PLoS One. 2025 Apr 25;20(4):e0314837. doi: 10.1371/journal.pone.0314837. eCollection 2025.

ABSTRACT

The accurate prediction of RNA secondary structure, and pseudoknots in particular, is of great importance in understanding the functions of RNAs since they give insights into their folding in three-dimensional space. However, existing approaches often face computational challenges or lack precision when dealing with long RNA sequences and/or pseudoknots. To address this, we propose a divide-and-conquer method based on deep learning, called DivideFold, for predicting the secondary structures including pseudoknots of long RNAs. Our approach is able to scale to long RNAs by recursively partitioning sequences into smaller fragments until they can be managed by an existing model able to predict RNA secondary structure including pseudoknots. We show that our approach exhibits superior performance compared to state-of-the-art methods for pseudoknot prediction and secondary structure prediction including pseudoknots for long RNAs. The source code of DivideFold, along with all the datasets used in this study, is accessible at https://evryrna.ibisc.univ-evry.fr/evryrna/dividefold/home.

PMID:40279361 | DOI:10.1371/journal.pone.0314837

Curr Opin Ophthalmol. 2025 Apr 22. doi: 10.1097/ICU.0000000000001150. Online ahead of print.

ABSTRACT

PURPOSE OF REVIEW: The integration of artificial intelligence (AI) in the diagnosis and management of anterior segment diseases has rapidly expanded, demonstrating significant potential to revolutionize clinical practice.

RECENT FINDINGS: AI technologies, including machine learning and deep learning models, are increasingly applied in the detection and management of a variety of conditions, such as corneal diseases, refractive surgery, cataract, conjunctival disorders (e.g., pterygium), trachoma, and dry eye disease. By analyzing large-scale imaging data and clinical information, AI enhances diagnostic accuracy, predicts treatment outcomes, and supports personalized patient care.

SUMMARY: As AI models continue to evolve, particularly with the use of large models and generative AI techniques, they will further refine diagnosis and treatment planning. While challenges remain, including issues related to data diversity and model interpretability, AI's integration into ophthalmology promises to improve healthcare outcomes, making it a cornerstone of data-driven medical practice. The continued development and application of AI will undoubtedly transform the future of anterior segment ophthalmology, leading to more efficient, accurate, and individualized care.

PMID:40279352 | DOI:10.1097/ICU.0000000000001150

IEEE Trans Biomed Eng. 2025 Apr 25;PP. doi: 10.1109/TBME.2025.3562067. Online ahead of print.

ABSTRACT

Sleep spindles, which are key biomarkers of non-rapid eye movement stage 2 sleep, play a crucial role in predicting outcomes for patients with acute disorders of consciousness (ADOC). However, several critical challenges remain in spindle detection: 1) the limited use of automated spindle detection in ADOC; 2) the difficulty in identifying low-frequency spindles in patient populations; and 3) the lack of effective tools for quantitatively analyzing the relationship between spindle density and patient outcomes. To address these challenges, we propose a novel Deep Learning-Augmented algorithm for automated sleep spindle detection in ADOC patients. This method combines Convolutional Neural Networks with decision tree-assisted validation, using wavelet transform principles to enhance detection accuracy and sensitivity, especially for the slow spindles commonly found in ADOC patients. Our approach not only demonstrates superior performance and reliability but also has the potential to significantly improve diagnostic precision and guide treatment strategies when integrated into clinical practice. Our algorithm was evaluated on the Montreal Archive of Sleep Studies - Session 2 (MASS SS2, n = 19), achieving average F1 scores of 0.798 and 0.841 compared to annotations from two experts. On a self-recorded dataset from ADOC patients (n = 24), it achieved an F1 score of 0.745 compared to expert annotations. Additionally, our analysis using the Spearman correlation coefficient revealed a moderate positive correlation between sleep spindle density and 28-day Glasgow Outcome Scale scores in ADOC patients. This suggests that spindle density could serve as a prognostic marker for predicting clinical outcomes and guiding personalized patient care.

