing data, assumption violations, survival outcomes, and interpretability of models.

PMID:40676602 | DOI:10.1186/s12874-025-02618-x

Reprod Biol Endocrinol. 2025 Jul 17;23(1):102. doi: 10.1186/s12958-025-01437-5.

ABSTRACT

BACKGROUND: Conservative treatment remains a viable option for selected patients with ectopic pregnancy (EP), but failure may lead to rupture and serious complications. Currently, serum β-hCG is the main predictor for treatment outcomes, yet its accuracy is limited. This study aimed to develop and validate a predictive model that integrates radiomic features derived from super-resolution (SR) ultrasound images with clinical biomarkers to improve risk stratification.

METHODS: A total of 228 patients with EP receiving conservative treatment were retrospectively included, with 169 classified as treatment success and 59 as failure. SR images were generated using a deep learning-based generative adversarial network (GAN). Radiomic features were extracted from both normal-resolution (NR) and SR ultrasound images. Features with intraclass correlation coefficient (ICC) ≥ 0.75 were retained after intra- and inter-observer evaluation. Feature selection involved statistical testing and Least Absolute Shrinkage and Selection Operator (LASSO) regression. Random forest algorithms were used to construct NR and SR models. A clinical model based on serum β-hCG was also developed. The Clin-SR model was constructed by fusing SR radiomics with β-hCG values. Model performance was evaluated using area under the curve (AUC), calibration, and decision curve analysis (DCA). An independent temporal validation cohort (n = 40; 20 failures, 20 successes) was used to validation of the nomogram derived from the Clin-SR model.

RESULTS: The SR model significantly outperformed the NR model in the test cohort (AUC: 0.791 ± 0.015 vs. 0.629 ± 0.083). In a representative iteration, the Clin-SR fusion model achieved an AUC of 0.870 ± 0.015, with good calibration and net clinical benefit, suggesting reliable performance in predicting conservative treatment failure. In the independent validation cohort, the nomogram demonstrated good generalizability with an AUC of 0.808 and consistent calibration across risk thresholds. Key contributing radiomic features included Gray Level Variance and Voxel Volume, reflecting lesion heterogeneity and size.

CONCLUSIONS: The Clin-SR model, which integrates deep learning-enhanced SR ultrasound radiomics with serum β-hCG, offers a robust and non-invasive tool for predicting conservative treatment failure in ectopic pregnancy. This multimodal approach enhances early risk stratification and supports personalized clinical decision-making, potentially reducing overtreatment and emergency interventions.

PMID:40676578 | DOI:10.1186/s12958-025-01437-5

BMC Med Imaging. 2025 Jul 17;25(1):288. doi: 10.1186/s12880-025-01828-5.

ABSTRACT

BACKGROUND: PD-1/PD-L1 immunotherapy represents the primary treatment for advanced NSCLC patients; however, response rates to this therapy vary among individuals. This dual-center study aimed to integrate habitat radiomics and multi-instance deep learning to predict durable clinical benefits from immunotherapy.

METHODS: We retrospectively collected 590 NSCLC patients from two medical centers who received PD-1/PD-L1 inhibitor immunotherapy. Patients from the GMU center were divided into a training cohort (n = 375) and an internal validation cohort (n = 161) for habitat analysis and multi-instance deep learning model development. Patients from the YJ center formed an external testing cohort (n = 54) for model validation. We implemented a DenseNet121-based architecture extracting radiomics features from triplanar (axial/coronal/sagittal) tumor sequences to construct a 2.5D deep-learning dataset. Then, we fuse 2.5D features through multi-instance learning. Additionally, we use K-means clustering to divide the tumor VOI into three subregions to extract radiological features for building a Habitat model. Finally, we use the Extra-Trees classifier to construct MIL, Habitat, and Combined models, the Combined model integrating age factors into the analysis. The primary endpoint was durable clinical benefit. Finally, a separate PD-L1 expression dataset was used to compare the predictive performance of imaging models against PD-L1 status (positive/negative) and expression levels (high/low) to identify the optimal model for predicting immunotherapy clinical benefit.

RESULTS: The Combined model combining Habitat, MIL, and patient age demonstrated robust DCB prediction with AUCs of 0.906(95% CI: 0.874-0.936), 0.889(95% CI: 0.826-0.948), and 0.831 (95% CI: 0.710-0.927)in training, validation, and testing cohorts respectively. Comparative analysis revealed all imaging models outperformed PD-L1 expression status (positive/negative) and levels (high/low) in predicting therapeutic response, with Habitat analysis showing superior performance to MIL alone. Notably, peritumoral structural features emerged as significant predictors of treatment efficacy.

