NPJ Precis Oncol. 2025 Jul 25;9(1):259. doi: 10.1038/s41698-025-01057-7.

ABSTRACT

Pathological complete response (pCR) can guide surgical strategy and postoperative treatments in triple-negative breast cancer (TNBC). In this study, we developed a Breast Cancer Response Prediction (BCRP) model to predict the pCR in patients with TNBC. The BCRP model integrated multi-dimensional longitudinal quantitative imaging features, clinical factors and features from the Breast Imaging Data and Reporting System (BI-RADS). Multi-dimensional longitudinal quantitative imaging features, including deep learning features and radiomics features, were extracted from multiview B-mode and colour Doppler ultrasound images before and after treatment. The BCRP model achieved the areas under the receiver operating curves (AUCs) of 0.94 [95% confidence interval (CI), 0.91-0.98] and 0.84 [95%CI, 0.75-0.92] in the training and external test cohorts, respectively. Additionally, the low BCRP score was an independent risk factor for event-free survival (P < 0.05). The BCRP model showed a promising ability in predicting response to neoadjuvant chemotherapy in TNBC, and could provide valuable information for survival.

PMID:40715366 | DOI:10.1038/s41698-025-01057-7

Magn Reson Med. 2025 Jul 25. doi: 10.1002/mrm.30634. Online ahead of print.

ABSTRACT

PURPOSE: To assess the efficacy of self-supervised deep learning (DL) denoising in reducing measurement variability in 4D-Flow MRI, and to clarify the contributions of physiological variation to cerebrovascular hemodynamics.

METHODS: A self-supervised DL denoising framework was trained on 3D radially sampled 4D-Flow MRI data. The model was evaluated in a prospective test-retest imaging study in which 10 participants underwent multiple 4D-Flow MRI scans. This included back-to-back scans and a single scan interleaved acquisition designed to isolate noise from physiological variations. The effectiveness of DL denoising was assessed by comparing pixelwise velocity and hemodynamic metrics before and after denoising.

RESULTS: DL denoising significantly enhanced the reproducibility of 4D-Flow MRI measurements, reducing the 95% confidence interval of cardiac-resolved velocity from 215 to 142 mm/s in back-to-back scans and from 158 to 96 mm/s in interleaved scans, after adjusting for physiological variation. In derived parameters, DL denoising did not significantly improve integrated measures, such as flow rates, but did significantly improve noise sensitive measures, such as pulsatility index. Physiologic variation in back-to-back time-resolved scans contributed 26.37% ± 0.08% and 32.42% ± 0.05% of standard error before and after DL.

CONCLUSION: Self-supervised DL denoising enhances the quantitative repeatability of 4D-Flow MRI by reducing technical noise; however, variations from physiology and post-processing are not removed. These findings underscore the importance of accounting for both technical and physiological variability in neurovascular flow imaging, particularly for studies aiming to establish biomarkers for neurodegenerative diseases with vascular contributions.

PMID:40711943 | DOI:10.1002/mrm.30634

Dentomaxillofac Radiol. 2025 Jul 25:twaf057. doi: 10.1093/dmfr/twaf057. Online ahead of print.

ABSTRACT

OBJECTIVES: Precise 3D segmentation of paranasal sinuses is essential for accurate diagnosis and treatment. This study aimed to develop a fully automated segmentation algorithm for the paranasal sinuses using the nnU-Net v2 architecture.

METHODS: The nnU-Net v2-based segmentation algorithm was developed using Python 3.6.1 and the PyTorch library, and its performance was evaluated on a dataset of 97 cone-beam computed tomography (CBCT) scans. Ground truth annotations were manually generated by expert radiologists using the 3D Slicer software, employing a polygonal labeling technique across sagittal, coronal, and axial planes. Model performance was assessed using several quantitative metrics, including accuracy, Dice Coefficient (DC), sensitivity, precision, Jaccard Index, Area Under the Curve (AUC), and 95% Hausdorff Distance (95% HD).