PMID:40279237 | DOI:10.1109/TBME.2025.3562067

IEEE Trans Biomed Eng. 2025 Apr 25;PP. doi: 10.1109/TBME.2025.3564567. Online ahead of print.

ABSTRACT

A transfer function approach has recently proven effective for calibrating deep learning (DL) algorithms in quantitative ultrasound (QUS), addressing data shifts at both the acquisition and machine levels. Expanding on this approach, we develop a strategy to acquire the functionality of a DL model from one ultrasound machine and implement it on another in a black-box setting, in the context of QUS. This demonstrates the ease with which the functionality of a DL model can be transferred between machines. While the proposed approach can also assist regulatory bodies in comparing and approving DL models, it also highlights the security risks associated with deploying such models in a commercial scanner for clinical use. The method is a black-box unsupervised domain adaptation technique that integrates the transfer function approach with an iterative schema. It does not utilize any information related to model internals but it solely relies on the availability of an input-output interface. Additionally, we assume the availability of unlabeled data from a testing machine. This scenario could become relevant as companies begin deploying their DL functionalities for clinical use. In the experiments, we used a SonixOne and a Verasonics machine. The model was trained on SonixOne data, and its functionality was then transferred to the Verasonics machine. The proposed method successfully transferred the functionality to the Verasonics machine, achieving a remarkable 98 percent classification accuracy in a binary decision task. This study underscores the need to establish security measures prior to deploying DL models in clinical settings.

PMID:40279236 | DOI:10.1109/TBME.2025.3564567

Mol Divers. 2025 Apr 25. doi: 10.1007/s11030-025-11197-4. Online ahead of print.

ABSTRACT

Predicting molecular properties with high accuracy is essential across scientific fields, from drug discovery and biotechnology to materials science and environmental research. In biomedical sciences, accurate molecular property prediction is crucial for elucidating disease mechanisms, identifying potential drug candidates, and optimising various processes. However, existing approaches, often based on low-dimensional representations, fail to capture the intricate spatial and structural complexities of molecular data. This study introduces a novel hybrid deep learning model, the Graph-Aware AURA-LSTM (Attentive Unified Representation Architecture-Long Short-Term Memory), designed to determine molecular properties with unprecedented accuracy using advanced graphical representations. AURA-LSTM combines multiple Graph Neural Network (GNN) architectures, specifically Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), and Graph Isomorphism Networks (GINs), in a parallel structure to comprehensively capture the multidimensional structural features of molecules. Within this architecture, GCNs incorporate local structural relationships, GATs apply attention mechanisms to highlight critical structural elements, and GINs capture intricate molecular details through isomorphic distinction, resulting in a richly detailed feature matrix. The feature layer then processes this BiLSTM matrix, which evaluates temporal relationships to enhance molecular feature classification. Evaluated on eight benchmark datasets, AURA-LSTM demonstrated superior performance, consistently achieving over 90% accuracy and outperforming state-of-the-art methods. These results position AURA-LSTM as a robust tool for molecular feature classification, uniquely capable of integrating temporally aware insights from distinct GNN architectures.

PMID:40279083 | DOI:10.1007/s11030-025-11197-4

Discov Oncol. 2025 Apr 25;16(1):614. doi: 10.1007/s12672-025-02346-0.

ABSTRACT

OBJECTIVE: Oral cancer is an important health challenge worldwide and accurate survival time prediction of this disease can guide treatment decisions. This study aims to propose a deep learning-based model, DeepOmicsSurv, to predict survival in oral cancer patients using clinical and multi-omics data.

METHODS: DeepOmicsSurv builds on the DeepSurv model, incorporating multi-head attention convolutional layers, dropout, pooling, and batch normalization to boost its strength and precision. Various dimensionality reduction techniques, including Principal Component Analysis (PCA), Kernel PCA, Non-Negative Matrix Factorization (NMF), Singular Value Decomposition (SVD), Partial Least Squares (PLS), Multidimensional Scaling (MDS), and Autoencoders, were employed to manage the high-dimensional omics data. The model's performance was evaluated against DeepSurv, DeepHit, Cox Proportional Hazards (CoxPH), Convolutional Neural Networks (CNN), and Recurrent Neural Networks (RNN). Additionally, SHapley Additive Explanations (SHAP) was used to analyze the impact of clinical features on survival predictions.

