Front Oncol. 2025 Jul 10;15:1582322. doi: 10.3389/fonc.2025.1582322. eCollection 2025.

ABSTRACT

OBJECTIVES: Non-contrast MRI(NC-MRI) is an attractive option for liver tumors screening and follow-up. This study aims to develop and validate a deep convolutional neural network for the classification of liver lesions using non-contrast MRI.

METHODS: A total of 50418 enhanced MRI images from 1959 liver tumor patients across three centers were included. Inception-ResNet V2 was used to generate four models through transfer-learning for three-way lesion classification, which processed T2-weighted, diffusion-weighted (DWI) and multiphasic T1-weighted images. The models were then validated using one independent internal and two external datasets with 5172, 2916, and 1338 images, respectively. The efficacy of non-contrast models (T2,T2+DWI) in differentiating between benign and malignant liver lesions at the patient level was also evaluated and compared with radiologists. The performance of models was evaluated using the area under the receiver operating characteristic curve (AUC),sensitivity and specificity.

RESULTS: Similar to multi-sequence and enhanced image-based models, the non-contrast models showed comparable accuracy in classifying liver lesions as benign, primary malignant or metastatic. In the independent internal cohort, the T2+DWI model achieved AUC of 0.91(95% CI,0.888-0.932), 0.873(0.848-0.899), and 0.876(0.840-0.911) for three tumour categories, respectively. The sensitivities for distinguishing malignant tumors in three validation sets were 98.1%, 89.7%, and 87.5%%, with specificities over 70% in all three sets.

CONCLUSIONS: Our deep-learning-based model yielded good applicability in classifying liver lesions in non-contrast MRI. It provides a potential alternative for screening liver tumors with the advantage of reducing costs, scanning time and contrast-agents risks. It is more suitable for benign tumours follow-up, surveillance of HCC and liver metastasis that need periodic repetitive examinations.

PMID:40708941 | PMC:PMC12287020 | DOI:10.3389/fonc.2025.1582322

Front Oncol. 2025 Jul 10;15:1576461. doi: 10.3389/fonc.2025.1576461. eCollection 2025.

ABSTRACT

BACKGROUND: With the rapid advances in artificial intelligence-particularly convolutional neural networks-researchers now exploit CT, PET/CT and other imaging modalities to predict epidermal growth factor receptor (EGFR) mutation status in non-small-cell lung cancer (NSCLC) non-invasively, rapidly and repeatably. End-to-end deep-learning models simultaneously perform feature extraction and classification, capturing not only traditional radiomic signatures such as tumour density and texture but also peri-tumoural micro-environmental cues, thereby offering a higher theoretical performance ceiling than hand-crafted radiomics coupled with classical machine learning. Nevertheless, the need for large, well-annotated datasets, the domain shifts introduced by heterogeneous scanning protocols and preprocessing pipelines, and the "black-box" nature of neural networks all hinder clinical adoption. To address fragmented evidence and scarce external validation, we conducted a systematic review to appraise the true performance of deep-learning and radiomics models for EGFR prediction and to identify barriers to clinical translation, thereby establishing a baseline for forthcoming multicentre prospective studies.

METHODS: Following PRISMA 2020, we searched PubMed, Web of Science and IEEE Xplore for studies published between 2018 and 2024. Fifty-nine original articles met the inclusion criteria. QUADAS-2 was applied to the eight studies that developed models using real-world clinical data, and details of external validation strategies and performance metrics were extracted systematically.

RESULTS: The pooled internal area under the curve (AUC) was 0.78 for radiomics-machine-learning models and 0.84 for deep-learning models. Only 17 studies (29%) reported independent external validation, where the mean AUC fell to 0.77, indicating a marked domain-shift effect. QUADAS-2 showed that 31% of studies had high risk of bias in at least one domain, most frequently in Index Test and Patient Selection.

CONCLUSION: Although deep-learning models achieved the best internal performance, their reliance on single-centre data, the paucity of external validation and limited code availability preclude their use as stand-alone clinical decision tools. Future work should involve multicentre prospective designs, federated learning, decision-curve analysis and open sharing of models and data to verify generalisability and facilitate clinical integration.

PMID:40708937 | PMC:PMC12286997 | DOI:10.3389/fonc.2025.1576461

Bioinform Adv. 2025 Jun 27;5(1):vbaf153. doi: 10.1093/bioadv/vbaf153. eCollection 2025.

