IEEE J Biomed Health Inform. 2025 Aug 1;PP. doi: 10.1109/JBHI.2025.3594014. Online ahead of print.

ABSTRACT

Parkinson's Disease (PD) treatment is challenging due to symptom heterogeneity and the lack of a definitive cure. Lifelong medication requires personalized treatment plans developed by physicians, but such approaches are constrained by high costs and limited physician capacity. Although deep learning (DL) methods have been explored, they lack interpretability and are restricted to numerical data inputs. In this study, we propose a novel framework that leverages large language models (LLMs) to design personalized PD treatment strategies, integrating both patient information in natural language form and external textual knowledge sources (e.g., medical guidelines). To enhance effectiveness, we use Monte Carlo Tree Search (MCTS) to refine strategies and establish a robust medication recommendation dataset. To enhance reliability and interpretability, we incorporate Retrieval-Augmented Generation (RAG) and Chain-of-Thought (CoT) reasoning within the LLM system, ensuring that each proposed strategy is accompanied by step-by-step explanations and references to similar historical cases. Experimental evaluations using the Parkinson's Progression Marking Initiative (PPMI) dataset show that our method surpasses physician-prescribed treatments, achieving an average reduction of over 1.4 points in the revised unified Parkinson's disease rating scale part III (MDS-UPDRS-III) scores. Our method also outperforms the RL-method by 1.01 points on average. Furthermore, over 43% of patients achieve more than 2 point-reduction of MDS-UPDRS-III scores. A detailed case study highlights the flexibility of LLMs in dynamically adjusting medication plans for patients at different disease stages, highlighting its potential to advance personalized PD management in real-world settings.

PMID:40748804 | DOI:10.1109/JBHI.2025.3594014

IEEE J Biomed Health Inform. 2025 Aug 1;PP. doi: 10.1109/JBHI.2025.3594210. Online ahead of print.

ABSTRACT

Accurate prediction of protein-ligand affinity (PLA) is critical for drug discovery. Recent deep learning approaches have adopted data-driven models for PLA prediction by learning intrinsic patterns from one-dimensional (1D) sequential or two-dimensional (2D) graph representations of proteins and ligands. However, these low-dimensional methods overlook the three-dimensional (3D) geometric features, which are hypothesized to be critical in binding interaction. To address the above problem, we present a Geometric deep learning approach with Hybrid message passing strategies-HybridGeo, for protein-ligand affinity prediction. We adopt dual-view graph learning to model the intra- and inter-molecular atomic interactions and propose to aggregate the spatial information with hybrid strategies. In addition, to fully model the inter-residue dependency upon message aggregation, we adopt a geometric graph transformer on the residue-scale graph of protein pockets. Extensive experiments on the PDBbind dataset show that HybridGeo achieves state-of-the-art performance with a Root Mean Square Error (RMSE) of 1.172. HybridGeo also achieves the best among all baseline models on three external test sets, showcasing good generalizability and robustness. Through systematic ablation experiments, we validated the effectiveness of the proposed modules, and further demonstrated the superior performance of HybridGeo in predicting the binding affinity of macrocyclic compound complexes through case studies. Visualization analysis further indicates the biological interpretability of the model predictions. Our code is publicly available at https://github.com/anxiangbiye1231/HybridGeo.

PMID:40748800 | DOI:10.1109/JBHI.2025.3594210

IEEE J Biomed Health Inform. 2025 Aug 1;PP. doi: 10.1109/JBHI.2025.3593825. Online ahead of print.

ABSTRACT

Deep learning-based molecular generation models have shown great potential in efficiently exploring vast chemical spaces by generating potential drug candidates with desired properties. However, these models often produce chemically invalid molecules, which limits the usable scope of the learned chemical space and poses significant challenges for practical applications. To address this issue, we propose ChemFixer, a framework designed to correct invalid molecules into valid ones. Chem- Fixer is built on a transformer architecture, pre-trained using masking techniques, and fine-tuned on a large-scale dataset of valid/invalid molecular pairs that we constructed. Through comprehensive evaluations across diverse generative models, ChemFixer improved molecular validity while effectively preserving the chemical and biological distributional properties of the original outputs. This indicates that ChemFixer can recover molecules that could not be previously generated, thereby expanding the diversity of potential drug candidates. Furthermore, ChemFixer was effectively applied to a drug-target interaction (DTI) prediction task using limited data, improving the validity of generated ligands and discovering promising ligand-protein pairs. These results suggest that ChemFixer is not only effective in data-limited scenarios, but also extensible to a wide range of downstream tasks. Taken together, ChemFixer shows promise as a practical tool for various stages of deep learning-based drug discovery, enhancing molecular validity and expanding accessible chemical space.