CONCLUSION: This non-invasive predictive framework provides clinically actionable insights for immunotherapy stratification, potentially overcoming limitations of current biomarker testing while highlighting the prognostic value of spatial tumor heterogeneity analysis.

PMID:40676504 | DOI:10.1186/s12880-025-01828-5

J Clin Monit Comput. 2025 Jul 17. doi: 10.1007/s10877-025-01328-5. Online ahead of print.

ABSTRACT

Deterioration of ventriculoarterial coupling is detrimental to cardiovascular and left ventricular function. To enable continuous monitoring of left ventricular function, we have developed autoMAPSE, a new tool that combines transoesophageal echocardiography with deep learning for automatic measurement of mitral annular plane systolic excursion. We hypothesised that autoMAPSE could be used to monitor systemic ventriculoarterial coupling and detect alterations in postoperative cardiac biomarkers. To test this hypothesis, we monitored 50 patients for 120 min immediately after cardiac surgery by measuring autoMAPSE and mean arterial pressure (MAP) every 5 min. Postoperative N-terminal pro B-type natriuretic peptide (ProBNP) and high-sensitivity troponin-T (TnT) were measured twice daily until the evening of postoperative day 1. Ventriculoarterial coupling was assessed non-invasively by calculating arterial elastance and end-systolic elastance (Ea/Ees-ratio). The relationship between autoMAPSE and ventriculoarterial coupling was assessed by 1) correlating Ea/Ees-ratio with one simultaneous autoMAPSE measurement, and 2) relating the measurements of autoMAPSE with corresponding MAP within each patient using a linear mixed model with random slopes. We found that autoMAPSE correlated negatively with Ea/Ees-ratio (rho = - 0.61, P < 0.05). Furthermore, the individual slopes relating autoMAPSE to MAP were highly significant (P < 0.001) and markedly heterogeneous (both positive and negative), suggesting that ventriculoarterial coupling differs substantially in different individual patients. Finally, continuous autoMAPSE measurements were negatively correlated with both peak postoperative ProBNP (rho = - 0.46, P < 0.001) and TnT (rho = - 0.29, P < 0.05). In conclusion, continuous monitoring using autoMAPSE in the first two postoperative hours reflected ventriculoarterial coupling as well as peak ProBNP and TnT during the subsequent 24 h.

PMID:40676456 | DOI:10.1007/s10877-025-01328-5

J Sleep Res. 2025 Jul 17:e70149. doi: 10.1111/jsr.70149. Online ahead of print.

ABSTRACT

Children's ambulatory sleep is commonly measured via actigraphy. However, traditional actigraphy measured sleep (e.g., Sadeh algorithm) struggles to predict wake (i.e., specificity, values typically < 70) and cannot predict sleep stages. Long short-term memory (LSTM) is a machine learning algorithm that may address these deficiencies. This study evaluated the agreement of LSTM sleep estimates from actigraphy and heartrate (HR) data with polysomnography (PSG). Children (N = 238, 5-12 years, 52.8% male, 50% Black 31.9% White) participated in an overnight laboratory polysomnography. Participants were referred because of suspected sleep disruptions. Children wore an ActiGraph GT9X accelerometer and two of three consumer wearables (i.e., Apple Watch Series 7, Fitbit Sense, Garmin Vivoactive 4) on their non-dominant wrist during the polysomnogram. LSTM estimated sleep versus wake and sleep stage (wake, not-REM, REM) using raw actigraphy and HR data for each 30-s epoch. Logistic regression and random forest were also estimated as a benchmark for performance with which to compare the LSTM results. A 10-fold cross-validation technique was employed, and confusion matrices were constructed. Sensitivity and specificity were calculated to assess the agreement between research-grade and consumer wearables with the criterion polysomnography. For sleep versus wake classification, LSTM outperformed logistic regression and random forest with accuracy ranging from 94.1 to 95.1, sensitivity ranging from 94.9 to 95.9 across different devices, and specificity ranging from 84.5 to 89.6. The addition of HR improved the prediction of sleep stages but not binary sleep versus wake. LSTM is promising for predicting sleep and sleep staging from actigraphy data, and HR may improve sleep stage prediction.

PMID:40676371 | DOI:10.1111/jsr.70149

Int J Comput Assist Radiol Surg. 2025 Jul 17. doi: 10.1007/s11548-025-03473-3. Online ahead of print.