ABSTRACT

MOTIVATION: The latest generation of deep learning-based structure prediction methods enable accurate modelling of most proteins and many complexes. However, preparing inputs for the locally installed software is not always straightforward, and the results of local runs are not always presented in an ideally accessible fashion. Furthermore, it is not yet clear whether the latest tools perform equivalently for all types of target.

RESULTS: ABCFold facilitates the use of AlphaFold 3, Boltz-1, and Chai-1 with a standardized input to predict atomic structures, with Boltz-1 and Chai-1 being installed on runtime (if required). MSAs can be generated internally using either the JackHMMER MSA search within AlphaFold 3, or with the MMseqs2 API. Alternatively, users can provide their own custom MSAs. This therefore allows AlphaFold 3 to be installed and run without downloading the large databases needed for JackHMMER. There are also straightforward options to use templates, including custom templates. Results from all packages are treated in a unified fashion, enabling easy comparison of results from different methods. A variety of visualization options are available which include information on steric clashes.

AVAILABILITY AND IMPLEMENTATION: ABCFold is coded in Python and JavaScript. All scripts and associated documentation are available from https://github.com/rigdenlab/ABCFold or https://pypi.org/project/ABCFold/.

PMID:40708869 | PMC:PMC12287924 | DOI:10.1093/bioadv/vbaf153

Bioinform Adv. 2025 Jul 14;5(1):vbaf170. doi: 10.1093/bioadv/vbaf170. eCollection 2025.

ABSTRACT

MOTIVATION: N 6 - methyladenosine ( m 6 A ) is the most abundant internal modification in eukaryotic mRNA and plays essential roles in post-transcriptional gene regulation. While several deep learning approaches have been proposed to predict m 6 A sites, most suffer from limited chromosome-level generalizability due to evaluation on randomly split datasets.

RESULTS: In this study, we propose two novel hybrid deep learning models-Hybrid Model and Hybrid Deep Model-that integrate local sequence features (k-mers) and contextual embeddings via convolutional neural networks to improve predictive performance and generalization. We evaluate these models using both a Random-Split strategy and a more biologically realistic Leave-One-Chromosome-Out setting to ensure robustness across genomic regions. Our proposed models outperform the state-of-the-art m6A-TCPred model across all key evaluation metrics. Hybrid Deep Model achieves the highest accuracy under Random-Split, while Hybrid Model demonstrates superior generalization under Leave-One-Chromosome-Out, indicating that deep global representations may overfit in chromosome-independent settings. These findings underscore the importance of rigorous validation strategies and offer insights into designing robust m 6 A predictors.

AVAILABILITY AND IMPLEMENTATION: Source code and datasets are available at: https://github.com/malikmtahir/LOCO-m6A.

PMID:40708868 | PMC:PMC12288952 | DOI:10.1093/bioadv/vbaf170

Front Hum Neurosci. 2025 Jul 10;19:1637174. doi: 10.3389/fnhum.2025.1637174. eCollection 2025.

ABSTRACT

Inner speech recognition (ISR) is an emerging field with significant potential for applications in brain-computer interfaces (BCIs) and assistive technologies. This review focuses on the critical role of machine learning (ML) in decoding inner speech, exploring how various ML techniques improve the analysis and classification of neural signals. We analyze both traditional methods such as support vector machines (SVMs) and random forests, as well as advanced deep learning approaches like convolutional neural networks (CNNs), which are particularly effective at capturing the dynamic and non-linear patterns of inner speech-related brain activity. Also, the review covers the challenges of acquiring high-quality neural signals and discusses essential preprocessing methods for enhancing signal quality. Additionally, we outline and synthesize existing approaches for improving ISR through ML, that can lead to many potential implications in several domains, including assistive communication, brain-computer interfaces, and cognitive monitoring. The limitations of current technologies were also discussed, along with insights into future advancements and potential applications of machine learning in inner speech recognition (ISR). Building on prior literature, this work synthesizes and organizes existing ISR methodologies within a structured mathematical framework, reviews cognitive models of inner speech, and presents a detailed comparative analysis of existing ML approaches, thereby offering new insights into advancing the field.

