Endocr Pathol. 2025 Jan 23;36(1):2. doi: 10.1007/s12022-025-09846-3.

ABSTRACT

Pancreatic neuroendocrine tumors (PanNETs) are a heterogeneous group of neoplasms that include tumors with different histomorphologic characteristics that can be correlated to sub-categories with different prognoses. In addition to the WHO grading scheme based on tumor proliferative activity, a new parameter based on the scoring of infiltration patterns at the interface of tumor and non-neoplastic parenchyma (tumor-NNP interface) has recently been proposed for PanNET categorization. Despite the known correlations, these categorizations can still be problematic due to the need for human judgment, which may involve intra- and inter-observer variability. Although there is a great need for automated systems working on quantitative metrics to reduce observer variability, there are no such systems for PanNET categorization. Addressing this gap, this study presents a computational pipeline that uses deep learning models to automatically categorize PanNETs for the first time. This pipeline proposes to quantitatively characterize PanNETs by constructing entity-graphs on the cells, and to learn the PanNET categorization using a graph neural network (GNN) trained on these graphs. Different than the previous studies, the proposed model integrates pathology domain knowledge into the GNN construction and training for the purpose of a deeper utilization of the tumor microenvironment and its architectural changes for PanNET categorization. We tested our model on 105 HE stained whole slide images of PanNET tissues. The experiments revealed that this domain knowledge integrated pipeline led to a 76.70% test set F1-score, resulting in significant improvements over its counterparts.

PMID:39847242 | DOI:10.1007/s12022-025-09846-3

Med Biol Eng Comput. 2025 Jan 23. doi: 10.1007/s11517-025-03296-z. Online ahead of print.

ABSTRACT

Positron emission tomography (PET) imaging plays a pivotal role in oncology for the early detection of metastatic tumors and response to therapy assessment due to its high sensitivity compared to anatomical imaging modalities. The balance between image quality and radiation exposure is critical, as reducing the administered dose results in a lower signal-to-noise ratio (SNR) and information loss, which may significantly affect clinical diagnosis. Deep learning (DL) algorithms have recently made significant progress in low-dose (LD) PET reconstruction. Nevertheless, a successful clinical application requires a thorough evaluation of uncertainty to ensure informed clinical judgment. We propose NPB-LDPET, a DL-based non-parametric Bayesian framework for LD PET reconstruction and uncertainty assessment. Our framework utilizes an Adam optimizer with stochastic gradient Langevin dynamics (SGLD) to sample from the underlying posterior distribution. We employed the Ultra-low-dose PET Challenge dataset to assess our framework's performance relative to the Monte Carlo dropout benchmark. We evaluated global reconstruction accuracy utilizing SSIM, PSNR, and NRMSE, local lesion conspicuity using mean absolute error (MAE) and local contrast, and the clinical relevance of uncertainty maps employing correlation between the uncertainty measures and the dose reduction factor (DRF). Our NPB-LDPET reconstruction method exhibits a significantly superior global reconstruction accuracy for various DRFs (paired t-test, p < 0.0001 , N=10, 631). Moreover, we demonstrate a 21% reduction in MAE (573.54 vs. 723.70, paired t-test, p < 0.0001 , N=28) and an 8.3% improvement in local lesion contrast (2.077 vs. 1.916, paired t-test, p < 0.0001 , N=28). Furthermore, our framework exhibits a stronger correlation between the predicted uncertainty 95th percentile score and the DRF ( r 2 = 0.9174 vs. r 2 = 0.6144 , N=10, 631). The proposed framework has the potential to improve clinical decision-making for LD PET imaging by providing a more accurate and informative reconstruction while reducing radiation exposure.

PMID:39847156 | DOI:10.1007/s11517-025-03296-z

Med Biol Eng Comput. 2025 Jan 23. doi: 10.1007/s11517-025-03301-5. Online ahead of print.