ABSTRACT

PURPOSE: Augmented Reality in Minimally Invasive Surgery has made tremendous progress in organs including the liver and the uterus. The core problem of Augmented Reality is registration, where a preoperative patient's geometric digital twin must be aligned with the image of the surgical camera. The case of the kidney is yet unresolved, owing to the absence of anatomical landmarks visible in both the patient's digital twin and the surgical images.

METHODS: We propose a landmark-free approach to registration, which is particularly well-adapted to the kidney. The approach involves a generic kidney model and an end-to-end neural network, which we train with a proposed dataset to regress the registration directly from a surgical RGB image.

RESULTS: Experimental evaluation across four clinical cases demonstrates strong concordance with expert-labelled registration, despite anatomical and motion variability. The proposed method achieved an average tumour contour alignment error of 7.3 ± 4.1 mm in 9.4 ± 0.2 ms.

CONCLUSION: This landmark-free registration approach meets the accuracy, speed and resource constraints required in clinical practice, making it a promising tool for Augmented Reality-Assisted Partial Nephrectomy.

PMID:40676342 | DOI:10.1007/s11548-025-03473-3

Commun Med (Lond). 2025 Jul 17;5(1):296. doi: 10.1038/s43856-025-01015-1.

ABSTRACT

BACKGROUND: Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder that affects behavior, attention, and learning. Current diagnoses rely heavily on subjective assessments, underscoring the need for objective imaging-based methods. This study aims to explore whether structural connectivity networks derived from MRI can reveal alterations associated with ADHD and support data-driven understanding.

METHODS: We collected brain MRI data from 947 individuals (aged 7-26 years; 590 males, 356 females, 1 unspecified) across eight centers, sourced from the Neuro Bureau ADHD-200 preprocessed dataset. Transformer-based deep learning models were used to learn relationships between different brain regions and construct structural connectivity networks. To prepare input for the model, each region was transformed into a standardized data sequence using four different strategies. The strength of connectivity between brain regions was then measured to identify structural differences related to ADHD. Five-fold cross-validation and statistical analyses were used to evaluate model robustness and group differences, respectively.

RESULTS: Here we show that the proposed method performs well in distinguishing ADHD individuals from healthy controls, with accuracy reaching 71.9 percent and an area under curve of 0.74. The structural networks also reveal significant differences in connectivity patterns (paired t-test: P = 0.81 × 10-6), particularly involving regions responsible for motor and executive function. Notably, the importance rankings of several brain regions, including the thalamus and caudate, differ markedly between groups.

CONCLUSIONS: This study shows that ADHD may be associated with connectivity alterations in multiple brain regions. Our findings suggest that brain structural connectivity networks built using Transformer-based methods offer a promising tool for both diagnosis and further research into brain structure.

PMID:40676171 | DOI:10.1038/s43856-025-01015-1

Sci Rep. 2025 Jul 17;15(1):26002. doi: 10.1038/s41598-025-11089-5.

ABSTRACT

Sparse-view computed tomography (CT) holds promise for reducing radiation exposure and enabling novel system designs. Traditional reconstruction algorithms, including Filtered Backprojection (FBP) and Model-Based Iterative Reconstruction (MBIR), often produce artifacts in sparse-view data. Deep Learning Reconstruction (DLR) offers potential improvements, but task-based evaluations of DLR in sparse-view CT remain limited. This study employs an Artificial Intelligence (AI) observer to evaluate the diagnostic accuracy of FBP, MBIR, and DLR for intracranial hemorrhage detection and classification, offering a cost-effective alternative to human radiologist studies. A public brain CT dataset with labeled intracranial hemorrhages was used to train an AI observer model. Sparse-view CT data were simulated, with reconstructions performed using FBP, MBIR, and DLR. Reconstruction quality was assessed using metrics such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), and Learned Perceptual Image Patch Similarity (LPIPS). Diagnostic utility was evaluated using Receiver Operating Characteristic (ROC) analysis and Area Under the Curve (AUC) values for One-vs-Rest and One-vs-One classification tasks. DLR outperformed FBP and MBIR in all quality metrics, demonstrating reduced noise, improved structural similarity, and fewer artifacts. The AI observer achieved the highest classification accuracy with DLR, while FBP surpassed MBIR in task-based accuracy despite inferior image quality metrics, emphasizing the value of task-based evaluations. DLR provides an effective balance of artifact reduction and anatomical detail in sparse-view CT brain imaging. This proof-of-concept study highlights AI observer models as a viable, cost-effective alternative for evaluating CT reconstruction techniques.