RESULTS: The nnU-Net v2-based algorithm demonstrated high segmentation performance across all paranasal sinuses. Dice Coefficient (DC) values were 0.94 for the frontal, 0.95 for the sphenoid, 0.97 for the maxillary, and 0.88 for the ethmoid sinuses. Accuracy scores exceeded 99% for all sinuses. The 95% Hausdorff Distance (95% HD) values were 0.51 mm for both the frontal and maxillary sinuses, 0.85 mm for the sphenoid sinus, and 1.17 mm for the ethmoid sinus. Jaccard indices were 0.90, 0.91, 0.94, and 0.80, respectively.

CONCLUSIONS: This study highlights the high accuracy and precision of the nnU-Net v2-based CNN model in the fully automated segmentation of paranasal sinuses from CBCT images. The results suggest that the proposed model can significantly contribute to clinical decision-making processes, facilitating diagnostic and therapeutic procedures.

PMID:40711942 | DOI:10.1093/dmfr/twaf057

IEEE Trans Med Imaging. 2025 Jul 25;PP. doi: 10.1109/TMI.2025.3592625. Online ahead of print.

ABSTRACT

In recent years, deep learning has been widely utilized in the fields of biomedical image segmentation and cellular image analysis. Supervised deep neural networks trained on annotated data have demonstrated good performance in tasks related to cell and nucleus segmentation. However, the use of supervised models necessitates carefully constructed training data and a substantial amount of ground truth information. Unfortunately, high-quality annotated data for cellular images are scarce. To address the issue of limited datasets, we propose a cell and nucleus segmentation method based on deep learning multimodal fusion. The proposed method includes three modules: a segmentation fundamental module, a multimodal prompter module, and an object output module. This comprehensive approach enables cell and nucleus segmentation tasks to be performed without the need for retraining on new data. The segmentation fundamental module is the core of the framework, as it provides essential segmentation capabilities. By leveraging preexisting models trained on natural imagery, this module effectively performs cell segmentation by incorporating prior knowledge. The multimodal prompter module, a pretrained model, aids in combining image and textual information. It employs a data fusion technique for multiple modalities to deliver prompts that steer the network's output, thereby avoiding the constraints inherent to single-modality approaches. The object output module combines the inputs from the preceding modules to generate the final segmentation output. The experimental validation confirms the superiority of the proposed method, which outperforms comparative methods in cell and nucleus segmentation tasks and has promise for future applications in cell tracking.

PMID:40711898 | DOI:10.1109/TMI.2025.3592625

J Chem Inf Model. 2025 Jul 25. doi: 10.1021/acs.jcim.5c01020. Online ahead of print.

ABSTRACT

To understand the benefits and drawbacks of 3D combinatorial and deep learning generators, a novel benchmark was created focusing on the recreation of important protein-ligand interactions and 3D ligand conformations. Using the BindingMOAD data set with a hold-out blind set, the sequential graph neural network generators, Pocket2Mol and PocketFlow, diffusion models, DiffSBDD and MolSnapper, and combinatorial genetic algorithms, AutoGrow4 and LigBuilderV3, were evaluated. It was discovered that deep learning methods fail to generate structurally valid molecules and 3D conformations, whereas combinatorial methods are slow and generate molecules that are prone to failing 2D MOSES filters. The results from this evaluation guide us toward improving deep learning structure-based generators by placing higher importance on structural validity, 3D ligand conformations, and recreation of important known active site interactions. This benchmark should be used to understand the limitations of future combinatorial and deep learning generators. The package is freely available under an Apache 2.0 license at github.com/gskcheminformatics/SBDD-benchmarking.

PMID:40711830 | DOI:10.1021/acs.jcim.5c01020

ACS Nano. 2025 Jul 25. doi: 10.1021/acsnano.5c04200. Online ahead of print.