ABSTRACT

The generalization of deep learning (DL) models is critical for accurate lesion segmentation in breast ultrasound (BUS) images. Traditional DL models often struggle to generalize well due to the high frequency and scale variations inherent in BUS images. Moreover, conventional loss functions used in these models frequently result in imbalanced optimization, either prioritizing region overlap or boundary accuracy, which leads to suboptimal segmentation performance. To address these issues, we propose UMA-Net, an enhanced UNet architecture specifically designed for BUS image segmentation. UMA-Net integrates residual connections, attention mechanisms, and a bottleneck with atrous convolutions to effectively capture multi-scale contextual information without compromising spatial resolution. Additionally, we introduce an adaptive ensemble loss function that dynamically balances the contributions of different loss components during training, ensuring optimization across key segmentation metrics. This novel approach mitigates the imbalances found in conventional loss functions. We validate UMA-Net on five diverse BUS datasets-BUET, BUSI, Mendeley, OMI, and UDIAT-demonstrating superior performance. Our findings highlight the importance of addressing frequency and scale variations, confirming UMA-Net as a robust and generalizable solution for BUS image segmentation.

PMID:39847155 | DOI:10.1007/s11517-025-03301-5

J Microsc. 2025 Jan 23. doi: 10.1111/jmi.13385. Online ahead of print.

ABSTRACT

The types and quantities of microorganisms in activated sludge are directly related to the stability and efficiency of sewage treatment systems. This paper proposes a sludge microorganism detection method based on microscopic phase contrast image optimisation and deep learning. Firstly, a dataset containing eight types of microorganisms is constructed, and an augmentation strategy based on single and multisamples processing is designed to address the issues of sample deficiency and uneven distribution. Secondly, a phase contrast image quality optimisation algorithm based on fused variance is proposed, which can effectively improve the standard deviation, entropy, and detection performance. Thirdly, a lightweight YOLOv8n-SimAM model is designed, which introduces a SimAM attention module to suppress the complex background interference and enhance attentions to the target objects. The lightweight of the network is realised using a detection head based on multiscale information fusion convolutional module. In addition, a new loss function IW-IoU is proposed to improve the generalisation ability and overall performance. Comparative and ablative experiments are conducted, demonstrating the great application potential for rapid and accurate detection of microbial targets. Compared to the baseline model, the proposed method improves the detection accuracy by 12.35% and hastens the running speed by 37.9 frames per second while evidently reducing the model size.

PMID:39846854 | DOI:10.1111/jmi.13385

J Funct Morphol Kinesiol. 2025 Jan 1;10(1):14. doi: 10.3390/jfmk10010014.

ABSTRACT

Background/Objectives: Chronic low back pain (CLBP) is prevalent among older adults and leads to significant functional limitations and reduced quality of life. Segmental stabilization exercises (SSEs) are commonly used to treat CLBP, but the selective activation of deep abdominal muscles during these exercises can be challenging for patients. To support muscle activation, physiotherapists use biofeedback methods such as palpation and ultrasound imaging. This randomized controlled pilot study aimed to compare the effectiveness of these two biofeedback techniques in older adults with CLBP. Methods: A total of 24 participants aged 65 years or older with CLBP were randomly assigned to one of two groups: one group performed self-palpation biofeedback, while the other group used real-time ultrasound imaging to visualize abdominal muscle activation. Muscle activation and thickness were continuously tracked using a semi-automated algorithm. The preferential activation ratio (PAR) was calculated to measure muscle activation, and statistical comparisons between groups were made using ANOVA. Results: Both groups achieved positive PAR values during all repetitions of the abdominal-draw-in maneuver (ADIM) and abdominal bracing (AB). Statistical analysis revealed no significant differences between the groups in terms of PAR during ADIM (F(2, 42) = 0.548, p = 0.58, partial η2 = 0.025) or AB (F(2, 36) = 0.812, p = 0.45, partial η2 = 0.043). Both groups reported high levels of exercise enjoyment and low task load. Conclusions: In conclusion, both palpation and ultrasound biofeedback appear to be effective for guiding older adults with CLBP during SSE. Larger studies are needed to confirm these results and examine the long-term effectiveness of these biofeedback methods.

PMID:39846655 | DOI:10.3390/jfmk10010014

Animal Model Exp Med. 2025 Jan 23. doi: 10.1002/ame2.12530. Online ahead of print.

ABSTRACT

BACKGROUND: Quantifying the rich home-cage activities of tree shrews provides a reliable basis for understanding their daily routines and building disease models. However, due to the lack of effective behavioral methods, most efforts on tree shrew behavior are limited to simple measures, resulting in the loss of much behavioral information.

METHODS: To address this issue, we present a deep learning (DL) approach to achieve markerless pose estimation and recognize multiple spontaneous behaviors of tree shrews, including drinking, eating, resting, and staying in the dark house, etc. RESULTS: This high-throughput approach can monitor the home-cage activities of 16 tree shrews simultaneously over an extended period. Additionally, we demonstrated an innovative system with reliable apparatus, paradigms, and analysis methods for investigating food grasping behavior. The median duration for each bout of grasping was 0.20 s.