PMID:40708808 | PMC:PMC12287026 | DOI:10.3389/fnhum.2025.1637174

J Cheminform. 2025 Jul 24;17(1):110. doi: 10.1186/s13321-025-01063-8.

ABSTRACT

The human Ether-à-go-go-Related Gene (hERG) potassium channel is crucial for repolarizing the cardiac action potential and regulating the heartbeat. Molecules that inhibit this protein can cause acquired long QT syndrome, increasing the risk of arrhythmias and sudden fatal cardiac arrests. Detecting compounds with potential hERG inhibitory activity is therefore essential to mitigate cardiotoxicity risks. In this article, we present a new hERG data set of unprecedented size, comprising nearly 300,000 molecules reported in PubChem and ChEMBL, approximately 2000 of which were confirmed hERG blockers identified through in vitro assays. Multiple structure-based artificial intelligence (AI) binary classifiers for predicting hERG inhibitors were developed, employing, as descriptors, protein-ligand extended connectivity (PLEC) fingerprints fed into random forest, extreme gradient boosting, and deep neural network (DNN) algorithms. Our best-performing model, a stacking ensemble classifier with a DNN meta-learner, achieved state-of-the-art classification performance, accurately identifying 86% of molecules having half-maximal inhibitory concentrations (IC50s) not exceeding 20 µM in our challenging test set, including 94% of hERG blockers whose IC50s were not greater than 1 µM. It also demonstrated superior screening power compared to virtual screening schemes that used existing scoring functions. This model, named "HERGAI," along with relevant input/output data and user-friendly source code, is available in our GitHub repository ( https://github.com/vktrannguyen/HERGAI ) and can be used to predict drug-induced hERG blockade, even on large data sets.

PMID:40708034 | DOI:10.1186/s13321-025-01063-8

Genome Biol. 2025 Jul 24;26(1):223. doi: 10.1186/s13059-025-03692-6.

ABSTRACT

Moving from a one-size-fits-all to an individual approach in precision medicine requires a deeper understanding of disease molecular mechanisms. Especially in heterogeneous complex diseases such as inflammatory bowel disease (IBD), better molecular stratification will help select the correct therapy. For this, we build end-to-end biologically sparsified neural network architectures for IBD subtyping based on whole exome sequence representations with gene-level and variant-level resolution. By moving beyond univariate methods, we capitalize on the model's ability to extract complex molecular patterns to improve prediction. Model interpretation identifies the most predictive pathways, genes, and variants, uncovering important intestinal barrier, immunological, and microbiome factors.

PMID:40708014 | DOI:10.1186/s13059-025-03692-6

Phys Eng Sci Med. 2025 Jul 24. doi: 10.1007/s13246-025-01583-5. Online ahead of print.

ABSTRACT

Medical image segmentation is a complex and challenging task, which aims to accurately segment various structures or abnormal regions in medical images. However, obtaining accurate segmentation results is difficult because of the great uncertainty in the shape, location, and scale of the target region. To address these challenges, we propose a higher-order spatial interaction framework with dual cross global efficient attention (DGEAHorNet), which employs a neural network architecture based on recursive gate convolution to adequately extract multi-scale contextual information from images. Specifically, a Dual Cross-Attentions (DCA) is added to the skip connection that can effectively blend multi-stage encoder features and narrow the semantic gap. In the bottleneck stage, global channel spatial attention module (GCSAM) is used to extract image global information. To obtain better feature representation, we feed the output from the GCSAM into the multi-branch dense layer (SENetV2) for excitation. Furthermore, we adopt Depthwise Over-parameterized Convolutional Layer (DO-Conv) in order to replace the common convolutional layer in the input and output part of our network, then add Efficient Attention (EA) to diminish computational complexity and enhance our model's performance. For evaluating the effectiveness of our proposed DGEAHorNet, we conduct comprehensive experiments on four publicly-available datasets, and achieving 0.9320, 0.9337, 0.9312 and 0.7799 in Dice similarity coefficient on ISIC2018, ISIC2017, CVC-ClinicDB and HRF respectively. Our results show that DGEAHorNet has better performance compared with advanced methods. The code is publicly available at https://github.com/penghaixin/mymodel .

PMID:40707863 | DOI:10.1007/s13246-025-01583-5

Nat Microbiol. 2025 Jul 24. doi: 10.1038/s41564-025-02034-3. Online ahead of print.