ABSTRACT

This review highlights the recent transformative impact of artificial intelligence (AI), machine learning (ML), and deep learning (DL) on materials science, emphasizing their applications in materials discovery, development, and optimization. AI-driven methods have revolutionized materials discovery through structure generation, property prediction, high-throughput (HT) screening, and computational design while advancing development with improved characterization and autonomous experimentation. Optimization has also benefited from AI's ability to enhance materials design and processes. The review will introduce fundamental AI and ML concepts, including supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning (RL), alongside advanced DL models such as recurrent neural networks (RNNs), convolutional neural networks (CNNs), graph neural networks (GNNs), generative models, and Transformer-based models, which are critical for analyzing complex material data sets. It also covers core topics in materials informatics, including structure-property relationships, material descriptors, quantitative structure-property relationships (QSPR), and strategies for managing missing data and small data sets. Despite these advancements, challenges such as inconsistent data quality, limited model interpretability, and a lack of standardized data-sharing frameworks persist. Future efforts will focus on improving robustness, integrating causal reasoning and physics-informed AI, and leveraging multimodal models to enhance scalability and transparency, unlocking new opportunities for more advanced materials discovery, development, and optimization. Furthermore, the integration of quantum computing with AI will enable faster and more accurate results, and ethical frameworks will ensure responsible human-AI collaboration, addressing concerns of bias, transparency, and accountability in decision-making.

PMID:40711807 | DOI:10.1021/acsnano.5c04200

Endocrine. 2025 Jul 25. doi: 10.1007/s12020-025-04351-3. Online ahead of print.

ABSTRACT

PURPOSE: Postoperative remission in pituitary adenoma (PA) patients significantly affects treatment outcomes and quality of life. Accurate prediction of remission is crucial for neurosurgeons and oncologists as it aids in personalizing treatment plans, optimizing follow-up care, and preventing unnecessary interventions. Unlike diagnostic classification, this review specifically focuses on remission prediction as a distinct prognostic application of AI. This systematic review and meta-analysis aim to evaluate the performance of machine learning (ML) algorithms in predicting remission outcomes in PA patients.

METHODS: A comprehensive search of PubMed, Scopus, Embase, Web of Science, and the google scholar was conducted to identify eligible studies until Dec 2024. Data on sensitivity, specificity, accuracy, precision, F1-score, and area under the curve (AUC) were extracted from the included studies.

RESULTS: Out of 1530 studies screened, 10 met our eligibility criteria involving ML approaches in patients with confirmed PA. ML algorithms, particularly artificial neural networks (ANN), offer promising performance for predicting remission outcomes in PA patients. Meta-analysis of 10 studies resulted in a pooled sensitivity of 0.84 (95% CI: 0.74-0.91), specificity of 0.84 (95% CI: 0.74-0.91), positive diagnostic likelihood ratio (DLR) of 0.19 (95% CI: 0.11-0.32), negative DLR of 15.26 (95% CI: 8.23-28.26), diagnostic odds ratio (DOR) of 28.25 (95% CI: 10.85-73.57), the diagnostic score was 3.34 (95% CI: 2.38-4.3) and an AUC of 0.91 (95% CI: 0.88-0.93).

CONCLUSION: ML-based models demonstrate moderate to high diagnostic accuracy in predicting remission outcomes in PA patients. While these models show promise in enhancing clinical decision-making post-surgery, further prospective validation and larger studies are necessary before their routine clinical integration.

PMID:40711667 | DOI:10.1007/s12020-025-04351-3

Eur Radiol. 2025 Jul 25. doi: 10.1007/s00330-025-11835-3. Online ahead of print.

ABSTRACT

OBJECTIVES: Cystic fibrosis (CF) is a prevalent autosomal recessive disorder, with lung complications being the primary cause of morbidity and mortality. In paediatric patients, structural lung changes begin early, necessitating prompt detection to guide treatment and delay disease progression. This study evaluates ultra-low-dose CT (ULDCT) versus chest x-rays (CXR) for children with CF (CwCF) lung disease assessment. ULDCT uses AI-enhanced deep-learning iterative reconstruction to achieve radiation doses comparable to a CXR.

