Biomedical Engineering Letters最新文献

The distinction between benign and malignant thoracic vertebral compression fractures (VCFs) on magnetic resonance imaging (MRI) is often subtle, and distinguishing between them is a fine-grained classification challenge. We propose a method for benign and malignant classification of thoracic VCFs based on multi-layer feature fusion and attention-guided patch reorganization to address the problem of low-level feature loss and noise associated with computing background patches when vision transformer (ViT) is applied to the task of fine-grained MRI image classification. The approach is based on the ViT architecture, which fuses low-level features with high-level features by selecting discriminative tokens using multiple layers of mutual attention weights between the classification tokens and the tokens. In addition, we incorporate an attention-guided patch recombination module that uses attention weights to select and combine patches of any two input images, which enhances the richness of the input images while reducing the noise computation. Experiments were conducted on the thoracic VCFs dataset with quantitative assessment metrics, achieving slice-level classification accuracy and AUC of 84.87% and 84.19%, respectively. Aggregating slice-level predictions, the patient-level classification accuracy reached 93.18%. Compared to other fine-grained ViT-based methods, our approach demonstrated varying improvements in slice-level accuracy, with a maximum increase of 3.65%. Ablation experiments further validated the effectiveness of the multi-layer feature fusion and patch recombination modules. The proposed MFAR-ViT utilizes the advantages of multi-layer feature fusion and patch reorganization to identify the fine-grained differences in MRI images of thoracic VCFs more efficiently, which is expected to help doctors diagnose the patient's condition quickly and accurately.

Generative artificial intelligence (AI) models, such as diffusion models and OpenAI's ChatGPT, are transforming medicine by enhancing diagnostic accuracy and automating clinical workflows. The field has advanced rapidly, evolving from text-only large language models for tasks such as clinical documentation and decision support to multimodal AI systems capable of integrating diverse data modalities, including imaging, text, and structured data, within a single model. The diverse landscape of these technologies, along with rising interest, highlights the need for a comprehensive review of their applications and potential. This scoping review explores the evolution of multimodal AI, highlighting its methods, applications, datasets, and evaluation in clinical settings. Adhering to PRISMA-ScR guidelines, we systematically queried PubMed, IEEE Xplore, and Web of Science, prioritizing recent studies published up to the end of 2024. After rigorous screening, 145 papers were included, revealing key trends and challenges in this dynamic field. Our findings underscore a shift from unimodal to multimodal approaches, driving innovations in diagnostic support, medical report generation, drug discovery, and conversational AI. However, critical challenges remain, including the integration of heterogeneous data types, improving model interpretability, addressing ethical concerns, and validating AI systems in real-world clinical settings. This review summarizes the current state of the art, identifies critical gaps, and provides insights to guide the development of scalable, trustworthy, and clinically impactful multimodal AI solutions in healthcare.

Supplementary information: The online version contains supplementary material available at 10.1007/s13534-025-00497-1.

This study aims to demonstrate that surface engineering of cytotoxic T cells with drug-loaded nanoparticles enhances nanoparticle delivery to induce a more potent combinatorial chemotherapeutic and immunotherapeutic effect, as well as enabling spatial tracking through the use of non-invasive, real-time ultrasound-guided photoacoustic imaging. Ovalbumin (OVA)-targeting OT-1 T cells were functionalized with doxorubicin-loaded, mesoporous silica-coated gold nanorods. In vitro toxicity and synergistic effects were assessed using antigen-matched OVA-expressing melanoma cells, while in vivo studies evaluated therapeutic efficacy. Ultrasound-guided photoacoustic imaging was employed to confirm the targeted delivery of the nanoengineered cells. The integration of optically active, drug-loaded nanoparticles with T cells facilitates precise image-guided delivery and enhances nanoparticle accumulation within the tumor environment, thereby maximizing the combinatorial chemo-immunotherapeutic effect. The integration of optically active, drug-loaded nanoparticles with T cells facilitates precise image-guided delivery and enhances nanoparticle accumulation within the tumor environment, thereby maximizing the combinatorial chemo-immunotherapeutic effect.