CONCLUSION: This study provides an efficient tool for quantifying and understand tree shrews' natural behaviors.

PMID:39846430 | DOI:10.1002/ame2.12530

IET Syst Biol. 2025 Jan-Dec;19(1):e70000. doi: 10.1049/syb2.70000.

ABSTRACT

Spatially resolved transcriptomics technologies potentially provide the extra spatial position information and tissue image to better infer spatial cell-cell interactions (CCIs) in processes such as tissue homeostasis, development, and disease progression. However, methods for effectively integrating spatial multimodal data to infer CCIs are still lacking. Here, the authors propose a deep learning method for integrating features through co-convolution, called SpaGraphCCI, to effectively integrate data from different modalities of SRT by projecting gene expression and image feature into a low-dimensional space. SpaGraphCCI can achieve significant performance on datasets from multiple platforms including single-cell resolution datasets (AUC reaches 0.860-0.907) and spot resolution datasets (AUC ranges from 0.880 to 0.965). SpaGraphCCI shows better performance by comparing with the existing deep learning-based spatial cell communication inference methods. SpaGraphCCI is robust to high noise and can effectively improve the inference of CCIs. We test on a human breast cancer dataset and show that SpaGraphCCI can not only identify proximal cell communication but also infer new distal interactions. In summary, SpaGraphCCI provides a practical tool that enables researchers to decipher spatially resolved cell-cell communication based on spatial transcriptome data.

PMID:39846423 | DOI:10.1049/syb2.70000

Acta Radiol. 2025 Jan 23:2841851241305736. doi: 10.1177/02841851241305736. Online ahead of print.

ABSTRACT

BACKGROUND: The wide variability in thresholds on computed tomography (CT) perfusion parametric maps has led to controversy in the stroke imaging community about the most accurate measurement of core infarction.

PURPOSE: To investigate the feasibility of using U-Net to perform infarct core segmentation in CT perfusion imaging.

MATERIAL AND METHODS: CT perfusion parametric maps were the input of U-Net, while the ground truth segmentation was determined based on diffusion-weighted imaging (DWI). The dataset used in this study was from the ISLES2018 challenge, which contains 63 acute stroke patients receiving CT perfusion imaging and DWI within 8 h of stroke onset. The segmentation accuracy of model outputs was assessed by calculating Dice similarity coefficient (DSC), sensitivity, and intersection over union (IoU).

RESULTS: The highest DSC was observed in U-Net taking mean transit time (MTT) or time-to-maximum (Tmax) as input. Meanwhile, the highest sensitivity and IoU were observed in U-Net taking Tmax as input. A DSC in the range of 0.2-0.4 was found in U-Net taking Tmax as input when the infarct area contains < 1000 pixels. A DSC of 0.4-0.6 was found in U-Net taking Tmax as input when the infarct area contains 1000-1999 pixels. A DSC value of 0.6-0.8 was found in U-Net taking Tmax as input when the infarct area contains ≥ 2000 pixels.

CONCLUSION: Our model achieved good performance for infarct area containing ≥ 2000 pixels, so it may assist in identifying patients who are contraindicated for intravenous thrombolysis.

PMID:39846186 | DOI:10.1177/02841851241305736

Clin Implant Dent Relat Res. 2025 Feb;27(1):e70000. doi: 10.1111/cid.70000.

ABSTRACT

OBJECTIVES: This study aimed to develop an artificial intelligence (AI)-based deep learning model for the detection and numbering of dental implants in panoramic radiographs. The novelty of this model lies in its ability to both detect and number implants, offering improvements in clinical decision support for dental implantology.

MATERIALS AND METHODS: A retrospective dataset of 32 585 panoramic radiographs, collected from patients at Sivas Cumhuriyet University between 2014 and 2024, was utilized. Two deep-learning models were trained using the YOLOv8 algorithm. The first model classified the regions of the jaw to number the teeth and identify implant regions, while the second model performed implant segmentation. Performance metrics including precision, recall, and F1-score were used to evaluate the model's effectiveness.

RESULTS: The implant segmentation model achieved a precision of 91.4%, recall of 90.5%, and an F1-score of 93.1%. For the implant-numbering task, precision ranged from 0.94 to 0.981, recall from 0.895 to 0.956, and F1-scores from 0.917 to 0.966 across various jaw regions. The analysis revealed that implants were most frequently located in the maxillary posterior region.