MATERIALS AND METHODS: This prospective study recruited radiographers and radiologists to assess the image quality (IQ) of ten paired ULDCT and CXR images of CwCF from a single centre. Statistical analyses, including the Wilcoxon Signed Rank test and visual grading characteristic (VGC) analysis, compared diagnostic confidence and anatomical detail.

RESULTS: Seventy-five participants were enrolled, 25 radiologists and 50 radiographers. The majority (88%) preferred ULDCT over CXR for monitoring CF lung disease due to higher perceived confidence (p ≤ 0.001) and better IQ ratings (p ≤ 0.05), especially among radiologists (area under the VGC curve and its 95% CI was 0.63 (asymmetric 95% CI: 0.51-0.73; p ≤ 0.05). While ULDCT showed no significant differences in anatomical visualisation compared to CXR, the overall IQ for lung pathology assessment was rated superior.

CONCLUSION: ULDCT offers superior IQ over CXR in CwCF, with similar radiation doses. It also enhances diagnostic confidence, supporting its use as a viable CXR alternative. Standardising CT protocols to optimise IQ and minimise radiation is essential to improve disease monitoring in this vulnerable group.

KEY POINTS: Question How does chest X-ray (CXR) IQ in children compare to ULDCT at similar radiation doses for assessing CF-related lung disease? Findings ULDCT offers superior IQ over CXR in CwCF. Participants preferred ULDCT due to higher perceived confidence levels and superior IQ. Clinical relevance ULDCT can enhance diagnosis in CwCF while maintaining comparable radiation doses. ULDCT also enhances diagnostic confidence, supporting its use as a viable CXR alternative.

PMID:40711551 | DOI:10.1007/s00330-025-11835-3

J Chem Inf Model. 2025 Jul 25. doi: 10.1021/acs.jcim.5c01362. Online ahead of print.

ABSTRACT

Molecular carcinogenicity is a crucial factor in the development of cancer, and accurate prediction of it is vital for cancer prevention, treatment, and drug development. In recent years, deep learning has been applied to predict molecular carcinogenicity, but due to limitations in data quality and feature richness, these methods still need improvement in terms of accuracy, robustness, and interpretability. In this article, we propose a deep transfer learning model based on multimodal information fusion, called MMF-MCP, for molecular feature extraction and carcinogenicity prediction. We extract molecular graph features and fingerprint features using graph attention networks and convolutional neural networks, respectively, and process molecular images through a deep residual network, SE-ResNet18, equipped with a squeeze-and-excitation module. To more effectively utilize limited carcinogenicity data and enhance the model's predictive performance and generalization ability, we further apply a transfer learning strategy by pretraining the model on a molecular mutagenicity data set and then fine-tuning it on the carcinogenicity data set, enabling knowledge transfer and significant improvement in model performance. MMF-MCP achieves average ACC, AUC, SE, and SP scores of 0.8452, 0.8513, 0.8571, and 0.8333 on benchmark data sets for molecular carcinogenicity, significantly outperforming state-of-the-art molecular carcinogenicity prediction methods. Additionally, the visualization results of MMF-MCP on molecular images demonstrate its strong interpretability, providing significant assistance in visually observing and understanding the critical structures and features of molecular carcinogenicity. The source code for MMF-MCP is available at https://github.com/liuliwei1980/MCP.

PMID:40711460 | DOI:10.1021/acs.jcim.5c01362

Vet Sci. 2025 Jul 13;12(7):664. doi: 10.3390/vetsci12070664.