ABSTRACT

Filoviruses such as Ebola virus (EBOV) give rise to frequent epidemics with high case fatality rates while therapeutic options remain limited. Earlier genetic screens aimed to identify potential drug targets for EBOV relied on systems that may not fully recapitulate the virus life cycle. Here we applied an image-based genome-wide CRISPR screen to identify 998 host regulators of EBOV infection in 39,085,093 cells. A deep learning model associated each host factor with a distinct viral replication step. From this we confirmed UQCRB as a post-entry regulator of EBOV RNA replication and show that small-molecule UQCRB inhibition reduced virus infection in vitro. Using a random forest model, we found that perturbations on STRAP (a spliceosome-associated factor) disrupted the equilibrium between viral RNA and protein. STRAP was associated with VP35, a viral RNA processing protein. This genome-wide screen coupled with 12 secondary screens including validation experiments with Sudan and Marburg virus, presents a rich resource for host regulators of virus replication and potential targets for therapeutic intervention.

PMID:40707832 | DOI:10.1038/s41564-025-02034-3

Eur Spine J. 2025 Jul 25. doi: 10.1007/s00586-025-09179-z. Online ahead of print.

ABSTRACT

PURPOSE: Intervertebral disc degeneration (IDD) is a major contributor to chronic low back pain. Magnetic resonance imaging (MRI) serves as the gold standard for IDD assessment, yet manual grading is often subjective and inconsistent. With advances in artificial intelligence (AI), particularly deep learning, automated detection and classification of IDD from MRI has become increasingly feasible. This narrative review aims to provide a comprehensive overview of AI applications-especially machine learning and deep learning techniques-for MRI-based detection and grading of lumbar disc degeneration, highlighting their clinical value, current limitations, and future directions.

METHODS: Relevant studies were reviewed and summarized based on thematic structure. The review covers classical methods (e.g., support vector machines), deep learning models (e.g., CNNs, SpineNet, ResNet, U-Net), and hybrid approaches incorporating transformers and multitask learning. Technical details, model architectures, performance metrics, and representative datasets were synthesized and discussed.

RESULTS: AI systems have demonstrated promising performance in automatic IDD grading, in some cases matching or surpassing expert radiologists. CNN-based models showed high accuracy and reproducibility, while hybrid models further enhanced segmentation and classification tasks. However, challenges remain in generalizability, data imbalance, interpretability, and regulatory integration. Tools such as Grad-CAM and SHAP improve model transparency, while methods like few-shot learning and data augmentation can alleviate data limitations.

CONCLUSION: AI-assisted analysis of MRI for lumbar disc degeneration offers significant potential to enhance diagnostic efficiency and consistency. While current models are encouraging, real-world clinical implementation requires further advancements in interpretability, data diversity, ethical standards, and large-scale validation.

PMID:40707791 | DOI:10.1007/s00586-025-09179-z

Radiol Med. 2025 Jul 24. doi: 10.1007/s11547-025-02043-6. Online ahead of print.

ABSTRACT

PURPOSE: Contrast-enhanced magnetic resonance imaging (CE-MRI) monitoring across multiple time points is critical for optimizing hepatocellular carcinoma (HCC) prognosis during transarterial chemoembolization (TACE) treatment. The aim of this retrospective study is to develop and validate an artificial intelligence (AI)-powered models utilizing multi-time-point arterial phase CE-MRI data for HCC prognosis stratification in TACE patients.

MATERIAL AND METHODS: A total of 543 individual arterial phase CE-MRI scans from 181 HCC patients were retrospectively collected in this study. All patients underwent TACE and longitudinal arterial phase CE-MRI assessments at three time points: prior to treatment, and following the first and second TACE sessions. Among them, 110 patients received TACE monotherapy, while the remaining 71 patients underwent TACE in combination with microwave ablation (MWA). All images were subjected to standardized preprocessing procedures. We developed an end-to-end deep learning model, ProgSwin-UNETR, based on the Swin Transformer architecture, to perform four-class prognosis stratification directly from input imaging data. The model was trained using multi-time-point arterial phase CE-MRI data and evaluated via fourfold cross-validation. Classification performance was assessed using the area under the receiver operating characteristic curve (AUC). For comparative analysis, we benchmarked performance against traditional radiomics-based classifiers and the mRECIST criteria. Prognostic utility was further assessed using Kaplan-Meier (KM) survival curves. Additionally, multivariate Cox proportional hazards regression was performed as a post hoc analysis to evaluate the independent and complementary prognostic value of the model outputs and clinical variables. GradCAM + + was applied to visualize the imaging regions contributing most to model prediction.