RESULTS: DeepOmicsSurv achieved a C-index of 0.966, MSE of 0.0138, RMSE of 0.1174, MAE of 0.0795, and MedAE of 0.0515, outperforming other deep learning models. Among various dimensionality reduction techniques, autoencoder performed the best with DeepOmicsSurv. SHAP analysis showed that Age, AJCC N Stage, alcohol history and patient smoking history are prevalent clinical features for survival time.

CONCLUSION: In conclusion, DeepOmicsSurv has the potential to predict survival time in oral cancer patients. This model achieved high accuracy with various data types including Clinical, DNAmethylation + clinical, mRNA + clinical, Copy number alteration + clinical, or multi-omics data. Additionally, SHAP analysis reveals clinical factors that influence survival time.

PMID:40278990 | DOI:10.1007/s12672-025-02346-0

Tomography. 2025 Apr 17;11(4):48. doi: 10.3390/tomography11040048.

ABSTRACT

BACKGROUND: Deep learning (DL) reconstruction techniques have shown promise in reducing MRI acquisition times while maintaining image quality. However, the impact of different acceleration factors on diagnostic accuracy in shoulder MRI remains unexplored in clinical practice.

PURPOSE: The purpose of this study was to evaluate the diagnostic accuracy of 2-fold and 4-fold DL-accelerated shoulder MRI protocols compared to standard protocols in clinical practice.

MATERIALS AND METHODS: In this prospective single-center study, 88 consecutive patients (49 males, 39 females; mean age, 51 years) underwent shoulder MRI examinations using standard, 2-fold (DL2), and 4-fold (DL4) accelerated protocols between June 2023 and January 2024. Four independent radiologists (experience range: 4-25 years) evaluated the presence of bone marrow edema (BME), rotator cuff tears, and labral lesions. The sensitivity, specificity, and interobserver agreement were calculated. Diagnostic confidence was assessed using a 4-point scale. The impact of reader experience was analyzed by stratifying the radiologists into ≤10 and >10 years of experience.

RESULTS: Both accelerated protocols demonstrated high diagnostic accuracy. For BME detection, DL2 and DL4 achieved 100% sensitivity and specificity. In rotator cuff evaluation, DL2 showed a sensitivity of 98-100% and specificity of 99-100%, while DL4 maintained a sensitivity of 95-98% and specificity of 99-100%. Labral tear detection showed perfect sensitivity (100%) with DL2 and slightly lower sensitivity (89-100%) with DL4. Interobserver agreement was excellent across the protocols (Kendall's W = 0.92-0.98). Reader experience did not significantly impact diagnostic performance. The area under the ROC curve was 0.94 for DL2 and 0.90 for DL4 (p = 0.32).

CLINICAL IMPLICATIONS: The implementation of DL-accelerated protocols, particularly DL2, could improve workflow efficiency by reducing acquisition times by 50% while maintaining diagnostic reliability. This could increase patient throughput and accessibility to MRI examinations without compromising diagnostic quality.

CONCLUSIONS: DL-accelerated shoulder MRI protocols demonstrate high diagnostic accuracy, with DL2 showing performance nearly identical to that of the standard protocol. While DL4 maintains acceptable diagnostic accuracy, it shows a slight sensitivity reduction for subtle pathologies, particularly among less experienced readers. The DL2 protocol represents an optimal balance between acquisition time reduction and diagnostic confidence.

PMID:40278715 | DOI:10.3390/tomography11040048

Tomography. 2025 Apr 1;11(4):41. doi: 10.3390/tomography11040041.