ABSTRACT

With the increasing importance of animal welfare, behavioral indicators such as changes in head and ear posture are widely recognized as non-invasive and field-applicable markers for evaluating the emotional state and stress levels of animals. However, traditional visual observation methods are often subjective, as assessments can vary between observers, and are unsuitable for long-term, quantitative monitoring. This study proposes an artificial intelligence (AI)-based system for the detection and pose estimation of cattle heads and ears using deep learning techniques. The system integrates Mask R-CNN for accurate object detection and FSA-Net for robust 3D pose estimation (yaw, pitch, and roll) of cattle heads and left ears. Comprehensive datasets were constructed from images of Japanese Black cattle, collected under natural conditions and annotated for both detection and pose estimation tasks. The proposed framework achieved mean average precision (mAP) values of 0.79 for head detection and 0.71 for left ear detection and mean absolute error (MAE) of approximately 8-9° for pose estimation, demonstrating reliable performance across diverse orientations. This approach enables long-term, quantitative, and objective monitoring of cattle behavior, offering significant advantages over traditional subjective stress assessment methods. The developed system holds promise for practical applications in animal welfare research and real-time farm management.

PMID:40711324 | DOI:10.3390/vetsci12070664

Nanomaterials (Basel). 2025 Jul 17;15(14):1112. doi: 10.3390/nano15141112.

ABSTRACT

Accurate identification of active-layer compositions in organic photovoltaic (OPV) devices often relies on invasive techniques such as electrical measurements or material extraction, which risk damaging the device. In this study, we propose a non-invasive classification approach based on simulated full-device absorption spectra. To account for fabrication-related variability, the active-layer thickness varied by over ±15% around the optimal value, creating a realistic and diverse training dataset. A multilayer perceptron (MLP) neural network was applied with various activation functions, optimization algorithms, and data split ratios. The optimized model achieved classification accuracies exceeding 99% on both training and testing sets, with minimal sensitivity to random initialization or data partitioning. These results demonstrate the potential of applying deep learning to spectral data for reliable, non-destructive OPV composition classification, paving the way for integration into automated manufacturing diagnostics and quality control workflows.

PMID:40711231 | DOI:10.3390/nano15141112

Toxics. 2025 Jun 23;13(7):525. doi: 10.3390/toxics13070525.

ABSTRACT

Drug toxicity prediction plays a crucial role in the drug research and development process, ensuring clinical drug safety. However, traditional methods are hampered by high cost, low throughput, and uncertainty of cross-species extrapolation, which has become a key bottleneck restricting the efficiency of new drug research and development. The breakthrough development of Artificial Intelligence (AI) technology, especially the application of deep learning and multimodal data fusion strategy, is reshaping the scientific paradigm of drug toxicology assessment. In this review, we focus on the application of AI in the field of drug toxicity prediction and systematically summarize the relevant literature and development status globally in the past years. The application of various toxicity databases in the prediction was elaborated in detail, and the research results and methods for the prediction of different toxicity endpoints were analyzed in depth, including acute toxicity, carcinogenicity, organ-specific toxicity, etc. Furthermore, this paper discusses the application progress of AI technologies (e.g., machine learning and deep learning model) in drug toxicity prediction, analyzes their advantages and challenges, and outlines the future development direction. It aims to provide a comprehensive and in-depth theoretical framework and actionable technical strategies for toxicity prediction in drug development.

PMID:40710970 | DOI:10.3390/toxics13070525

Tomography. 2025 Jun 30;11(7):78. doi: 10.3390/tomography11070078.

ABSTRACT

The rapid growth of artificial intelligence, particularly in the field of deep learning, has opened up new advances in analyzing and processing large and complex datasets. Prospects and emerging trends in this area engage the development of methods, techniques, and algorithms to build autonomous systems that perform tasks with minimal human action. In medical practice, radiological imaging technologies systematically boost progress in the clinical monitoring of cancer through the information that can be analyzed in these images. This review gives insight into deep learning-based approaches that strengthen the assessment of the response to the treatment of non-small-cell lung cancer. This systematic survey delves into the various approaches to morphological and metabolic changes observed in computerized tomography (CT) and positron emission tomography (PET) imaging. We highlight the challenges and opportunities for feasible integration of deep learning computer-based tools in evaluating treatments in lung cancer patients, after which CT and PET-based strategies are contrasted. The investigated deep learning methods are organized and described as instruments for classification, clustering, and prediction, which can contribute to the design of automated and objective assessment of lung tumor responses to treatments.