PMID:40748798 | DOI:10.1109/JBHI.2025.3593825

Oral Radiol. 2025 Aug 1. doi: 10.1007/s11282-025-00848-9. Online ahead of print.

ABSTRACT

OBJECTIVE: The purpose of this study is the development of a deep learning model based on nnU-Net v2 for the automated segmentation of sphenoid sinus and middle skull base anatomic structures in cone-beam computed tomography (CBCT) volumes, followed by an evaluation of the model's performance.

MATERIAL AND METHODS: In this retrospective study, the sphenoid sinus and surrounding anatomical structures in 99 CBCT scans were annotated using web-based labeling software. Model training was conducted using the nnU-Net v2 deep learning model with a learning rate of 0.01 for 1000 epochs. The performance of the model in automatically segmenting these anatomical structures in CBCT scans was evaluated using a series of metrics, including accuracy, precision, recall, dice coefficient (DC), 95% Hausdorff distance (95% HD), intersection on union (IoU), and AUC.

RESULTS: The developed deep learning model demonstrated a high level of success in segmenting sphenoid sinus, foramen rotundum, and Vidian canal. Upon evaluation of the DC values, it was observed that the model demonstrated the highest degree of ability to segment the sphenoid sinus, with a DC value of 0.96.

CONCLUSION: The nnU-Net v2-based deep learning model achieved high segmentation performance for the sphenoid sinus, foramen rotundum, and Vidian canal within the middle skull base, with the highest DC observed for the sphenoid sinus (DC: 0.96). However, the model demonstrated limited performance in segmenting other foramina of the middle skull base, indicating the need for further optimization for these structures.

PMID:40748555 | DOI:10.1007/s11282-025-00848-9

Acta Biotheor. 2025 Aug 1;73(3):11. doi: 10.1007/s10441-025-09500-9.

ABSTRACT

While RNA folding prediction remains challenging, even with machine and deep learning methods, it can also be approached from a topological mathematics perspective. The purpose of the present paper is to elucidate this problem for students and researchers in both the mathematical physics and biology fields, fostering interest in developing novel theoretical and applied solutions that could propel RNA research forward. With this intention, the mathematical method, based on matrix field theory, to compute the topological classification of RNA structures is reviewed. Similarly, McGenus, a computational software that exploits matrix field theory for topological and folding predictions, is examined. To further illustrate the outcomes of this mathematical approach, two types of analyses are performed: the prediction results from McGenus are compared with topological information extracted from experimentally-determined RNA structures, and the topology of RNA structures is investigated for biological significance, both in evolutionary and functional terms. Lastly, we advocate for more research efforts to be conducted at the intersection between physics, mathematics and biology, with a particular focus on the potential contributions that topology can make to the study of RNA folding and structure.

PMID:40748481 | DOI:10.1007/s10441-025-09500-9

Abdom Radiol (NY). 2025 Aug 1. doi: 10.1007/s00261-025-05100-w. Online ahead of print.

ABSTRACT

PURPOSE: This study explored the effects of using Intelligent Quick Magnetic Resonance (IQMR) image post-processing on image quality in Field of View Optimized and Constrained Single-Shot Diffusion-Weighted Imaging (FOCUS-DWI) sequences for prostate cancer detection, and assessed its efficacy in distinguishing malignant from benign lesions.

METHODS: The clinical data and MRI images from 62 patients with prostate masses (31 benign and 31 malignant) were retrospectively analyzed. Axial T2-weighted imaging with fat saturation (T2WI-FS) and FOCUS-DWI sequences were acquired, and the FOCUS-DWI images were processed using the IQMR post-processing system to generate IQMR-FOCUS-DWI images. Two independent radiologists undertook subjective scoring, grading using the Prostate Imaging Reporting and Data System (PI-RADS), diagnosis of benign and malignant lesions, and diagnostic confidence scoring for images from the FOCUS-DWI and IQMR-FOCUS-DWI sequences. Additionally, quantitative analyses, specifically, the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM), were conducted using T2WI-FS as the reference standard. The apparent diffusion coefficients (ADCs) of malignant and benign lesions were compared between the two imaging sequences. Spearman correlation coefficients were calculated to evaluate the associations between diagnostic confidence scores and diagnostic accuracy rates of the two sequence groups, as well as between the ADC values of malignant lesions and Gleason grading in the two sequence groups. Receiver operating characteristic (ROC) curves were utilized to assess the efficacy of ADC in distinguishing lesions.