CONCLUSIONS: The AI model demonstrated high accuracy in detecting and numbering dental implants in panoramic radiographs. This technology offers the potential to reduce clinicians' workload and improve diagnostic accuracy in dental implantology. Further validation across more diverse datasets is recommended to enhance its clinical applicability.

CLINICAL RELEVANCE: This AI model could revolutionize dental implant detection and classification, providing fast, objective analyses to support clinical decision-making in dental practices.

PMID:39846131 | DOI:10.1111/cid.70000

Bioinform Adv. 2024 Nov 25;5(1):vbae187. doi: 10.1093/bioadv/vbae187. eCollection 2025.

ABSTRACT

SUMMARY: Protein structure prediction aims to infer a protein's three-dimensional (3D) structure from its amino acid sequence. Protein structure is pivotal for elucidating protein functions, interactions, and driving biotechnological innovation. The deep learning model AlphaFold2, has revolutionized this field by leveraging phylogenetic information from multiple sequence alignments (MSAs) to achieve remarkable accuracy in protein structure prediction. However, a key question remains: how well does AlphaFold2 understand protein structures? This study investigates AlphaFold2's capabilities when relying primarily on high-quality template structures, without the additional information provided by MSAs. By designing experiments that probe local and global structural understanding, we aimed to dissect its dependence on specific features and its ability to handle missing information. Our findings revealed AlphaFold2's reliance on sterically valid C β for correctly interpreting structural templates. Additionally, we observed its remarkable ability to recover 3D structures from certain perturbations and the negligible impact of the previous structure in recycling. Collectively, these results support the hypothesis that AlphaFold2 has learned an accurate biophysical energy function. However, this function seems most effective for local interactions. Our work advances understanding of how deep learning models predict protein structures and provides guidance for researchers aiming to overcome limitations in these models.

AVAILABILITY AND IMPLEMENTATION: Data and implementation are available at https://github.com/ibmm-unibe-ch/template-analysis.

PMID:39846081 | PMC:PMC11751578 | DOI:10.1093/bioadv/vbae187

Eur Heart J Digit Health. 2024 Sep 24;6(1):137-146. doi: 10.1093/ehjdh/ztae072. eCollection 2025 Jan.

ABSTRACT

AIMS: Accurate heart function estimation is vital for detecting and monitoring cardiovascular diseases. While two-dimensional echocardiography (2DE) is widely accessible and used, it requires specialized training, is prone to inter-observer variability, and lacks comprehensive three-dimensional (3D) information. We introduce CardiacField, a computational echocardiography system using a 2DE probe for precise, automated left ventricular (LV) and right ventricular (RV) ejection fraction (EF) estimations, which is especially easy to use for non-cardiovascular healthcare practitioners. We assess the system's usability among novice users and evaluate its performance against expert interpretations and advanced deep learning (DL) tools.

METHODS AND RESULTS: We developed an implicit neural representation network to reconstruct a 3D cardiac volume from sequential multi-view 2DE images, followed by automatic segmentation of LV and RV areas to calculate volume sizes and EF values. Our study involved 127 patients to assess EF estimation accuracy against expert readings and two-dimensional (2D) video-based DL models. A subset of 56 patients was utilized to evaluate image quality and 3D accuracy and another 50 to test usability by novice users and across various ultrasound machines. CardiacField generated a 3D heart from 2D echocardiograms with <2 min processing time. The LVEF predicted by our method had a mean absolute error (MAE) of 2.48 % , while the RVEF had an MAE of 2.65 % .

CONCLUSION: Employing a straightforward apical ring scan with a cost-effective 2DE probe, our method achieves a level of EF accuracy for assessing LV and RV function that is comparable to that of three-dimensional echocardiography probes.

PMID:39846074 | PMC:PMC11750196 | DOI:10.1093/ehjdh/ztae072

Eur Heart J Digit Health. 2024 Oct 27;6(1):7-22. doi: 10.1093/ehjdh/ztae080. eCollection 2025 Jan.