PMID:40676122 | DOI:10.1038/s41598-025-11089-5

Sci Rep. 2025 Jul 17;15(1):25946. doi: 10.1038/s41598-025-11518-5.

ABSTRACT

B-ultrasound results are widely used in early pregnancy loss (EPL) prediction, but there are inevitable intra-observer and inter-observer errors in B-ultrasound results especially in early pregnancy, which lead to inconsistent assessment of embryonic status, and thus affect the judgment of EPL. To address this, we need a rapid and accurate model to predict pregnancy loss in the first trimester. This study aimed to construct an artificial intelligence model to automatically extract biometric parameters from ultrasound videos of early embryos and predict pregnancy loss. This can effectively eliminate the measurement error of B-ultrasound results, accurately predict EPL, and provide decision support for doctors with relatively little clinical experience. A total of 630 ultrasound videos from women with early singleton pregnancies of gestational age between 6 and 10 weeks were used for training. A two-stage artificial intelligence model was established. First, some biometric parameters such as gestational sac areas (GSA), yolk sac diameter (YSD), crown rump length (CRL) and fetal heart rate (FHR), were extract from ultrasound videos by a deep neural network named A3F-net, which is a modified neural network based on U-Net designed by ourselves. Then an ensemble learning model predicted pregnancy loss risk based on these features. Dice, IOU and Precision were used to evaluate the measurement results, and sensitivity, AUC etc. were used to evaluate the predict results. The fetal heart rate was compared with those measured by doctors, and the accuracy of results was compared with other AI models. In the biometric features measurement stage, the precision of GSA, YSD and CRL of A3F-net were 98.64%, 96.94% and 92.83%, it was the highest compared to other 2 models. Bland-Altman analysis did not show systematic deviations between doctors and AI. The mean and standard deviation of the mean relative error between doctors and the AI model was 0.060 ± 0.057. In the EPL prediction stage, the ensemble learning models demonstrated excellent performance, with CatBoost being the best-performing model, achieving a precision of 98.0% and an AUC of 0.969 (95% CI: 0.962-0.975). In this study, a hybrid AI model to predict EPL was established. First, a deep neural network automatically measured the biometric parameters from ultrasound video to ensure the consistency and accuracy of the measurements, then a machine learning model predicted EPL risk to support doctors making decisions. The use of our established AI model in EPL prediction has the potential to assist physicians in making more accurate and timely clinical decision in clinical application.

PMID:40676105 | DOI:10.1038/s41598-025-11518-5

Sci Rep. 2025 Jul 18;15(1):26018. doi: 10.1038/s41598-025-10738-z.

ABSTRACT

Automatic modulation classification (AMC) is a critical component in modern communication systems, particularly within software-defined radios, cognitive radio networks, smart grid and and distributed renewable energy systems (RESs) where adaptive and efficient signal processing is essential. This paper proposes a novel deep learning-based AMC method for identifying M-PSK and M-QAM waveform signals in single-relay cooperative MIMO 5G systems operating under partial channel state information (CSI) and spatially correlated channels. The proposed method leverages a convolutional neural network (CNN) classifier trained on a reduced set of discriminative features, including higher-order statistics and the differential nonlinear phase peak factor, which are extracted from the received signal. Feature dimensionality is reduced using the Gram-Schmidt orthogonalization procedure to enhance training efficiency. A centralized decision-making strategy aggregates predictions from multiple antennas. The method is evaluated through simulations using various modulation orders and under challenging conditions such as low signal-to-noise ratios (SNR). Results demonstrate that the proposed CNN-based approach significantly outperforms benchmark machine learning classifiers in terms of classification accuracy, precision, recall, and F-measure. These findings underscore the practical potential of the method for enhancing AMC performance in realistic 5G cooperative scenarios.

PMID:40676088 | DOI:10.1038/s41598-025-10738-z

J Med Internet Res. 2025 Jul 17;27:e74402. doi: 10.2196/74402.

ABSTRACT

BACKGROUND: Traditional ocular gaze photograph preprocessing, relying on manual cropping and head tilt correction, is time-consuming and inconsistent, limiting artificial intelligence (AI) model development and clinical application.

OBJECTIVE: This study aimed to address these challenges using an advanced preprocessing algorithm to enhance the accuracy, efficiency, and standardization of eye region cropping for clinical workflows and AI data preprocessing.