ABSTRACT

INTRODUCTION: An efficient pipeline for rosette trajectory magnetic resonance imaging reconstruction is proposed, combining the inverse Fourier transform with a vision transformer (ViT) network enhanced with a convolutional layer. This method addresses the challenges of reconstructing high-quality images from non-Cartesian data by leveraging the ViT's ability to handle complex spatial dependencies without extensive preprocessing.

MATERIALS AND METHODS: The inverse fast Fourier transform provides a robust initial approximation, which is refined by the ViT network to produce high-fidelity images.

RESULTS AND DISCUSSION: This approach outperforms established deep learning techniques for normalized root mean squared error, peak signal-to-noise ratio, and entropy-based image quality scores; offers better runtime performance; and remains competitive with respect to other metrics.

PMID:40278708 | DOI:10.3390/tomography11040041

Toxins (Basel). 2025 Apr 1;17(4):171. doi: 10.3390/toxins17040171.

ABSTRACT

The aaccurate prediction of enzymes with environment detoxification functions is crucial, not only to achieve a better understanding of bioremediation strategies, but also to alleviate environmental pollution. In the present study, a novel machine learning model was introduced which classifies enzymes by their toxin degradation ability. In this model, two different sets of data were used which include enzymes that can catalyze the toxin degradation as a positive dataset and non-toxin-degrading enzymes as a negative dataset. Further, a comparison of multiple classifiers was performed to find the best model and a Random Forest (RF) classifier was selected due to its strong performance. To enhance the accuracy, we combined RF with a Deep Neural Network (DNN), forming an ensemble model which effectively integrated both techniques. This combination achieved 95% precision, surpassing individual models. Our ensemble model not only ensures high prediction accuracy but also reliably differentiates toxin-degrading enzymes from non-degrading ones. This study highlights the power of combining classical machine learning with deep learning to advance prediction. Our model represents a significant step in enzyme classification and serves as a valuable resource for environmental biotechnology, food nutrition, and health applications.

PMID:40278669 | DOI:10.3390/toxins17040171

Toxins (Basel). 2025 Mar 22;17(4):156. doi: 10.3390/toxins17040156.

ABSTRACT

Aflatoxin B1, a toxic carcinogen frequently contaminating almonds, nuts, and food products, poses significant health risks. Therefore, a rapid and non-destructive detection method is crucial to detect aflatoxin B1-contaminated almonds to ensure food safety. This study introduces a novel deep learning approach utilizing 3D Inception-ResNet architecture with fine-tuning to classify aflatoxin B1-contaminated almonds using hyperspectral images. The proposed model achieved higher classification accuracy than traditional methods, such as support vector machine (SVM), random forest (RF), quadratic discriminant analysis (QDA), and decision tree (DT), for classifying aflatoxin B1 contaminated almonds. A feature selection algorithm was employed to enhance processing efficiency and reduce spectral dimensionality while maintaining high classification accuracy. Experimental results demonstrate that the proposed 3D Inception-ResNet (Lightweight) model achieves superior classification performance with a 90.81% validation accuracy, an F1-score of 0.899, and an area under the curve value of 0.964, outperforming traditional machine learning approaches. The Lightweight 3D Inception-ResNet model, with 381 layers, offers a computationally efficient alternative suitable for real-time industrial applications. These research findings highlight the potential of hyperspectral imaging combined with deep learning for aflatoxin B1 detection in almonds with higher accuracy. This approach supports the development of real-time automated screening systems for food safety, reducing contamination-related risks in almonds.

PMID:40278655 | DOI:10.3390/toxins17040156

Toxics. 2025 Apr 13;13(4):305. doi: 10.3390/toxics13040305.