PMID:40710896 | DOI:10.3390/tomography11070078

J Imaging. 2025 Jul 20;11(7):244. doi: 10.3390/jimaging11070244.

ABSTRACT

In clinical practice, left ventricle hypertrophy (LVH) continues to pose a considerable challenge, highlighting the need for more reliable diagnostic approaches. This study aims to propose an automated framework for the quantification of LVH extent and the classification of myocardial segments according to hypertrophy severity using a deep learning-based algorithm. The proposed method was validated on 133 subjects, including both healthy individuals and patients with LVH. The process starts with automatic LV segmentation using U-Net and the segmentation of the left ventricle cavity based on the American Heart Association (AHA) standards, followed by the division of each segment into three equal sub-segments. Then, an automated quantification of regional wall thickness (RWT) was performed. Finally, a convolutional neural network (CNN) was developed to classify each myocardial sub-segment according to hypertrophy severity. The proposed approach demonstrates strong performance in contour segmentation, achieving a Dice Similarity Coefficient (DSC) of 98.47% and a Hausdorff Distance (HD) of 6.345 ± 3.5 mm. For thickness quantification, it reaches a minimal mean absolute error (MAE) of 1.01 ± 1.16. Regarding segment classification, it achieves competitive performance metrics compared to state-of-the-art methods with an accuracy of 98.19%, a precision of 98.27%, a recall of 99.13%, and an F1-score of 98.7%. The obtained results confirm the high performance of the proposed method and highlight its clinical utility in accurately assessing and classifying cardiac hypertrophy. This approach provides valuable insights that can guide clinical decision-making and improve patient management strategies.

PMID:40710630 | DOI:10.3390/jimaging11070244

J Imaging. 2025 Jul 18;11(7):243. doi: 10.3390/jimaging11070243.

ABSTRACT

Accurate analysis of medical videos remains a major challenge in deep learning (DL) due to the need for effective spatiotemporal feature mapping that captures both spatial detail and temporal dynamics. Despite advances in DL, most existing models in medical AI focus on static images, overlooking critical temporal cues present in video data. To bridge this gap, a novel DL-based framework is proposed for spatiotemporal feature extraction from medical video sequences. As a feasibility use case, this study focuses on gastrointestinal (GI) endoscopic video classification. A 3D convolutional neural network (CNN) is developed to classify upper and lower GI endoscopic videos using the hyperKvasir dataset, which contains 314 lower and 60 upper GI videos. To address data imbalance, 60 matched pairs of videos are randomly selected across 20 experimental runs. Videos are resized to 224 × 224, and the 3D CNN captures spatiotemporal information. A 3D version of the parallel spatial and channel squeeze-and-excitation (P-scSE) is implemented, and a new block called the residual with parallel attention (RPA) block is proposed by combining P-scSE3D with a residual block. To reduce computational complexity, a (2 + 1)D convolution is used in place of full 3D convolution. The model achieves an average accuracy of 0.933, precision of 0.932, recall of 0.944, F1-score of 0.935, and AUC of 0.933. It is also observed that the integration of P-scSE3D increased the F1-score by 7%. This preliminary work opens avenues for exploring various GI endoscopic video-based prospective studies.

PMID:40710629 | DOI:10.3390/jimaging11070243

J Cardiovasc Electrophysiol. 2025 Jul 25. doi: 10.1111/jce.70028. Online ahead of print.