METHODS: This retrospective and prospective cross-sectional study utilized 5832 images from 648 inpatients and outpatients, capturing 3 gaze positions under diverse conditions, including obstructions and varying distances. The preprocessing algorithm, based on a rotating bounding box detection framework, was trained and evaluated using precision, recall, and mean average precision (mAP) at various intersections over union thresholds. A 5-fold cross-validation was performed on an inpatient dataset, with additional testing on an independent outpatient dataset and an external cross-population dataset of 500 images from the IMDB-WIKI collection, representing diverse ethnicities and ages. Expert validation confirmed alignment with clinical standards across 96 images (48 images from a Chinese dataset of patients with strabismus and 48 images from IMDB-WIKI). Gradient-weighted class activation mapping heatmaps were used to assess model interpretability. A control experiment with 5 optometry specialists compared manual and automated cropping efficiency. Downstream task validation involved preprocessing 1000 primary gaze photographs using the Dlib toolkit, faster region-based convolutional neural network (R-CNN; both without head tilt correction), and our model (with correction), evaluating the impact of head tilt correction via the vision transformer strabismus screening network through 5-fold cross-validation.

RESULTS: The model achieved exceptional performance across datasets: on the 5-fold cross-validation set, it recorded a mean precision of 1.000 (95% CI 1.000-1.000), recall of 1.000 (95% CI 1.000-1.000), mAP50 of 0.995 (95% CI 0.995-0.995), and mAP95 of 0.893 (95% CI 0.870-0.918); on the internal independent test set, precision and recall were 1.000, with mAP50 of 0.995 and mAP95 of 0.801; and on the external cross-population test set, precision and recall were 1.000, with mAP50 of 0.937 and mAP95 of 0.792. The control experiment reduced image preparation time from 10 hours for manual cropping of 900 photos to 30 seconds with the automated model. Downstream strabismus screening task validation showed our model (with head tilt correction) achieving an area under the curve of 0.917 (95% CI 0.901-0.933), surpassing Dlib-toolkit and faster R-CNN (both without head tilt correction) with an area under the curve of 0.856 (P=.02) and 0.884 (P=.05), respectively. Heatmaps highlighted core ocular focus, aligning with head tilt directions.

CONCLUSIONS: This study delivers an AI-driven platform featuring a preprocessing algorithm that automates eye region cropping, correcting head tilt variations to improve image quality for AI development and clinical use. Integrated with electronic archives and patient-physician interaction, it enhances workflow efficiency, ensures telemedicine privacy, and supports ophthalmological research and strabismus care.

PMID:40674714 | DOI:10.2196/74402

Magn Reson Med. 2025 Jul 17. doi: 10.1002/mrm.30632. Online ahead of print.

ABSTRACT

PURPOSE: Propose and evaluate an automatic approach for predicting the optimal inversion time (TI) for dark and gray blood images for flow-independent dark-blood delayed-enhancement (FIDDLE) acquisition based on free-breathing FIDDLE TI-scout images.

METHODS: In 267 patients, the TI-scout sequence acquired single-shot magnetization-prepared and associated reference images (without preparation) on a 3 T Magnetom Vida and a 1.5 T Magnetom Sola scanner. Data were reconstructed into phase-corrected TI-scout images typically covering TIs from 140 to 440 ms (20 ms increment). A deep learning network was trained to segment the myocardium and blood pool in reference images. These segmentation masks were transferred to the TI-scout images to derive intensity features of myocardium and blood, with which T1-recovery curves were determined by logarithmic fitting. The optimal TI for dark and gray blood images were derived as linear functions of the TI in which both T1-curves cross. This TI-prediction pipeline was evaluated in 64 clinical subjects.

RESULTS: The pipeline predicted optimal TIs with an average error less than 10 ms compared to manually annotated optimal TIs.

CONCLUSION: The presented approach reliably and automatically predicted optimal TI for dark and gray blood FIDDLE acquisition, with an average error less than the TI increment of the FIDDLE TI-scout sequence.

PMID:40674608 | DOI:10.1002/mrm.30632

PLoS One. 2025 Jul 17;20(7):e0327586. doi: 10.1371/journal.pone.0327586. eCollection 2025.