ABSTRACT

Cardiovascular disease (CVD) remains a major cause of mortality in the UK, prompting the need for improved risk predictive models for primary prevention. Machine learning (ML) models utilizing electronic health records (EHRs) offer potential enhancements over traditional risk scores like QRISK3 and ASCVD. To systematically evaluate and compare the efficacy of ML models against conventional CVD risk prediction algorithms using EHR data for medium to long-term (5-10 years) CVD risk prediction. A systematic review and random-effect meta-analysis were conducted according to preferred reporting items for systematic reviews and meta-analyses guidelines, assessing studies from 2010 to 2024. We retrieved 32 ML models and 26 conventional statistical models from 20 selected studies, focusing on performance metrics such as area under the curve (AUC) and heterogeneity across models. ML models, particularly random forest and deep learning, demonstrated superior performance, with the highest recorded pooled AUCs of 0.865 (95% CI: 0.812-0.917) and 0.847 (95% CI: 0.766-0.927), respectively. These significantly outperformed the conventional risk score of 0.765 (95% CI: 0.734-0.796). However, significant heterogeneity (I² > 99%) and potential publication bias were noted across the studies. While ML models show enhanced calibration for CVD risk, substantial variability and methodological concerns limit their current clinical applicability. Future research should address these issues by enhancing methodological transparency and standardization to improve the reliability and utility of these models in clinical settings. This study highlights the advanced capabilities of ML models in CVD risk prediction and emphasizes the need for rigorous validation to facilitate their integration into clinical practice.

PMID:39846062 | PMC:PMC11750195 | DOI:10.1093/ehjdh/ztae080

Health Inf Sci Syst. 2025 Jan 17;13(1):17. doi: 10.1007/s13755-024-00333-3. eCollection 2025 Dec.

ABSTRACT

PURPOSE: This paper aims to develop a three-dimensional (3D) Alzheimer's disease (AD) prediction method, thereby bettering current predictive methods, which struggle to fully harness the potential of structural magnetic resonance imaging (sMRI) data.

METHODS: Traditional convolutional neural networks encounter pressing difficulties in accurately focusing on the AD lesion structure. To address this issue, a 3D decoupling, self-attention network for AD prediction is proposed. Firstly, a multi-scale decoupling block is designed to enhance the network's ability to extract fine-grained features by segregating convolutional channels. Subsequently, a self-attention block is constructed to extract and adaptively fuse features from three directions (sagittal, coronal and axial), so that more attention is geared towards brain lesion areas. Finally, a clustering loss function is introduced and combined with the cross-entropy loss to form a joint loss function for enhancing the network's ability to discriminate between different sample types.

RESULTS: The accuracy of our model is 0.985 for the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset and 0.963 for the Australian Imaging, Biomarker & Lifestyle (AIBL) dataset, both of which are higher than the classification accuracy of similar tasks in this category. This demonstrates that our model can accurately distinguish between normal control (NC) and Alzheimer's Disease (AD), as well as between stable mild cognitive impairment (sMCI) and progressive mild cognitive impairment (pMCI).

CONCLUSION: The proposed AD prediction network exhibits competitive performance when compared with state-of-the-art methods. The proposed model successfully addresses the challenges of dealing with 3D sMRI image data and the limitations stemming from inadequate information in 2D sections, advancing the utility of predictive methods for AD diagnosis and treatment.

PMID:39846055 | PMC:PMC11748674 | DOI:10.1007/s13755-024-00333-3

Health Sci Rep. 2025 Jan 22;8(1):e70372. doi: 10.1002/hsr2.70372. eCollection 2025 Jan.

ABSTRACT

BACKGROUND AND AIM: Epilepsy is a major neurological challenge, especially for pediatric populations. It profoundly impacts both developmental progress and quality of life in affected children. With the advent of artificial intelligence (AI), there's a growing interest in leveraging its capabilities to improve the diagnosis and management of pediatric epilepsy. This review aims to assess the effectiveness of AI in pediatric epilepsy detection while considering the ethical implications surrounding its implementation.

METHODOLOGY: A comprehensive systematic review was conducted across multiple databases including PubMed, EMBASE, Google Scholar, Scopus, and Medline. Search terms encompassed "pediatric epilepsy," "artificial intelligence," "machine learning," "ethical considerations," and "data security." Publications from the past decade were scrutinized for methodological rigor, with a focus on studies evaluating AI's efficacy in pediatric epilepsy detection and management.