ABSTRACT

BACKGROUND: Screening of asymptomatic/occult atrial fibrillation (AF) remains challenging. This study aimed to use a deep learning model to predict hidden AF in patients who showed normal sinus rhythm (SR) during 24-h Holter monitoring.

METHODS: This was a retrospective cohort study that enrolled 934 patients receiving 24-h ambulatory Holter monitoring. Of them, 640 patients (AF group) had the documented paroxysmal AF in the index Holter monitoring. The rest of 294 patients (Control group) did not have medical record of AF and the index Holter exam did not detect AF. A ConvNeXt model (1st stage) and Long Short-Term Memory (LSTM) (2nd stage) was used to predict the probability of AF.

RESULTS: 368,550 eligible SR ECG segments (60 s/segment) were taken into 1st staged classification, and the Area Under Curve (AUC) was 0.7755 with accuracy of 0.7755, sensitivity of 0.9105, and specificity of 0.5718. After 2nd staged classification (10-min SR ECG recording), the AUC reached 0.874 (accuracy of 0.8213, sensitivity of 0.8339, and specificity of 0.8115). The fact that longer time length examined in the 2nd stage leaded to a dilution of features more related to AF might decrease specificity compared with 1st stage. Subgroup analysis demonstrated that night-time settings had better performance (AUC: 0.902 in night-time, 0.8726 in day-time).

CONCLUSION: We developed an AI-enabled model with 10 min ECG recording from ambulatory Holter monitoring during SR to predict hidden AF with high accuracy. Subgroup analysis according to the diurnal period, night-time settings showed more favorable performance compared to day-time recordings.

PMID:40709543 | DOI:10.1111/jce.70028

ArXiv [Preprint]. 2025 Jul 14:arXiv:2507.09996v1.

ABSTRACT

OBJECTIVE: This study aims to support early diagnosis of Alzheimer's disease and detection of amyloid accumulation by leveraging the microstructural information available in multi-shell diffusion MRI (dMRI) data, using a vision transformer-based deep learning framework.

METHODS: We present a classification pipeline that employs the Swin Transformer, a hierarchical vision transformer model, on multi-shell dMRI data for the classification of Alzheimer's disease and amyloid presence. Key metrics from DTI and NODDI were extracted and projected onto 2D planes to enable transfer learning with ImageNet-pretrained models. To efficiently adapt the transformer to limited labeled neuroimaging data, we integrated Low-Rank Adaptation. We assessed the framework on diagnostic group prediction (cognitively normal, mild cognitive impairment, Alzheimer's disease dementia) and amyloid status classification.

RESULTS: The framework achieved competitive classification results within the scope of multi-shell dMRI-based features, with the best balanced accuracy of 95.2% for distinguishing cognitively normal individuals from those with Alzheimer's disease dementia using NODDI metrics. For amyloid detection, it reached 77.2% balanced accuracy in distinguishing amyloid-positive mild cognitive impairment/Alzheimer's disease dementia subjects from amyloid-negative cognitively normal subjects, and 67.9% for identifying amyloid-positive individuals among cognitively normal subjects. Grad-CAM-based explainability analysis identified clinically relevant brain regions, including the parahippocampal gyrus and hippocampus, as key contributors to model predictions.

CONCLUSION: This study demonstrates the promise of diffusion MRI and transformer-based architectures for early detection of Alzheimer's disease and amyloid pathology, supporting biomarker-driven diagnostics in data-limited biomedical settings.

PMID:40709302 | PMC:PMC12288649

Front Psychol. 2025 Jul 10;16:1567523. doi: 10.3389/fpsyg.2025.1567523. eCollection 2025.