ABSTRACT

Copy move forgery is a type of image forgery in which a portion of the original image is copied and pasted in a new location on the same image. The consistent illumination and noise pattern make this kind of forgery more difficult to detect. In copy-move forgery detection, conventional approaches are generally effective at identifying simple multiple copy-move forgeries. However, the conventional approaches and deep learning approaches often fall short in detecting multiple forgeries when transformations are applied to the copied regions. Motivated from these findings, a transform domain method for generating and analyzing multiple copy-move forgeries is proposed in this paper. This method utilizes the discrete wavelet transform (DWT) to decompose the original and patch image into approximate (low frequency) and detail coefficients (high frequency). The patch image approximate and details coefficients are inserted into the corresponding positions of the original image wavelet coefficients. The inverse DWT (IDWT) reconstructs the processed image planes after modification which simulates the multiple copy move forgery. In addition, this approach is tested by resizing the region of interest with varying patch sizes resulting in an interesting set of outcomes when evaluated against existing state-of-the-art techniques. This evaluation allows us to identify gaps in existing approaches and suggest improvements for creating more robust detection techniques for multiple copy-move forgeries.

PMID:40674455 | DOI:10.1371/journal.pone.0327586

PLoS One. 2025 Jul 17;20(7):e0324957. doi: 10.1371/journal.pone.0324957. eCollection 2025.

ABSTRACT

Pneumonia, a severe lung infection caused by various viruses, presents significant challenges in diagnosis and treatment due to its similarities with other respiratory conditions. Additionally, the need to protect patient privacy complicates the sharing of sensitive clinical data. This study introduces FLPneXAINet, an effective framework that combines federated learning (FL) with deep learning (DL) and explainable AI (XAI) to securely and accurately predict pneumonia using chest X-ray (CXR) images. We utilized a benchmark dataset from Kaggle, comprising 8,402 CXR images (3,904 normal and 4,498 pneumonia). The dataset was preprocessed and augmented using a cycle-consistent generative adversarial (CycleGAN) network to increase the volume of training data. Three pre-trained DL models named VGG16, NASNetMobile, and MobileNet were employed to extract features from the augmented dataset. Further, four ensemble DL (EDL) models were used to enhance feature extraction. Feature optimization was performed using recursive feature elimination (RFE), analysis of variance (ANOVA), and random forest (RF) to select the most relevant features. These optimized features were then inputted into machine learning (ML) models, including K-nearest neighbor (KNN), naive bayes (NB), support vector machine (SVM), and RF, for pneumonia prediction. The performance of the models was evaluated in a FL environment, with the EDL network achieving the best results: accuracy 97.61%, F1 score 98.36%, recall 98.13%, and precision 98.59%. The framework's predictions were further validated using two XAI techniques-Local Interpretable Model-Agnostic Explanations (LIME) and Grad-CAM. FLPneXAINet offers a robust solution for healthcare professionals to accurately diagnose pneumonia, ensuring timely treatment while safeguarding patient privacy.

PMID:40674439 | DOI:10.1371/journal.pone.0324957

Cancer Rep (Hoboken). 2025 Jul;8(7):e70240. doi: 10.1002/cnr2.70240.

ABSTRACT

BACKGROUND AND AIMS: Lung cancer remains a leading cause of cancer-related deaths worldwide, with early detection critical for improving prognosis. Traditional machine learning (ML) models have shown limited generalizability in clinical settings. This study proposes a deep learning-based approach using transfer learning to accurately segment lung tumor regions from CT scans and classify images as cancerous or noncancerous, aiming to overcome the limitations of conventional ML models.

METHODS: We developed a two-stage model utilizing a ResNet50 backbone within a U-Net architecture for lesion segmentation, followed by a multi-layer perceptron (MLP) for binary classification. The model was trained on publicly available CT scan datasets and evaluated on an independent clinical dataset from Hazrat Rasool Hospital, Iran. Training employed binary cross-entropy and Dice loss functions. Data augmentation, dropout, and regularization were used to enhance model generalizability and prevent overfitting.

RESULTS: The model achieved 94% accuracy on the real-world clinical test set. Evaluation metrics, including F1 score, Matthews correlation coefficient (MCC), Cohen's kappa, and Dice index, confirmed the model's robustness and diagnostic reliability. In comparison, traditional ML models performed poorly on external test data despite high training accuracy, highlighting a significant generalization gap.

CONCLUSION: This research presents a reliable deep learning framework for lung cancer detection that outperforms traditional ML approaches on external validation. The results demonstrate its potential for clinical deployment. Future work will focus on prospective validation, interpretability techniques, and integration into hospital workflows to support real-time decision making and regulatory compliance.

PMID:40674395 | DOI:10.1002/cnr2.70240