RESULTS: The ProgSwin-UNETR model achieved an accuracy of 0.86 and an AUC of 0.92 (95% CI: 0.90-0.95) for the four-class prognosis stratification task, outperforming radiomic models across all risk groups. Furthermore, KM survival analyses were performed using three different approaches-AI model, radiomics-based classifiers, and mRECIST criteria-to stratify patients by risk. Of the three approaches, only the AI-based ProgSwin-UNETR model achieved statistically significant risk stratification across the entire cohort and in both TACE-alone and TACE + MWA subgroups (p < 0.005). In contrast, the mRECIST and radiomics models did not yield significant survival differences across subgroups (p > 0.05). Multivariate Cox regression analysis further demonstrated that the model was a robust independent prognostic factor (p = 0.01), effectively stratifying patients into four distinct risk groups (Class 0 to Class 3) with Log(HR) values of 0.97, 0.51, -0.53, and -0.92, respectively. Additionally, GradCAM + + visualizations highlighted critical regional features contributing to prognosis prediction, providing interpretability of the model.

CONCLUSION: ProgSwin-UNETR can well predict the various risk groups of HCC patients undergoing TACE therapy and can further be applied for personalized prediction.

PMID:40707767 | DOI:10.1007/s11547-025-02043-6

NPJ Digit Med. 2025 Jul 24;8(1):477. doi: 10.1038/s41746-025-01884-9.

ABSTRACT

Glaucoma is a progressive disease that can lead to permanent vision loss, making progression prediction vital for guiding effective treatment. Deep learning aids progression prediction but may yield unequal outcomes across demographic groups. We proposed a model called FairDist, which utilized baseline optical coherence tomography scans to predict glaucoma progression. An equity-aware EfficientNet was trained for glaucoma detection, which was then adapted for progression prediction with knowledge distillation. Model accuracy was measured by the AUC, Sensitivity, Specificity, and equity was assessed using equity-scaled AUC, which adjusts AUC by accounting for subgroup disparities. The mean deviation, fast progression, and total deviation pointwise progression were explored in this work. For both progression types, FairDist achieved the highest AUC and equity-scaled AUC for gender and racial groups, compared to methods with and without unfairness mitigation strategies. FairDist can be generalized to other disease progression prediction tasks to potentially achieve improved performance and fairness.

PMID:40707743 | DOI:10.1038/s41746-025-01884-9

Eur Radiol. 2025 Jul 24. doi: 10.1007/s00330-025-11843-3. Online ahead of print.

ABSTRACT

OBJECTIVES: To compare the diagnostic performance of deep learning reconstruction (DLR) of zero echo time (ZTE) MRI for structural lesions in patients with axial spondyloarthritis, against T1WI and ZTE MRI without DLR, using CT as the reference standard.

MATERIALS AND METHODS: From February 2021 to December 2022, 26 patients (52 sacroiliac joints (SIJ) and 104 quadrants) underwent SIJ MRIs. Three readers assessed overall image quality and structural conspicuity, scoring SIJs for structural lesions on T1WI, ZTE, and ZTE DLR 50%, 75%, and 100%, respectively. Diagnostic performance was evaluated using CT as the reference standard, and inter-reader agreement was assessed using weighted kappa.

RESULTS: ZTE DLR 100% showed the highest image quality scores for readers 1 and 2, and the best structural conspicuity scores for all three readers. In readers 2 and 3, ZTE DLR 75% showed the best diagnostic performance for bone sclerosis, outperforming T1WI and ZTE (all p < 0.05). In all readers, ZTE DLR 100% showed superior diagnostic performance for bone erosion compared to T1WI and ZTE (all p < 0.01). For bone sclerosis, ZTE DLR 50% showed the highest kappa coefficients between readers 1 and 2 and between readers 1 and 3. For bone erosion, ZTE DLR 100% showed the highest kappa coefficients between readers.

CONCLUSION: ZTE MRI with DLR outperformed T1WI and ZTE MRI without DLR in diagnosing bone sclerosis and erosion of the SIJ, while offering similar subjective image quality and structural conspicuity.