ABSTRACT

The aim of the present study was to analyse the association of bullying and cyberbullying with deep learning, superficial learning, planning and decision making, as well as school procrastination. A total of 1,263 Spanish schoolchildren (51.39% girls) aged 10-16 years (13.23 ± 1.77) participated. The association between variables and the analysis of exposure risk was performed by analysis of covariance (ANCOVA) and binary logistic regression, respectively. All analyses were conducted separately for boys and girls and adjusted for age, body mass index, mother's education and average weekly physical activity. Results showed that girls who were victims of bullying and cyberbullying had significantly higher procrastination toward class tasks (7 and 16%, respectively). In addition, cyberbullying victims acquire more superficial learning (5.28%). In general, victims of bullying have almost twice the risk of having higher values of superficial learning and procrastination than non-victims. This risk is multiplied by 3 and 4, respectively, in the case of cyberbullying victims. On the other hand, bullying aggressors were also found to have high superficial learning (7.34%) and higher procrastination (17.45%). In the case of cyberbullying, aggressors also had more superficial learning (boys = 13.38% and girls = 9.56%), worse values in planning and decision making (boys = 3.82% and girls = 3.3%) and more procrastination (boys = 16.81% and girls = 20.48%). In both sexes, the risk of exposure to aggression toward the above variables is multiplied by 8, 2, and 10, respectively. All these findings reveal that bullying and cyberbullying can affect young people in key learning variables, beyond those of physical, psychological or socio-emotional aspects already known. Immediate and systematic actions are needed to monitor and prevent bullying and cyberbullying inside and outside the school context, creating safe spaces and providing counseling for both victims and aggressors.

PMID:40709227 | PMC:PMC12286959 | DOI:10.3389/fpsyg.2025.1567523

Nat Mach Intell. 2025;7(7):1053-1075. doi: 10.1038/s42256-025-01052-4. Epub 2025 Jul 22.

ABSTRACT

Cancers are commonly characterized by a complex pathology encompassing genetic, microscopic and macroscopic features, which can be probed individually using imaging and omics technologies. Integrating these data to obtain a full understanding of pathology remains challenging. We introduce a method called deep latent variable path modelling, which combines the representational power of deep learning with the capacity of path modelling to identify relationships between interacting elements in a complex system. To evaluate the capabilities of deep latent variable path modelling, we initially trained a model to map dependencies between single-nucleotide variant, methylation profiles, microRNA sequencing, RNA sequencing and histological data using breast cancer data from The Cancer Genome Atlas. This method exhibited superior performance in mapping associations between data types compared with classical path modelling. We additionally performed successful applications of the model to stratify single-cell data, identify synthetic lethal interactions using CRISPR-Cas9 screens derived from cell lines and detect histologic-transcriptional associations using spatial transcriptomic data. Results from each of these data types can then be understood with reference to the same holistic model of illness.

PMID:40709098 | PMC:PMC12283373 | DOI:10.1038/s42256-025-01052-4

Bioeng Transl Med. 2025 Feb 3;10(4):e70002. doi: 10.1002/btm2.70002. eCollection 2025 Jul.

ABSTRACT

Cardiovascular diseases (CVDs) continue to drive global mortality rates, underscoring an urgent need for advancements in healthcare solutions. The development of point-of-care (POC) devices that provide rapid diagnostic services near patients has garnered substantial attention, especially as traditional healthcare systems face challenges such as delayed diagnoses, inadequate care, and rising medical costs. The advancement of machine learning techniques has sparked considerable interest in medical research and engineering, offering ways to enhance diagnostic accuracy and relevance. Improved data interoperability and seamless connectivity could enable real-time, continuous monitoring of cardiovascular health. Recent breakthroughs in computing power and algorithmic design, particularly deep learning frameworks that emulate neural processes, have revolutionized POC devices for CVDs, enabling more frequent detection of abnormalities and automated, expert-level diagnosis. However, challenges such as data privacy concerns and biases in dataset representation continue to hinder clinical integration. Despite these barriers, the translational potential of machine learning-assisted POC devices presents significant opportunities for advancement in CVDs healthcare.

PMID:40708978 | PMC:PMC12284442 | DOI:10.1002/btm2.70002