RESULTS: The qualitative analysis revealed that IQMR-FOCUS-DWI images showed significantly better noise suppression, reduced geometric distortion, and enhanced overall quality relative to the FOCUS-DWI images (P < 0.001). There was no significant difference in the PI-RADS scores between IQMR-FOCUS-DWI and FOCUS-DWI images (P = 0.0875), while the diagnostic confidence scores of IQMR-FOCUS-DWI sequences were markedly higher than those of FOCUS-DWI sequences (P = 0.0002). The diagnostic results of the FOCUS-DWI sequences for benign and malignant prostate lesions were consistent with those of the pathological results (P < 0.05), as were those of the IQMR-FOCUS-DWI sequences (P < 0.05). The quantitative analysis indicated that the PSNR, SSIM, and ADC values were markedly greater in IQMR-FOCUS-DWI images relative to FOCUS-DWI images (P < 0.01). In both imaging sequences, benign lesions exhibited ADC values markedly greater than those of malignant lesions (P < 0.001). The diagnostic confidence scores of both groups of sequences were significantly positively correlated with the diagnostic accuracy rate. In malignant lesions, the ADC values of the FOCUS-DWI sequences showed moderate negative correlations with the Gleason grading, while the ADC values of the IQMR-FOCUS-DWI sequences were strongly negatively associated with the Gleason grading. ROC curves indicated the superior diagnostic performance of IQMR-FOCUS-DWI (AUC = 0.941) compared to FOCUS-DWI (AUC = 0.832) for differentiating prostate lesions (P = 0.0487).

CONCLUSION: IQMR-FOCUS-DWI significantly enhances image quality and improves diagnostic accuracy for benign and malignant prostate lesions compared to conventional FOCUS-DWI.

PMID:40748461 | DOI:10.1007/s00261-025-05100-w

Brief Bioinform. 2025 Jul 2;26(4):bbaf372. doi: 10.1093/bib/bbaf372.

ABSTRACT

Systematically investigating interactions among molecules of the same type across different contexts is crucial for unraveling disease mechanisms and developing potential therapeutic strategies. The "A-A-B" triplet paradigm provides a principled approach to model such context-specific interactions, and leveraging third-order tensor to capture such type ternary relationships is an efficient strategy. However, effectively modeling both multilinear and nonlinear characteristics to accurately identify such triplets using tensor-based methods remains a challenge. In this paper, we propose a novel Convolutional Neural Tensor Completion (ConvNTC) framework that collaboratively learns the multilinear and nonlinear representations to model triplet-based network interactions. ConvNTC consists of a multilinear module and a nonlinear module. The former is a tensor decomposition approach that integrates multiple constraints to learn the tensor factor embeddings. The latter contains three components: an embedding generator to produce position-specific index embeddings for each tensor entry in addition to the factor embeddings, a convolutional encoder to perform nonlinear feature mapping while preserving the tensor's rank-one property, and a Kolmogorov-Arnold Network (KAN) based predictor to effectively capture high-dimensional relationships aligned with the intrinsic structure of real-world data. We evaluate ConvNTC on two types triplet datasets of the "A-A-B" type: miRNA-miRNA-disease and drug-drug-cell. Comprehensive experiments against 11 state-of-the-art methods demonstrate the superiority of ConvNTC in terms of triplet prediction. ConvNTC reveals promising prognostic values of the miRNA-miRNA interactions on breast cancer and detects synergistic drug combinations in cancer cell lines.

PMID:40748325 | DOI:10.1093/bib/bbaf372

Brief Bioinform. 2025 Jul 2;26(4):bbaf355. doi: 10.1093/bib/bbaf355.

ABSTRACT

The rapid advancement of high-throughput sequencing and other assay technologies has resulted in the generation of large and complex multi-omics datasets, offering unprecedented opportunities for advancing precision medicine. However, multi-omics data integration remains challenging due to the high-dimensionality, heterogeneity, and frequency of missing values across data types. Computational methods leveraging statistical and machine learning approaches have been developed to address these issues and uncover complex biological patterns, improving our understanding of disease mechanisms. Here, we comprehensively review state-of-the-art multi-omics integration methods with a focus on deep generative models, particularly variational autoencoders (VAEs) that have been widely used for data imputation, augmentation, and batch effect correction. We explore the technical aspects of VAE loss functions and regularisation techniques, including adversarial training, disentanglement, and contrastive learning. Moreover, we highlight recent advancements in foundation models and multimodal data integration, outlining future directions in precision medicine research.