KEY POINTS: Question With zero echo time (ZTE) alone, small structural lesions, such as bone sclerosis and erosion, are challenging to confirm in axial spondyloarthritis. Findings ZTE deep learning reconstruction (DLR) showed higher diagnostic performance for detecting bone sclerosis and erosion, compared with T1WI and ZTE. Clinical relevance Applying DLR to ZTE enhances diagnostic capability for detecting bone sclerosis and erosion in the sacroiliac joint, aiding in the early diagnosis of axial spondyloarthritis.

PMID:40707731 | DOI:10.1007/s00330-025-11843-3

Sci Rep. 2025 Jul 24;15(1):26992. doi: 10.1038/s41598-025-12825-7.

ABSTRACT

This study integrates ultrasound Radiomics with clinical data to enhance the diagnostic accuracy of HER-2 expression status in breast cancer, aiming to provide more reliable treatment strategies for this aggressive disease. We included ultrasound images and clinicopathologic data from 210 female breast cancer patients, employing a Generative Adversarial Network (GAN) to enhance image clarity and segment the region of interest (ROI) for Radiomics feature extraction. Features were optimized through Z-score normalization and various statistical methods. We constructed and compared multiple machine learning models, including Linear Regression, Random Forest, and XGBoost, with deep learning models such as CNNs (ResNet101, VGG19) and Transformer technology. The Grad-CAM technique was used to visualize the decision-making process of the deep learning models. The Deep Learning Radiomics (DLR) model integrated Radiomics features with deep learning features, and a combined model further integrated clinical features to predict HER-2 status. The LightGBM and ResNet101 models showed high performance, but the combined model achieved the highest AUC values in both training and testing, demonstrating the effectiveness of integrating diverse data sources. The study successfully demonstrates that the fusion of deep learning with Radiomics analysis significantly improves the prediction accuracy of HER-2 status, offering a new strategy for personalized breast cancer treatment and prognostic assessments.

PMID:40707691 | DOI:10.1038/s41598-025-12825-7

Sci Rep. 2025 Jul 24;15(1):26890. doi: 10.1038/s41598-025-11064-0.

ABSTRACT

Tomato leaf disease detection is critical in precision agriculture for safeguarding crop health and optimizing yields. This study compares the latest YOLO architectures, including YOLOv8, YOLOv9, YOLOv10, YOLOv11, and YOLOv12, using the Tomato-Village dataset, which contains 14,368 images across six disease classes. All models are trained under identical settings to ensure a fair evaluation based on precision, recall, mean Average Precision, training time, and inference speed. Results show that YOLOv11 consistently outperforms the other architectures, achieving the highest accuracy with competitive training times and acceptable latency. YOLOv10, YOLOv8, and YOLOv12 also deliver strong results, with YOLOv12n emerging as the most effective lightweight model for resource-constrained environments. In contrast, YOLOv9 demonstrates the weakest performance, requiring more training time and exhibiting higher latency. Overall, YOLOv11 is positioned as the most effective solution for tomato leaf disease detection, providing a strong benchmark for future advancements in agricultural technology.

PMID:40707664 | DOI:10.1038/s41598-025-11064-0

Artif Intell Med. 2025 Jul 19;168:103221. doi: 10.1016/j.artmed.2025.103221. Online ahead of print.

ABSTRACT

Diabetic retinopathy (DR) is a leading cause of blindness worldwide, requiring early detection and accurate grading for effective intervention. Advances in artificial intelligence (AI), computer vision, machine learning, and deep learning (DL) have enabled automated detection and classification of DR through various imaging modalities. This review comprehensively evaluates 91 studies employing AI-based methods in the detection and classification of DR using fundus color photography, optical coherence tomography (OCT), OCT-angiography (OCTA), and fundus fluorescein angiography, providing a holistic understanding of their strengths, challenges, and limitations. Additionally, this review compares the characteristics of 23 public datasets for DR. Across modalities, DL approaches generally outperform traditional methods. Among the studies reviewed, 81% utilized fundus images, followed by 9% using OCT, 6% using OCTA, and 2% incorporating multiple modalities. Regarding classification tasks, 62% used AI for multi-way classification, 28% for binary classification, and 10% incorporated both. The paper concludes with future directions, including explainable AI frameworks, multimodal data integration, and suggested protocols to integrate into existing healthcare workflows.