ABSTRACT

Bacillus subtilis exhibits a great affinity to soluble U(VI) through non-reducing biomineralization. The pH value, temperature, initial uranium concentration, bacterial concentration, and adsorption time are recognized as the five environmental sensitive factors that can regulate the degree of non-reductive biomineralization. Most of the current studies have focused on the regulatory mechanisms of these factors on uranium non-reductive mineralization. However, there are still few reports on the importance of these factors in influencing non-reductive mineralization, as well as on how to regulate these factors to increase the efficiency of non-reductive mineralization and enhance the enrichment of Bacillus subtilis on uranium. In this work, a deep learning neural network model was constructed to effectively predict the effects of changes in these five environmental sensitivity factors on the non-reducing mineralization of Bacillus subtilis to uranium. Accuracy (99.6%) and R2 (up to 0.89) confirm a high degree of agreement between the predicted output and the observed values. Sensitivity analysis shows that in this model, pH value is the most important influencing factor. However, under different pH values, temperature, initial uranium concentration, adsorption time, and bacterial concentration have different effects. When the pH value is lower than 6, the most important factor is temperature, and once the pH value is greater than 6, the initial concentration is the most important factor. The results are expected to provide a theoretical basis for regulating the enrichment degree of U(VI) by Bacillus subtilis, achieving the maximum long-term stable fixation of U(VI), and understanding the environmental chemical behavior of uranium under different conditions.

PMID:40278621 | DOI:10.3390/toxics13040305

Toxics. 2025 Mar 28;13(4):254. doi: 10.3390/toxics13040254.

ABSTRACT

Surface air pollution affects ecosystems and people's health. However, traditional models have low prediction accuracy. Therefore, a hybrid model for accurately predicting daily surface PM2.5 concentrations was integrated with wavelet (W), convolutional neural network (CNN), bidirectional long short-term memory (BiLSTM), and bidirectional gated recurrent unit (BiGRU). The data for meteorological factors and air pollutants in Guangzhou City from 2014 to 2020 were utilized as inputs to the models. The W-CNN-BiGRU-BiLSTM hybrid model demonstrated strong performance during the predicting phase, achieving an R (correlation coefficient) of 0.9952, a root mean square error (RMSE) of 1.4935 μg/m3, a mean absolute error (MAE) of 1.2091 μg/m3, and a mean absolute percentage error (MAPE) of 7.3782%. Correspondingly, the accurate prediction of surface PM2.5 concentrations is beneficial for air pollution control and urban planning.

PMID:40278570 | DOI:10.3390/toxics13040254

J Intell. 2025 Apr 7;13(4):45. doi: 10.3390/jintelligence13040045.

ABSTRACT

This study proposes a multimodal deep learning model for automated scoring of image-based divergent thinking tests, integrating visual and semantic features to improve assessment objectivity and efficiency. Utilizing 708 Chinese high school students' responses from validated tests, we developed a system combining pretrained ResNet50 (image features) and GloVe (text embeddings), fused through a fully connected neural network with MSE loss and Adam optimization. The training set (603 images, triple-rated consensus scores) showed strong alignment with human scores (Pearson r = 0.810). Validation on 100 images demonstrated generalization capacity (r = 0.561), while participant-level analysis achieved 0.602 correlation with total human scores. Results indicate multimodal integration effectively captures divergent thinking dimensions, enabling simultaneous evaluation of novelty, fluency, and flexibility. This approach reduces manual scoring subjectivity, streamlines assessment processes, and maintains cost-effectiveness while preserving psychometric rigor. The findings advance automated cognitive evaluation methodologies by demonstrating the complementary value of visual-textual feature fusion in creativity assessment.

PMID:40278054 | DOI:10.3390/jintelligence13040045

J Imaging. 2025 Apr 21;11(4):125. doi: 10.3390/jimaging11040125.

ABSTRACT

Computer vision aims to enable machines to understand the visual world. Computer vision encompasses numerous tasks, namely action recognition, object detection and image classification. Much research has been focused on solving these tasks, but one that remains relatively uncharted is light enhancement (LE). Low-light enhancement (LLE) is crucial as computer vision tasks fail in the absence of sufficient lighting, having to rely on the addition of peripherals such as sensors. This review paper will shed light on this (focusing on video enhancement) subfield of computer vision, along with the other forementioned computer vision tasks. The review analyzes both traditional and deep learning-based enhancers and provides a comparative analysis on recent models in the field. The review also analyzes how popular computer vision tasks are improved and made more robust when coupled with light enhancement algorithms. Results show that deep learners outperform traditional enhancers, with supervised learners obtaining the best results followed by zero-shot learners, while computer vision tasks are improved with light enhancement coupling. The review concludes by highlighting major findings such as that although supervised learners obtain the best results, due to a lack of real-world data and robustness to new data, a shift to zero-shot learners is required.