PMID:40748323 | DOI:10.1093/bib/bbaf355

Acc Chem Res. 2025 Aug 1. doi: 10.1021/acs.accounts.5c00396. Online ahead of print.

ABSTRACT

ConspectusAs an essential branch of chemical science, biochemical analysis is widely applied in disease diagnosis, food safety testing, environmental monitoring, and other fields. Artificial intelligence (AI) technology has substantially advanced biochemical analysis, enabling the prediction and extraction of key information from large volumes of image data; these tasks were previously unattainable, particularly in visualized single-microsphere counting for biosensing assays. By modification of different biorecognition molecules, encoded microspheres of various colors and sizes can serve as ideal optical multiprobes. In uniquely designed biochemical sensing systems, different encoded microspheres specifically associate with targets through signal recognition, resulting in corresponding changes in quantity as the target concentration varies after signal transduction. Images of these encoded microspheres are captured by optical imaging equipment and processed with high speed and precision by AI technology, which decodes them into the corresponding target concentration information. This approach, with its affordable cost and user-friendly operation, can be readily adopted for the rapid and sensitive multiplexed detection of various targets, including proteins, bacteria, viruses, and antibiotics.This Account summarizes recent studies conducted by our group on developing AI-assisted visualized microspheres for biosensing analysis and highlights unique encoding-decoding strategies and various biochemical approaches. We begin by introducing our encoding rules for different microsphere characteristics (e.g., color and size) and the preparation of encoded fluorescent microspheres, which can be conjugated with various biorecognition molecules to enable specific target capture or association. Subsequently, we summarize a series of biosensing platforms developed based on the MP encoding-AI decoding strategy; these platforms integrate different biochemical sensing approaches to meet diverse detection requirements and achieve efficient signal transduction tasks, such as immunoassays, click chemistry, Argonaute (Ago) systems, clustered regularly interspaced short palindromic repeats (CRISPR) systems, and microfluidics. Meanwhile, we present customized high-speed decoding algorithms based on AI technologies such as computer vision, machine learning, deep learning, and unsupervised learning to enable the accurate processing of encoded microsphere images acquired by different imaging systems. We show how the excellent integration of advanced biosensing techniques drives changes in the number of various encoded microspheres, enabling the accurate quantification of multitarget concentrations through the special AI decoding algorithm. Furthermore, we introduce portable optical imaging devices, including AI-integrated smartphones and portable lensless holographic microscopes, for probe visualization of encoded microspheres to facilitate rapid analysis from the laboratory to point-of-care testing (POCT). Then we outline strategies to address the challenges of future applications of AI-assisted visualized microsphere biosensors, focusing on efficient biosensing approaches with enhanced encoding capacity, lightweight smartphone-based decoding apps, and integrated systems for automated biochemical analysis. This Account aims to stimulate researchers' interest in the unique attributes of AI-assisted visualized microspheres for biochemical analysis and their corresponding encoding-decoding strategies across interdisciplinary fields, including chemistry, biology, optics, and computer science.

PMID:40748254 | DOI:10.1021/acs.accounts.5c00396

Spine (Phila Pa 1976). 2025 Aug 1. doi: 10.1097/BRS.0000000000005462. Online ahead of print.

ABSTRACT

STUDY DESIGN: Retrospective and cross-sectional study.

OBJECTIVE: The study aims to develop an open software for lumbar spine image analysis enabling no-code approach to lumbar spine segmentation, grading, and intervertebral disc height index (DHI) calculations with robust evaluation of the application on six external datasets from diverse geographical regions.

SUMMARY OF DATA: The datasets used include NFBC1966 (Finland), HKDDC (Hong Kong), TwinsUK (UK), CETIR (Spain), NCSD (Hungary), SPIDER (Netherlands), and Mendeley (global). Thirty participants from each dataset were sampled for external evaluation and NFBC1966 was used for training. Annotation was performed on T2-weighted mid-sagittal slices of vertebral bodies L1-S1 and intervertebral discs L1/2-L5/S1.

MATERIALS AND METHODS: Open Lumbar Spine Image Analysis (OLSIA) application was developed to include no-code approach tools for automated segmentation, grading, DHI calculation, and batch processing capabilities by integrating the deep learning (DL) models. DL models were trained on the NFBC1966 dataset with augmentation (histogram clipping, median filtering, geometric scaling) to improve generalization. Inter-rater agreement was assessed using Dice similarity coefficient (DSC), Bland-Altman (BA) analysis for DHI measurements and a paired t-test for statistical significance.