RESULTS: AI systems have demonstrated strong potential in diagnosing and monitoring pediatric epilepsy, often matching clinical accuracy. For example, AI-driven decision support achieved 93.4% accuracy in diagnosis, closely aligning with expert assessments. Specific methods, like EEG-based AI for detecting interictal discharges, showed high specificity (93.33%-96.67%) and sensitivity (76.67%-93.33%), while neuroimaging approaches using rs-fMRI and DTI reached up to 97.5% accuracy in identifying microstructural abnormalities. Deep learning models, such as CNN-LSTM, have also enhanced seizure detection from video by capturing subtle movement and expression cues. Non-EEG sensor-based methods effectively identified nocturnal seizures, offering promising support for pediatric care. However, ethical considerations around privacy, data security, and model bias remain crucial for responsible AI integration.

CONCLUSION: While AI holds immense potential to enhance pediatric epilepsy management, ethical considerations surrounding transparency, fairness, and data security must be rigorously addressed. Collaborative efforts among stakeholders are imperative to navigate these ethical challenges effectively, ensuring responsible AI integration and optimizing patient outcomes in pediatric epilepsy care.

PMID:39846037 | PMC:PMC11751886 | DOI:10.1002/hsr2.70372

MethodsX. 2024 Dec 21;14:103131. doi: 10.1016/j.mex.2024.103131. eCollection 2025 Jun.

ABSTRACT

Recent advancements in artificial intelligence (AI) have increased interest in intelligent transportation systems, particularly autonomous vehicles. Safe navigation in traffic-heavy environments requires accurate road scene segmentation, yet traditional computer vision methods struggle with complex scenarios. This study emphasizes the role of deep learning in improving semantic segmentation using datasets like the Indian Driving Dataset (IDD), which presents unique challenges in chaotic road conditions. We propose a modified CANet that incorporates U-Net and LinkNet elements, focusing on accuracy, efficiency, and resilience. The CANet features an encoder-decoder architecture and a Multiscale Context Module (MCM) with three parallel branches to capture contextual information at multiple scales. Our experiments show that the proposed model achieves a mean Intersection over Union (mIoU) value of 0.7053, surpassing state-of-the-art models in efficiency and performance. Here we demonstrate:•Traditional computer vision methods struggle with complex driving scenarios, but deep learning based semantic segmentation methods show promising results.•Modified CANet, incorporating U-Net and LinkNet elements is proposed for semantic segmentation of unstructured driving scenarios.•The CANet structure consists of an encoder-decoder architecture and a Multiscale Context Module (MCM) with three parallel branches to capture contextual information at multiple scales.

PMID:39846010 | PMC:PMC11751566 | DOI:10.1016/j.mex.2024.103131

Eur J Obstet Gynecol Reprod Biol X. 2024 Dec 24;25:100362. doi: 10.1016/j.eurox.2024.100362. eCollection 2025 Mar.

ABSTRACT

This review examines the emerging applications of machine learning (ML) and radiomics in the diagnosis and prediction of placenta accreta spectrum (PAS) disorders, addressing a significant challenge in obstetric care. It highlights recent advancements in ML algorithms and radiomic techniques that utilize medical imaging modalities like magnetic resonance imaging (MRI) and ultrasound for effective classification and risk stratification of PAS. The review discusses the efficacy of various deep learning models, such as nnU-Net and DenseNet-PAS, which have demonstrated superior performance over traditional diagnostic methods through high AUC scores. Furthermore, it underscores the importance of integrating quantitative imaging features with clinical data to enhance diagnostic accuracy and optimize surgical planning. The potential of ML to predict surgical morbidity by analyzing demographic and obstetric factors is also explored. Emphasizing the need for standardized methodologies to ensure consistent feature extraction and model performance, this review advocates for the integration of radiomics and ML into clinical workflows, aiming to improve patient outcomes and foster a multidisciplinary approach in high-risk pregnancies. Future research should focus on larger datasets and validation of biomarkers to refine predictive models in obstetric care.

PMID:39845985 | PMC:PMC11751428 | DOI:10.1016/j.eurox.2024.100362

Genome Integr. 2025 Jan 22;15:e20240002. doi: 10.14293/genint.15.1.002. eCollection 2024.

ABSTRACT

Artificial intelligence (AI) offers a broad range of enhancements in medicine. Machine learning and deep learning techniques have shown significant potential in improving diagnosis and treatment outcomes, from assisting clinicians in diagnosing medical images to ascertaining effective drugs for a specific disease. Despite the prospective benefits, adopting AI in clinical settings requires careful consideration, particularly concerning data generalisation and model explainability. This commentary aims to discuss two potential use cases for AI in the field of medicine and the overarching challenges involved in their implementation.