Abstract: Hypertrophic cardiomyopathy (HCM) is a common hereditary heart disease and is the leading cause of sudden cardiac death in adolescents. Septal hypertrophy (SH) and apical hypertrophy (AH) are two common types. The former is characterized by abnormal septal myocardial thickening and the latter by left ventricular apical hypertrophy, both of which significantly increase the risk of heart failure, arrhythmias, and other serious complications. Identifying hypertrophic sites in HCM patients using 12-lead electrocardiography (ECG) is crucial for early diagnosis, staging, and prognosis. However, most deep learning methods rely on 1D one-dimensional ECG signal detection, or 2D two-dimensional ECG image or spectrogram recognition, which may result in the loss of spatial or temporal information, thus limiting diagnostic accuracy. Therefore, an optimized multi-stage network with multi-dimensional spatiotemporal interactions (Ms-MdST) is proposed for detecting AH and SH in HCM. The optimized Ms-MdST model combines the advantages of different dimensional convolutions to capture the spatiotemporal characteristics of ECG and consists of a 1D convolution branch for overall temporal features and a 2D convolution branch for similar spatial features across multiple leads. Moreover, a global-local interactive attention mechanism (GLIA) and a multi-loss joint optimization strategy are employed to facilitate multi-stage multi-scale feature fusion. Experimental results show that Ms-MdST achieves F1-scores of 0.9672, 0.7250, and 0.8009 in the CONTROL, SH, and AH groups, respectively, demonstrating its superiority compared to existing ECG classification methods. In addition, the proposed model is interpretable and can be further extended to clinical applications.

Graphical abstract:

Cage subsidence is a common complication following transforaminal lumbar interbody fusion (TLIF) that can lead to poor clinical outcomes, including recurrent pain and segmental instability. Conventional TLIF cage designs often fail to distribute stress evenly, increasing the risk of endplate damage and subsequent subsidence. This study aims to evaluate the effect of a modified TLIF cage with upper and lower open windows (lattice structure) in reducing cage subsidence in patients with lumbar degenerative disc disease (LDDD). A finite element (FE) model of the lumbar spine was developed and validated. Three TLIF cage designs (Open, Lattice, Closed) were simulated under various loading conditions (flexion-extension, lateral bending, axial rotation), and von Mises stresses were analyzed within the TLIFs, endplates, and cancellous bone. The FE model demonstrated ROMs consistent with cadaveric studies. Elevated stresses were found in all cages, especially Open and Closed designs. The Lattice TLIF showed improved stress distribution, reducing peak stress on endplates. However, increased contact area had a limited effect on reducing subsidence under physiological loads. While contact area alone does not significantly mitigate subsidence risk, incorporating lattice structures may enhance resistance to physiological stress. These findings suggest that optimized TLIF designs integrating lattice structures can improve stability and reduce the likelihood of subsidence, leading to better clinical outcomes (e.g., reduced pain, improved fusion success, long-term stability) in LDDD patients.

Hypertension influences cardiovascular diseases, such as heart attacks and strokes. Blood pressure (BP) monitoring is essential for detecting hypertension and assessing its consequences. BP was traditionally measured using stethoscopes and pressure cuffs, which had several limitations. Additionally, automated blood pressure machines are not always accurate. Blood pressure measurement can be conducted more accurately and sensitively through a novel, non-invasive, and automated method. In this paper, a hybrid classification-mapping model is proposed to estimate Systolic (SBP) and Diastolic (DBP) blood pressure using 155 subjects from the University of New South Wales non-invasive BP (NIBP) dataset. In addition to exploring new beat-related features derived from oscillometric waveforms (OW), our study employs eight distinct feature ranking techniques to optimize the performance of different machine learning classifiers (K Nearest Neighbor (KNN), Ensemble KNN, Ensemble Bagged Tree, and Support Vector Machine (SVM)). As a comparison to existing methods for estimating DBP, which report a Mean Absolute Error (MAE) of 3.42 ± 5.38 mmHg, our approach achieves remarkably comparable results for estimating SBP, with an MAE of 1.28 ± 2.27 mmHg. Considering our promising results, implementing our methodology could provide a more reliable and convenient way to monitor blood pressure via remote healthcare.

Supplementary information: The online version contains supplementary material available at 10.1007/s13534-025-00496-2.