RESULTS: Our application demonstrated 222-fold improvement in processing time compared to performing manually lumbar spine segmentation, grading and DHI calculation tasks. OLSIA's segmentation performance exhibited close correspondence with the inter-rater agreement across all six external datasets. Inter-rater reliability was high (mean DSC>90). Although paired t-test on DHI measurements is significant (P<0.05), the mean difference (0.02) of DHI from the BA plots indicates low systematic bias.

CONCLUSION: We introduced OLSIA, a user-friendly interface for lumbar spine segmentation, grading, and intervertebral DHI calculation. OLSIA empowers researchers from diverse backgrounds to efficiently use the no-code tools to accelerate their radiomics and lumbar spine image analysis workflows.

PMID:40747922 | DOI:10.1097/BRS.0000000000005462

Anal Chem. 2025 Aug 1. doi: 10.1021/acs.analchem.5c01703. Online ahead of print.

ABSTRACT

Retention time (RT) can provide orthogonal information to mass spectra, supporting the qualitative identification. However, RT is influenced by experimental conditions and column parameters, and it is difficult to have a large amount of RT data in the user's experimental conditions. Hence, various machine learning methods, including advanced deep learning approaches, have been developed for RT prediction. However, most of them were limited to a given column and operational conditions. In the meantime, data sparsity often hinders the prediction performance. In this study, we propose an MDL-TL method that combines multiple data sets to jointly train the base model. MDL-TL vectorizes the column and conditions (chromatographic parameters, CPs) using word2vec and autoencoders, and distinguishes the data sets from different chromatographic experiments by including the CPs in the compound representation. This not only augments the data but also introduces the CPs into the RT prediction, allowing the pretrained model to be efficiently transferred to different target systems by fine-tuning. MDL-TL was evaluated against five popular deep learning approaches and four machine learning approaches on 14 reversed-phase liquid chromatography data sets and 14 hydrophilic interaction liquid chromatography data sets, respectively. The results show that our method surpassed the compared methods, including transfer learning methods based on the METLIN small molecule retention time (SMRT) data set, in mean absolute error, median absolute error, mean relative error, and R2 in most cases, demonstrating that MDL-TL is a promising approach for predicting RTs for various chromatographic systems and operational conditions.

PMID:40747624 | DOI:10.1021/acs.analchem.5c01703

Phys Chem Chem Phys. 2025 Aug 1. doi: 10.1039/d5cp02009c. Online ahead of print.

ABSTRACT

Efficient chemical kinetic model inference and application in combustion are challenging due to large ODE systems and widely separated time scales. Machine learning techniques have been proposed to streamline these models, though strong nonlinearity and numerical stiffness combined with noisy data sources make their application challenging. Here, we introduce ChemKANs, a novel neural network framework with applications both in model inference and simulation acceleration for combustion chemistry. ChemKAN's novel structure augments the generic Kolmogorov-Arnold network ordinary differential equations (KAN-ODEs) with knowledge of the information flow through the relevant kinetic and thermodynamic laws. This chemistry-specific structure combined with the expressivity and rapid neural scaling of the underlying KAN-ODE algorithm instills in ChemKANs a strong inductive bias, streamlined training, and higher accuracy predictions compared to standard benchmarks, while facilitating parameter sparsity through shared information across all inputs and outputs. In a model inference investigation, we benchmark the robustness of ChemKANs to sparse data containing up to 15% added noise, and superfluously large network parameterizations. We find that ChemKANs exhibit no overfitting or model degradation in any of these training cases, demonstrating significant resilience to common deep learning failure modes. Next, we find that a remarkably parameter-lean ChemKAN (344 parameters) can accurately represent hydrogen combustion chemistry, providing a 2× acceleration over the detailed chemistry in a solver that is generalizable to larger-scale turbulent flow simulations. These demonstrations indicate the potential for ChemKANs as robust, expressive, and efficient tools for model inference and simulation acceleration for combustion physics and chemical kinetics.

PMID:40747601 | DOI:10.1039/d5cp02009c

MethodsX. 2025 Jul 17;15:103519. doi: 10.1016/j.mex.2025.103519. eCollection 2025 Dec.