PMID:40278041 | DOI:10.3390/jimaging11040125

Front Med (Lausanne). 2025 Apr 10;12:1574428. doi: 10.3389/fmed.2025.1574428. eCollection 2025.

ABSTRACT

BACKGROUND: Brain tumor categorization on MRI is a challenging but crucial task in medical imaging, requiring high resilience and accuracy for effective diagnostic applications. This study describe a unique multimodal scheme combining the capabilities of deep learning with ensemble learning approaches to overcome these issues.

METHODS: The system integrates three new modalities, spatial feature extraction using a pre-trained VGG19 network, sequential dependency learning using a Bidirectional LSTM, and classification efficiency through a LightGBM classifier.

RESULTS: The combination of both methods leverages the complementary strengths of convolutional neural networks and recurrent neural networks, thus enabling the model to achieve state-of-the-art performance scores. The outcomes confirm the efficacy of this multimodal approach, which achieves a total accuracy of 97%, an F1-score of 0.97, and a ROC AUC score of 0.997.

CONCLUSION: With synergistic harnessing of spatial and sequential features, the model enhances classification rates and effectively deals with high-dimensional data, compared to traditional single-modal methods. The scalable methodology has the possibility of greatly augmenting brain tumor diagnosis and planning of treatment in medical imaging studies.

PMID:40276738 | PMC:PMC12020513 | DOI:10.3389/fmed.2025.1574428

Front Aging Neurosci. 2025 Apr 10;17:1547727. doi: 10.3389/fnagi.2025.1547727. eCollection 2025.

ABSTRACT

INTRODUCTION: Frontotemporal dementia (FTD) is a neurodegenerative disorder characterized by progressive degeneration of the frontal and temporal lobes, leading to significant changes in personality, behavior, and language abilities. Early and accurate differential diagnosis between FTD, its subtypes, and other dementias, such as Alzheimer's disease (AD), is crucial for appropriate treatment planning and patient care. Machine learning (ML) techniques have shown promise in enhancing diagnostic accuracy by identifying complex patterns in clinical and neuroimaging data that are not easily discernible through conventional analysis.

METHODS: This systematic review, following PRISMA guidelines and registered in PROSPERO, aimed to assess the strengths and limitations of current ML models used in differentiating FTD from other neurological disorders. A comprehensive literature search from 2013 to 2024 identified 25 eligible studies involving 6,544 patients with dementia, including 2,984 with FTD, 3,437 with AD, 103 mild cognitive impairment (MCI) and 20 Parkinson's disease dementia or probable dementia with Lewy bodies (PDD/DLBPD).

RESULTS: The review found that Support Vector Machines (SVMs) were the most frequently used ML technique, often applied to neuroimaging and electrophysiological data. Deep learning methods, particularly convolutional neural networks (CNNs), have also been increasingly adopted, demonstrating high accuracy in distinguishing FTD from other dementias. The integration of multimodal data, including neuroimaging, EEG signals, and neuropsychological assessments, has been suggested to enhance diagnostic accuracy.

DISCUSSION: ML techniques showed strong potential for improving FTD diagnosis, but challenges like small sample sizes, class imbalance, and lack of standardization limit generalizability. Future research should prioritize the development of standardized protocols, larger datasets, and explainable AI techniques to facilitate the integration of ML-based tools into real-world clinical practice.

SYSTEMATIC REVIEW REGISTRATION: https://www.crd.york.ac.uk/PROSPERO/view/CRD42024520902.

PMID:40276595 | PMC:PMC12018464 | DOI:10.3389/fnagi.2025.1547727

Head Neck Tumor Segm MR Guid Appl (2024). 2025;15273:179-190. doi: 10.1007/978-3-031-83274-1_13. Epub 2025 Mar 3.