Purpose: Idiopathic rapid eye movement (REM) sleep behavior disorder (iRBD) is a sleep disorder considered to be a prodromal stage of neurodegeneration disease and is often accompanied by cognitive impairments. The purpose of this study was to investigate spatiotemporal characteristics of abnormal oscillatory cortical activity associated with dysfunction of visuospatial attention in iRBD based on an explainable machine learning approach. Methods: EEGs were recorded from 49 iRBD patients and 49 normal controls while they were performing Posner's cueing task and transformed to cortical current density time-series. Spectral cortical activities for four frequency bands (theta, alpha, beta, and gamma) were estimated, and then converted to three-dimensional (3D) spatiotemporal data. A pattern classifier based on 3D convolutional neural network was devised to discriminate the cortical activities of iRBD patients and those of normal controls. Results: The location, time, and frequency which characterize the difference between the patients and normal controls, thereby deemed to be associated with cognitive impairment due to the iRBD, were identified by finding the input nodes which were most critical to the classifier's decision. Conclusion: Our results suggest that theta- and gamma-band activities in parietal and occipital regions, which may underlie efficient visuospatial processing and attentional reallocation, are impaired in iRBD patients, resulting in poor visuospatial attention performance.

Supplementary information: The online version contains supplementary material available at 10.1007/s13534-025-00493-5.

Conformal tissue-electrode interfaces play a vital role in the long-term high-performance operation of bioelectronic devices, enabling continuous health monitoring, precise diagnosis, and personalized therapeutics as well as human-machine interfaces in the form of electronic skin (E-skin) and prostheses. Softness and mechanical deformability of the tissue-electrode interface minimize the damage to the target tissue and allow long-term efficient signal transmission through conformal integration with dynamically moving and curved organs. We herein summarize the recent advances in the tissue-electrode interfaces for bioelectronic devices with a focus on materials, fabrication, and applications. First, we discuss material design strategies to achieve stretchable, conductive materials. Next, we present novel fabrication techniques that fulfill the requirements of tissue-electrode interfaces. Subsequently, we present the applications of these strategies to tissue-electrode interfaces, demonstrating the advancements in the functional properties of these interfaces. Finally, we conclude with a summary and a discussion on the remaining challenges and future prospects of tissue-electrode interfaces.

Hypothermia, a component of the "lethal triad," commonly complicates the condition of critically injured trauma patients, thereby substantially elevating the risk of mortality. This study develop and evaluate a dynamic warning system based on non-invasive features, aimed at predicting the likelihood of hypothermia occurring in trauma patients within the next hour. 462 patients from the eICU database were selected on the basis of meeting the inclusion criteria, and 19 non-invasive and 17 invasive features were extracted. Five classic machine learning methods were employed to develop dynamic early warning model for hypothermia based on various observation windows, with multi-center data used for model validation. The shapley additive explanations (SHAP) algorithm was utilized to analyze the interpretability of the model, and ablation experiments were conducted to further evaluate the contribution of significant feature to the prediction performance. The AUC values of the optimal models based on non-invasive features in the same test set are 0.838. When using cross-hospital data as the validation set, the highest AUC values for the same models based on non-invasive features decrease by only 0.015. In addition, ablation experiments reveal that the model's AUC exhibited a 0.010 improvement when the three most influential invasive features were incorporated into the non-invasive feature set. The results show that machine learning models have shown significant potential in predicting hypothermia through the utilization of solely non-invasive features.

Generative models have become innovative tools across various domains, including neuroscience, where they enable the synthesis of realistic brain imaging data that captures complex anatomical and functional patterns. These models, such as Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), and diffusion models, leverage deep learning to generate high-quality brain images while maintaining biological and clinical relevance. These models address critical challenges in brain imaging, e.g., the high cost and time required for data acquisition and the frequent imbalance in datasets, particularly for rare diseases or specific population groups. By conditioning the generative process on variables such as age, sex, clinical phenotypes, or genetic factors, these models enhance dataset diversity and provide opportunities to study underrepresented scenarios, model disease progression, and perform controlled experiments that are otherwise infeasible. Additionally, synthetic data generated by these models offer a potential solution to data privacy concerns, as they provide realistic non-identifiable data. As generative models continue to develop, they hold significant potential to substantially advance neuroscience by augmenting datasets, improving diagnostic accuracy, and accelerating the development of personalized treatments. This paper provides a comprehensive overview of the advancements in generative modeling techniques and their applications in brain imaging, with a particular emphasis on conditional generative methods. By categorizing existing approaches, addressing key challenges, and highlighting future directions, this paper aims to advance the integration of conditional generative models into neuroscience research and clinical workflows.