ABSTRACT

Oral squamous cell carcinoma (OSCC) continues to present a major worldwide healthcare problem because patients have poor survival outcomes alongside frequent disease returns. Globocan predicts that, OSCC will result in 389,846 new cases and 188,438 deaths globally during 2022 while maintaining an extremely poor 5-year survival rate at about 50%. Our method applies residual connections with Squeeze-and-Excitation blocks along with hybrid attention systems and enhanced activation functions and optimization algorithms to boost gradient movement throughout feature extraction. Compared against established conventional CNN backbones (VGG16, ResNet50, DenseNet121, and more), the proposed ConvNeXt-SE-Attn model outperformed them in all aspects of discrimination and calibration, including precision 97.88% (vs. ≤94.2%), sensitivity 96.82% (vs. ≤92.5%), specificity 95.94% (vs. ≤93.1%), F1 score 97.31% (vs. ≤93.8%), AUC 0.9644 (vs. ≤0.945), and MCC 0.9397 (vs. ≤0.910). The findings are critical to the increased feature-representation power and the robustness of classification of the architecture. The proposed architecture employs ConvNeXt backbone with SE blocks and hybrid attention to extract essential details within class boundaries which standard models usually miss. The activation through Gaussian-based GReLU incorporates Swish activation together with DropPath regularization for producing smooth gradient patterns which lead to generalizable features across imbalanced datasets. Grad-CAM enhances interpretability by showing which image sections lead to predictions in order to enable clinical decisions. The model demonstrates its capability as an effective detection method for minimal variations in oral cells which supports precise non-invasive treatment approaches for OSCC.

PMID:40747534 | PMC:PMC12312062 | DOI:10.1016/j.mex.2025.103519

Bladder (San Franc). 2025 Apr 10;12(2):e21200042. doi: 10.14440/bladder.2024.0054. eCollection 2025.

ABSTRACT

BACKGROUND: Bladder pathophysiology encompasses a wide array of disorders, including bladder cancer, interstitial cystitis, overactive and underactive bladder, and bladder outlet obstruction. It also involves conditions such as neurogenic bladder, bladder infections, trauma, and congenital anomalies. Each of these conditions presents unique challenges for diagnosis and treatment. Recent advancements in artificial intelligence (AI) have shown significant potential in revolutionizing diagnostic methodologies within this domain.

OBJECTIVE: This review provides an updated and comprehensive examination of the integration of AI into the diagnosis of bladder pathophysiology. It highlights key AI techniques, including machine learning and deep learning, and their applications in identifying and classifying bladder conditions. The review also assesses current AI-driven diagnostic tools, their accuracy, and clinical utility. Furthermore, it explores the challenges and limitations confronted in the implementation of AI technologies, such as data quality, interpretability, and integration into clinical workflows, among others. Finally, the paper discusses future directions and advancements, proposing pathways for enhancing AI applications in bladder pathophysiology diagnosis. This review aims to provide a valuable resource for clinicians, researchers, and technologists, fostering an in-depth understanding of AI's roles and potential in transforming bladder disease diagnosis.

CONCLUSION: While AI demonstrates considerable promise in enhancing the diagnosis of bladder pathophysiology, ongoing progresses in data quality, algorithm interpretability, and clinical integration are essential for maximizing its potential. The future of AI in bladder disease diagnosis holds great promise, with continued innovation and collaboration opening the possibility of more accurate, efficient, and personalized care for patients.

PMID:40747464 | PMC:PMC12308116 | DOI:10.14440/bladder.2024.0054

Eur Heart J Imaging Methods Pract. 2025 Jul 21;3(2):qyaf094. doi: 10.1093/ehjimp/qyaf094. eCollection 2025 Jul.

ABSTRACT

AIMS: The low reproducibility of echocardiographic measurements challenges the identification of subtle changes in left ventricular (LV) function. Deep learning (DL) methods enable real-time analysis of acquisitions and may improve echocardiography. The aim of this study was to evaluate the impact of DL-based guidance and automated measurements on the reproducibility of LV global longitudinal strain (GLS), end-diastolic (EDV) and end-systolic (ESV) volume, and ejection fraction (EF).

METHODS AND RESULTS: Forty-six patients (24 breast cancer and 22 general cardiology patients) were included and underwent four consecutive echocardiograms. Six were included twice, totalling 52 inclusions and 208 echocardiograms. One sonographer-cardiologist pair used DL guidance and measurements (DL group), while another did not use DL tools and performed manual measurements (manual group). DL group recordings were also measured using a commercially available DL-based EF tool. For GLS, the DL group had a 30% lower test-retest variability than the manual group (minimal detectable change 2.0 vs. 2.9, P = 0.036). LV volumes had ∼40% lower minimal detectable changes in the DL group vs. the manual group (32 mL vs. 52 mL for EDV and 18 mL vs. 32 mL for ESV, P ≤ 0.006). This did not translate to a significant improvement in EF reproducibility in the DL group. The benchmarking method showed similar results compared with the manual group.