ABSTRACT

The superior soft tissue differentiation provided by MRI may enable more accurate tumor segmentation compared to CT and PET, potentially enhancing adaptive radiotherapy treatment planning. The Head and Neck Tumor Segmentation for MR-Guided Applications challenge (HNTSMRG-24) comprises two tasks: segmentation of primary gross tumor volume (GTVp) and metastatic lymph nodes (GTVn) on T2-weighted MRI volumes obtained at (1) pre-radiotherapy (pre-RT) and (2) mid-radiotherapy (mid-RT). The training dataset consists of data from 150 patients, including MRI volumes of pre-RT, mid-RT, and pre-RT registered to the corresponding mid-RT volumes. Each MRI volume is accompanied by a label mask, generated by merging independent annotations from a minimum of three experts. For both tasks, we propose adopting the nnU-Net V2 framework by the use of a 15-fold cross-validation ensemble instead of the standard number of 5 folds for increased robustness and variability. For pre-RT segmentation, we augmented the initial training data (150 pre-RT volumes and masks) with the corresponding mid-RT data. For mid-RT segmentation, we opted for a three-channel input, which, in addition to the mid-RT MRI volume, comprises the registered pre-RT MRI volume and the corresponding mask. The mean of the aggregated Dice Similarity Coefficient for GTVp and GTVn is computed on a blind test set and determines the quality of the proposed methods. These metrics determine the final ranking of methods for both tasks separately. The final blind testing (50 patients) of the methods proposed by our team, RUG_UMCG, resulted in an aggregated Dice Similarity Coefficient of 0.81 (0.77 for GTVp and 0.85 for GTVn) for Task 1 and 0.70 (0.54 for GTVp and 0.86 for GTVn) for Task 2.

PMID:40276554 | PMC:PMC12018675 | DOI:10.1007/978-3-031-83274-1_13

Phys Imaging Radiat Oncol. 2025 Mar 8;34:100740. doi: 10.1016/j.phro.2025.100740. eCollection 2025 Apr.

ABSTRACT

BACKGROUND AND PURPOSE: Pancreatic gross tumor volume (GTV) delineation is challenging due to their variable morphology and uncertain ground truth. Previous deep learning-based auto-segmentation methods have struggled to handle tasks with uncertain ground truth and have not accommodated stylistic customizations. We aim to develop a human-in-the-loop pancreatic GTV segmentation tool using Tversky ensembles by leveraging uncertainty estimation techniques.

MATERIAL AND METHODS: In this study, we utilized a total of 282 patients from the pancreas task of the Medical Segmentation Decathlon. Thirty patients were randomly selected to form an independent test set, while the remaining 252 patients were divided into an 80-20 % training-validation split. We incorporated Tversky loss layer during training to train a five-member segmentation ensemble with varying contouring tendencies. The Tversky ensemble predicted probability maps by estimating pixel-level segmentation uncertainties. Probability thresholding was employed on the resulting probability maps to generate the final contours, from which eleven contours were extracted for quantitative evaluation against ground truths, with variations in the threshold values.

RESULTS: Our Tversky ensemble achieved DSC of 0.47, HD95 of 12.70 mm and MSD of 3.24 mm respectively using the optimal thresholding configuration. We outperformed the Swin-UNETR configuration that achieved the state-of-the-art result in the pancreas task of the medical segmentation decathlon.

CONCLUSIONS: Our study demonstrated the effectiveness of employing an ensemble-based uncertainty estimation technique for pancreatic tumor segmentation. The approach provided clinicians with a consensus probability map that could be fine-tuned in line with their preferences, generating contours with greater confidence.

PMID:40276495 | PMC:PMC12019452 | DOI:10.1016/j.phro.2025.100740

JHEP Rep. 2025 Feb 19;7(5):101356. doi: 10.1016/j.jhepr.2025.101356. eCollection 2025 May.