PMID:39845982 | PMC:PMC11752870 | DOI:10.14293/genint.15.1.002

Health Aff (Millwood). 2025 Jan 22:101377hlthaff202401003. doi: 10.1377/hlthaff.2024.01003. Online ahead of print.

ABSTRACT

The field of artificial intelligence (AI) has entered a new cycle of intense opportunity, fueled by advances in deep learning, including generative AI. Applications of recent advances affect many aspects of everyday life, yet nowhere is it more important to use this technology safely, effectively, and equitably than in health and health care. Here, as part of the National Academy of Medicine's Vital Directions for Health and Health Care: Priorities for 2025 initiative, which is designed to provide guidance on pressing health care issues for the incoming presidential administration, we describe the steps needed to achieve these goals. We focus on four strategic areas: ensuring safe, effective, and trustworthy use of AI; promotion and development of an AI-competent health care workforce; investing in AI research to support the science, practice, and delivery of health and health care; and promotion of policies and procedures to clarify AI liability and responsibilities.

PMID:39841940 | DOI:10.1377/hlthaff.2024.01003

Retina. 2025 Jan 17. doi: 10.1097/IAE.0000000000004409. Online ahead of print.

ABSTRACT

PURPOSE: Predicting long-term anatomical responses in neovascular age-related macular degeneration (nAMD) patients is critical for patient-specific management. This study validates a generative deep learning (DL) model to predict 12-month posttreatment optical coherence tomography (OCT) images and evaluates the impact of incorporating clinical data on predictive performance.

METHODS: A total of 533 eyes from 513 treatment-naïve nAMD patients were analyzed. A conditional generative adversarial network (cGAN) served as the baseline model, generating 12-month OCT images using pretreatment OCT, fluorescein angiography (FA), and indocyanine green angiography (ICGA). We then sequentially added OCT after three loading doses, baseline visual acuity (VA), treatment regimen (pro re nata or treat-and-extend), drug type, and switching events. The generated and patient OCT images were compared for intraretinal fluid, subretinal fluid, pigment epithelial detachment, and subretinal hyperreflective material, both qualitatively and quantitatively.

RESULTS: The baseline model achieved acceptable accuracy for four macular fluid compartments (range 0.74-0.96). Incorporating OCT after loading doses and other clinical parameters improved accuracy (range 0.91-0.98). With all the clinical inputs, the model achieved 92% accuracy in distinguishing wet macular status from dry macular status. The segmented fluid compartments in the generated images correlated positively with those in the patient images.

CONCLUSION: Integrating clinical and treatment data, particularly OCT data after loading doses, significantly enhanced the 12-month predictive performance of cGANs. This approach can help clinicians anticipate anatomical outcomes and guide personalized, long-term nAMD treatment strategies.

PMID:39841905 | DOI:10.1097/IAE.0000000000004409

J Chem Theory Comput. 2025 Jan 22. doi: 10.1021/acs.jctc.4c01525. Online ahead of print.

ABSTRACT

Metal-organic frameworks (MOFs) hold great potential in gas separation and storage. Graph neural networks (GNNs) have proven effective in exploring structure-property relationships and discovering new MOF structures. Unlike molecular graphs, crystal graphs must consider the periodicity and patterns. MOFs' specific features at different scales, such as covalent bonds, functional groups, and global structures, influenced by interatomic interactions, exert varying degrees of impact on gas adsorption or selectivity. Moreover, redundant interatomic interactions hinder training accuracy, leading to overfitting. This research introduces a construction method for multiscale crystal graphs, which considers specific features at different scales by decomposing the crystal graph into multiple subgraphs based on interatomic interactions within varying distance ranges. Additionally, it takes into account the global structure of the crystal by encoding the periodic patterns of the unit cells. We propose MSAIGNN, a multiscale atomic interaction graph neural network with self-attention-based graph pooling mechanism, which incorporates three-body bond angle information, accounts for structural features at different scales, and minimizes interference from redundant interactions. Compared with traditional methods, MSAIGNN demonstrates higher prediction accuracy in assessing single-component adsorption, gas separation, and structural features. Visualization of attention scores confirms effective learning of structural features at different scales, highlighting MSAIGNN's interpretability. Overall, MSAIGNN offers a novel, efficient, multilayered, and interpretable approach for property prediction of complex porous crystal structures like MOFs using deep learning.

PMID:39841881 | DOI:10.1021/acs.jctc.4c01525