CONCLUSION: Combining real-time DL guidance with automated measurements improved the reproducibility of LV size and function measurements compared with usual care, but future studies are needed to evaluate its clinical effect.

TRIAL REGISTRATION NUMBER: NCT06310330.

PMID:40747448 | PMC:PMC12311362 | DOI:10.1093/ehjimp/qyaf094

bioRxiv [Preprint]. 2025 Jun 17:2025.06.11.658673. doi: 10.1101/2025.06.11.658673.

ABSTRACT

The tumor microenvironment (TME) is a critical focus for biomarker discovery and therapeutic targeting in cancer. However, widespread clinical adoption of TME profiling is hindered by the high cost and technical complexity of current platforms such as spatial transcriptomics and proteomics. Artificial Intelligence (AI)-based analysis of the TME from routine Hematoxylin & Eosin (H&E)-stained pathology slides presents a promising alternative. Yet, most existing deep learning approaches depend on extensive high-quality single-cell or patch-level annotations, which are labor-intensive and costly to generate. To address these limitations, we previously introduced HistoTME, a weakly supervised deep learning framework that predicts the activity of cell type-specific transcriptomic signatures directly from whole slide H&E images of non-small cell lung cancer. This enables rapid, high throughput analysis of the TME composition from whole slide H&E images (WSI) without the need for segmenting and classifying individual cells. In this work, we present HistoTME-v2, a pan-cancer extension of HistoTME, applied across 25 solid tumor types, substantially broadening the scope of prior efforts. HistoTME-v2 demonstrates high accuracy for predicting cell type-specific transcriptomic signature activity from H&E images, achieving a median Pearson correlation of 0.61 with ground truth measurements in internal cross- validation on The Cancer Genome Atlas (TCGA), encompassing 7,586 WSIs, 6,901 patients, and 24 cancer types, and a median Pearson correlation of 0.53 on external validation datasets spanning 5,657 WSIs, 1,775 patients and 9 cancer types. Furthermore, HistoTME- v2 resolves the spatial distribution of key immune and stromal cell types, exhibiting strong spatial concordance with single-cell measurements derived from multiplex imaging (CODEX, IHC) as well as Visium spatial transcriptomics, spanning 259 WSI, 154 patients, and 7 cancer types. Overall, across both bulk and spatial settings, HistoTME-v2 significantly outperforms baselines, positioning it as a robust, interpretable and cost-efficient tool for TME profiling and advancing the integration of spatial biology into routine pathology workflows.

PMID:40747415 | PMC:PMC12312183 | DOI:10.1101/2025.06.11.658673

Int J Biomed Imaging. 2025 Jul 23;2025:7560099. doi: 10.1155/ijbi/7560099. eCollection 2025.

ABSTRACT

Background: This study is aimed at solving the misalignment and semantic gap caused by multiple convolutional and pooling operations in U-Net while segmenting subabdominal MR images during rectal cancer treatment. Methods: We propose a new approach for MR Image Segmentation based on a multiscale feature pyramid network and a bidirectional cross-attention mechanism. Our approach comprises two innovative modules: (1) We use dilated convolution and a multiscale feature pyramid network in the encoding phase to mitigate the semantic gap, and (2) we implement a bidirectional cross-attention mechanism to preserve spatial information in U-Net and reduce misalignment. Results: Experimental results on a subabdominal MR image dataset demonstrate that our proposed method outperforms existing methods. Conclusion: A multiscale feature pyramid network effectively reduces the semantic gap, and the bidirectional cross-attention mechanism facilitates feature alignment between the encoding and decoding stages.

PMID:40747370 | PMC:PMC12313379 | DOI:10.1155/ijbi/7560099

Front Psychiatry. 2025 Jul 17;16:1596269. doi: 10.3389/fpsyt.2025.1596269. eCollection 2025.

ABSTRACT

OBJECTIVE: Sleep is a vital component of individual health, and personality traits are key factors influencing it. This study aims to investigate the relationship between personality traits and both modelassessed sleep problems and self-reported sleep quality.