ABSTRACT

BACKGROUND & AIMS: Accurate prediction of disease severity and prognosis are challenging in patients with cirrhosis. We evaluated whether the deep learning-based AI-Cirrhosis-ECG (ACE) score could detect hepatic decompensation and predict clinical outcomes in cirrhosis.

METHODS: We analyzed 2,166 ECGs from 472 patients in a retrospective Mayo Clinic cohort, 420 patients in a prospective Mayo transplant cohort, and 341 patients in an external validation cohort from Hospital Clínic de Barcelona. The ACE score's performance was assessed using receiver-operating characteristic analysis for decompensation detection and competing risks Cox regression for outcome prediction.

RESULTS: The ACE score showed high accuracy in detecting hepatic decompensation (area under the curve 0.933, 95% CI: 0.923-0.942) with 88.0% sensitivity and 84.3% specificity at an optimal threshold of 0.25. In multivariable analysis, each 0.1-point increase in ACE score was independently associated with increased risk of liver-related death (hazard ratio [HR] 1.44, 95% CI 1.32-1.58, p <0.001). Adding ACE to model for end-stage liver disease-sodium significantly improved prediction of adverse outcomes across all cohorts (c-statistics: retrospective cohort 0.903 vs. 0.844; prospective cohort 0.779 vs. 0.735; external validation 0.744 vs. 0.732; all p <0.001).

CONCLUSIONS: The ACE score accurately identifies hepatic decompensation and independently predicts liver-related outcomes in cirrhosis. This non-invasive tool enhances current prognostic models and may improve risk stratification in cirrhosis management.

IMPACT AND IMPLICATIONS: This study demonstrates the potential of artificial intelligence to enhance prognostication in liver disease, addressing the critical need for improved risk stratification in cirrhosis management. The AI-Cirrhosis-ECG (ACE) score, derived from widely available ECGs, shows promise as a non-invasive tool for detecting hepatic decompensation and predicting liver-related outcomes, which could significantly impact clinical decision-making and resource allocation in hepatology. These findings are particularly important for hepatologists, transplant surgeons, and patients with cirrhosis, as they offer a novel approach to complement existing prognostic models such as model for end-stage liver disease-sodium. In practical terms, the ACE score could be integrated into routine clinical assessments to provide more accurate risk predictions, potentially improving the timing of interventions, optimizing transplant listing decisions, and ultimately enhancing patient outcomes. However, further validation in diverse populations and integration with other established predictors is necessary before widespread clinical implementation.

PMID:40276480 | PMC:PMC12018547 | DOI:10.1016/j.jhepr.2025.101356

Front Public Health. 2025 Apr 10;13:1540618. doi: 10.3389/fpubh.2025.1540618. eCollection 2025.

ABSTRACT

INTRODUCTION: Preventive healthcare policies are critical for improving public health outcomes and reducing the socioeconomic burden of diseases, aligning closely with the theme of enhancing residents' health welfare through robust social security systems. However, traditional approaches often overlook the dynamic interplay between economic factors and health outcomes, limiting their effectiveness in designing sustainable interventions.

METHODS: To address these gaps, this study leverages corporate financial monitoring as a novel lens for assessing the effectiveness of preventive healthcare policies. Utilizing the Advanced Financial Monitoring Neural Framework (AFMNF) and the Dynamic Risk-Adaptive Framework (DRAF), we integrate deep learning techniques with dynamic risk modeling to analyze the financial and health impacts of such policies. Our methodology involves monitoring corporate financial metrics, anomaly detection, and trend analysis to identify correlations between policy implementation and economic indicators.

RESULTS AND DISCUSSION: The results demonstrate that integrating financial insights with health policy evaluation improves prediction accuracy of socioeconomic outcomes by 40% and enhances anomaly detection in policy performance by 30%. This adaptive framework offers a scalable, real-time approach to monitoring, providing actionable insights for policymakers to optimize preventive healthcare strategies. This study underscores the importance of interdisciplinary methods in advancing public health outcomes through innovative, data-driven frameworks.

PMID:40276349 | PMC:PMC12019879 | DOI:10.3389/fpubh.2025.1540618