PMID:40706108 | DOI:10.1016/j.artmed.2025.103221

J Med Internet Res. 2025 Jul 24;27:e70118. doi: 10.2196/70118.

ABSTRACT

BACKGROUND: Incorporating initial serum chloride levels as a prognostic indicator in the intensive care environment has the potential to refine risk stratification and tailor treatment strategies, leading to more efficient use of clinical resources and improved patient outcomes.

OBJECTIVE: Quantitative analysis of the relationship between serum chloride levels at intensive care unit (ICU) admission and in-hospital mortality, and the establishment of a personalized survival curve prediction deep learning model to enhance risk stratification and clinical decision-making.

METHODS: A large-scale, cross-country, multicohort study of 189,462 ICU patients from four cohorts was conducted: 70,370 from Medical Information Mart for Intensive Care IV (MIMIC-IV), 112,457 from eICU Collaborative Research Database (eICU-CRD; 2 US cohorts), 4653 from Yantai Yuhuangding Hospital, and 1982 patients from Zigong Fourth People's Hospital (2 Chinese cohorts). We collected demographics, underlying diseases, ICU complications, electrolyte levels, biochemical parameters, and vital signs at ICU admission, along with length of stay and in-hospital survival outcomes. Causal graph analysis pinpointed clinical variables linked to mortality. Nonlinear associations between chloride levels and mortality were evaluated using restricted cubic splines and Cox proportional hazards models, validated with the Cox frailty model, Kaplan-Meier curves, and sensitivity analyses. A deep learning model was created for individualized survival predictions.

RESULTS: Causal inference revealed a significant association between admission serum chloride levels and 28-day mortality. The median serum chloride level at ICU admission was 104 (IQR 100-108) mEq/L. In analyzing all 42 variables, restricted cubic splines identified thresholds at 103 mEq/L and 115 mEq/L, categorizing patients into three groups: ≤103 mEq/L, 103-115 mEq/L, and >115 mEq/L. Cox proportional hazards models revealed higher death risks for patients outside this range, with hazard ratios (HRs) of 1.36 (95% CI 1.29-1.43) for ≤103 mEq/L and 1.27 (95% CI 1.14-1.41) for >115 mEq/L. Four cross-cohort validations confirmed these critical ranges. For the eICU-CRD dataset, the HRs for the key intervals are 1.30 (95% CI 1.24-1.36) and 0.97 (95% CI 0.89-1.06). In the Yantai Yuhuangding Hospital affiliated with Qingdao University (YHD-HOSP) dataset, the HRs for the key intervals are 1.23 (95% CI 1.09-1.38) and 1.58 (95% CI 1.27-1.96). In the Sichuan Zigong Fourth People's Hospital (SCZG-HOSP) dataset, the HR for the key interval is 2.20 (95% CI 1.43-3.39). The Causal SurvivalNet accurately predicted individual survival curves using admission chloride levels and other factors, achieving Brier scores of 0.09, 0.12, and 0.15. Results from cohort analyses in both China and the United States consistently and closely correlate the critical range of chloride with the prognosis of ICU patients.

CONCLUSIONS: Using initial serum chloride levels enhances prognostic accuracy and facilitates tailored treatment plans for ICU patients in critical care settings.

PMID:40706028 | DOI:10.2196/70118

JMIR Med Inform. 2025 Jul 24;13:e69286. doi: 10.2196/69286.

ABSTRACT

BACKGROUND: Several clinical cases and experiments have demonstrated the effectiveness of traditional Chinese medicine (TCM) formulas in treating and preventing diseases. These formulas contain critical information about their ingredients, efficacy, and indications. Classifying TCM formulas based on this information can effectively standardize TCM formulas management, support clinical and research applications, and promote the modernization and scientific use of TCM. To further advance this task, TCM formulas can be classified using various approaches, including manual classification, machine learning, and deep learning. Additionally, large language models (LLMs) are gaining prominence in the biomedical field. Integrating LLMs into TCM research could significantly enhance and accelerate the discovery of TCM knowledge by leveraging their advanced linguistic understanding and contextual reasoning capabilities.