METHODS: Using deep semantic understanding technology, we developed three deep learning models based on microblogs. Model 1 and Model 2 identified whether a post indicated a sleep problem, while Model 3 assessed the user's personality traits based on the Five-Factor Model (FFM). We surveyed a dataset comprising 336 active users and then applied the models to a large-scale microblog dataset containing 4,860,000 posts from 15,251 users.

RESULTS: Our experimental results revealed that: (1) conscientiousness, agreeableness, and extraversion are associated with better sleep quality, while neuroticism is linked to poorer sleep quality; (2) the relationships between sleep problems and personality traits remained consistent when the model, trained on a small survey dataset with expert annotations, was applied to the large-scale dataset.

CONCLUSIONS: These findings highlight the potential of using deep learning models to analyze the complex relationship between personality traits and sleep, offering valuable insights for future research and interventions.

PMID:40747256 | PMC:PMC12310568 | DOI:10.3389/fpsyt.2025.1596269

Cureus. 2025 Jul 1;17(7):e87130. doi: 10.7759/cureus.87130. eCollection 2025 Jul.

ABSTRACT

Heart failure (HF) is a prevalent global health concern, impacting millions and contributing to high morbidity, mortality, and healthcare costs. The management of HF involves complex strategies, and traditional approaches often fail to address the escalating burden of hospital readmissions and deteriorating patient quality of life. Artificial intelligence (AI) has emerged as a promising tool for enhancing diagnostic accuracy, personalizing treatment plans, and improving patient outcomes in HF care. This narrative review investigates how AI technologies can benefit HF patients' quality of life by improving risk assessment, patient self-management, and diagnostics. A comprehensive review of the literature was conducted through the studies on PubMed, Scopus, and Embase, primarily focusing on AI applications in HF diagnosis, management, and patient education, with key studies selected to highlight the role of AI in improving clinical outcomes and reducing hospital readmissions. AI-driven tools, such as neural networks and deep learning algorithms, have demonstrated high accuracy in the early detection of HF, enabling timely interventions that mitigate disease progression. Rule-based AI systems apply fixed clinical rules to standardize HF diagnostics but cannot adjust to individual patient differences. Machine learning methods analyze structured health records to forecast risks like hospitalizations or refine treatments. Deep learning techniques, using neural networks, detect subtle heart abnormalities in complex imaging data like echocardiograms that conventional approaches might overlook. Personalized digital health applications, including avatar-based self-management programs, have significantly improved quality of life by empowering patients to monitor symptoms, adhere to treatment regimens, and engage proactively in their care. Furthermore, AI's integration into cardiac imaging systems enhances precision in identifying subtle cardiac abnormalities. At the same time, remote monitoring technologies leverage predictive analytics to flag decompensation risks, allowing clinicians to adjust therapies preemptively. These advancements collectively optimize therapeutic strategies and reduce rehospitalization rates. Despite challenges such as implementation costs, data privacy concerns, and ethical considerations surrounding algorithmic bias, AI's evolving role in HF management highlights its transformative potential. By bridging gaps in personalized care, fostering patient engagement, and refining risk stratification, AI promises to revolutionize HF management paradigms, shifting the focus from reactive treatment to proactive, patient-centered precision medicine. This integration addresses systemic inefficiencies and holds promise for sustainable improvements in long-term outcomes and quality of life for HF patients globally.

PMID:40747166 | PMC:PMC12313163 | DOI:10.7759/cureus.87130

Photoacoustics. 2025 Jul 4;45:100736. doi: 10.1016/j.pacs.2025.100736. eCollection 2025 Oct.

ABSTRACT

The difficulty of obtaining absorption coefficient annotations hinders the practical application of deep learning in quantitative photoacoustic tomography. While training on synthetic data is easy to implement, the synthetic-to-real domain gap poses a significant challenge to model generalization. To address this, we propose a Decoder-enhanced unsupervised Domain Adaptation (DDA) framework to enable knowledge transfer from synthetic data to an unlabeled target domain. Experimental results show that DDA significantly improves estimation performance on target images and surpasses competing methods in quantitative evaluation and visual comparison. Additionally, we investigate the effect of cross-domain label distribution similarity on domain adaptation and recommend an effective approach for data synthesis. To mitigate the effect of absorption property mismatch, we propose fine-tuning the affine parameters of normalization layers, which significantly improves estimation accuracy using labeled multi-wavelength photoacoustic images from as few as two target samples.

PMID:40747132 | PMC:PMC12311538 | DOI:10.1016/j.pacs.2025.100736