OBJECTIVE: The objective of this study is to evaluate the performance of different LLMs in the TCM formula classification task. Additionally, by employing ensemble learning with multiple fine-tuned LLMs, this study aims to enhance classification accuracy.

METHODS: The data for the TCM formula were manually refined and cleaned. We selected 10 LLMs that support Chinese for fine-tuning. We then employed an ensemble learning approach that combined the predictions of multiple models using both hard and weighted voting, with weights determined by the average accuracy of each model. Finally, we selected the top 5 most effective models from each series of LLMs for weighted voting (top 5) and the top 3 most accurate models of 10 for weighted voting (top 3).

RESULTS: A total of 2441 TCM formulas were curated manually from multiple sources, including the Coding Rules for Chinese Medicinal Formulas and Their Codes, the Chinese National Medical Insurance Catalog for proprietary Chinese medicines, textbooks of TCM formulas, and TCM literature. The dataset was divided into a training set of 1999 TCM formulas and test set of 442 TCM formulas. The testing results showed that Qwen-14B achieved the highest accuracy of 75.32% among the single models. The accuracy rates for hard voting, weighted voting, weighted voting (top 5), and weighted voting (top 3) were 75.79%, 76.47%, 75.57%, and 77.15%, respectively.

CONCLUSIONS: This study aims to explore the effectiveness of LLMs in the TCM formula classification task. To this end, we propose an ensemble learning method that integrates multiple fine-tuned LLMs through a voting mechanism. This method not only improves classification accuracy but also enhances the existing classification system for classifying the efficacy of TCM formula.

PMID:40705933 | DOI:10.2196/69286

ACS Appl Mater Interfaces. 2025 Jul 24. doi: 10.1021/acsami.5c09372. Online ahead of print.

ABSTRACT

Real-time, accurate, and passive temperature monitoring is critical for industrial and scientific applications. However, conventional temperature sensors often require external power, rely on complex instrumentation, and may perturb the thermal field, compromising measurement accuracy in passive sensing scenarios. Although thermochromic materials offer visual and passive temperature feedback, their utility is limited by narrow sensitivity ranges and subjective interpretation. To address these challenges, this study introduces a machine-vision-enabled microarray passive temperature sensor (MAPTS) inspired by the cooperative perception mechanism of insect compound eyes. The system comprises arrays of organic thermochromic materials patterned via soft lithography on flexible, thermally conductive substrates, enabling wide-range passive thermal sensing. A deep learning-based ResNet-34 architecture deciphers the color-to-temperature relationship from optical images, facilitating high-precision, noncontact regression-based temperature prediction. Experimental results demonstrate that the MAPTS achieves dynamic thermal responses across 0-70 °C with a rapid prediction time of 50 ms. In a high-density 7 × 7 array configuration, the system exhibits better extrapolation performance (R2 = 0.9996) and higher prediction accuracy (mean absolute error ≤ ±0.3 °C), compared to conventional thermochromic sensing methods. This work presents a cost-effective, highly accurate, and reliable approach for intelligent temperature monitoring in diverse applications.

PMID:40705845 | DOI:10.1021/acsami.5c09372

Disabil Rehabil Assist Technol. 2025 Jul 24:1-16. doi: 10.1080/17483107.2025.2530674. Online ahead of print.

ABSTRACT

A multi-technology approach is helpful for challenging people to sense, localisation and user-centred route maps to navigate between destinations. Visually challenged people need navigation assistance between source and destination to choose a hazard-free or minimal-hazard optimal path. Navigator invokes online and offline hazard-free route maps to establish a safe and optimised route between source and destination. An online navigation system finds hazard-free optimal paths by artificial intelligence (AI), deep learning (DL), machine learning (ML) and cloud. The offline navigator uses a tensor processing unit (TPU) to generate a path map. Navigator uses AI and DL techniques to identify tree branches, signboards, underside of parked vehicles, open glass windows bumping into another walking person and fast-moving objects in outdoors and predicts artificial and natural hazards for selecting hazard-free optimised paths. The proposed fuzzy trusted hazard free routing path (FTHRP) algorithm utilises the data set "hazard-route data set" to identify the obstacle-free path between the source and destination by path planning and dynamic re-routing to avoid unexpected hazards. The navigation system leverages semantic route mapping based on AI-driven context inference processed by hardware and software.

PMID:40705835 | DOI:10.1080/17483107.2025.2530674