Medical & Biological Engineering & Computing最新文献

In clinical practice, the upper limb function of hemiplegic post-stroke patients is commonly evaluated using clinical tests and questionnaires. Performing a reliable investigation of compensatory strategies adopted for the upper limb movement may shed light on the basis of motor control and the mechanism of functional recovery. To quantitatively evaluate the compensatory strategies in post-stroke hemiplegic patients, we conducted an observational study in which 36 hemiplegic patients were enrolled and were stratified according to the Fugl-Meyer score. We assessed compensatory strategies in upper limb movements, specifically reaching (RCH) and hand-to-mouth (HTM) movements, using the Kinect V2 device. 11 severe, 8 severe-moderate, 9 moderate-mild, and 8 mild patients and 17 controls participated in the study. Our results showed that severe, severe-moderate, and moderate-mild patients can be discriminated from healthy participants in almost all parameters. In particular, patients showed a reduced ROM of the shoulder in RCH, higher shoulder and elbow vertical displacement, and lower wrist vertical displacement in HTM. Interestingly, compensatory parameters also discriminate mild patients from healthy controls, such as head frontal and vertical displacements. Our protocol works effectively and the instrumental assessment of compensatory strategies in post-stroke patients allows to discriminate different levels of impairments even with low-cost devices.

The 3D spinal model plays a crucial role in the assessment and treatment decision of adolescent idiopathic scoliosis. The complex 3D shape of the spine cannot be fully captured by a single radiograph. A 3D spine reconstruction framework is developed in this study. First, a dual-training strategy for Generative Adversarial Networks (GANs) is proposed, which generates high-quality 3D spinal structures. Second, an adaptive scale-agnostic attention mechanism is integrated to establish cross-layer feature correlations and dynamically allocate weights. This mechanism ensures the preservation of the crucial information across all scales throughout the feature extraction process. The proposed method has been validated on 49 cases of scoliosis. Experiments show that surface overlap and volume Dice coefficient are 0.92 and 0.94, respectively. Compared with the state-of-the-art methods, the proposed method reduces the average surface distance by 0.16 mm. The results demonstrate its effectiveness in reconstructing the 3D spine from a single radiograph.

Recognition of tumors is very important in clinical practice and radiomics; however, the segmentation task currently still needs to be done manually by experts. With the development of deep learning, automatic segmentation of tumors is gradually becoming possible. This paper combines the molecular information from PET and the pathology information from CT for tumor segmentation. A dual-branch encoder is designed based on SE-UNet (Squeeze-and-Excitation Normalization UNet) and Transformer, 3D Convolutional Block Attention Module (CBAM) is added to skip-connection, and BCE loss is used in training for improving segmentation accuracy. The new model is named TASE-UNet. The proposed method was tested on the HECKTOR2022 dataset, which obtains the best segmentation accuracy compared with state-of-the-art methods. Specifically, we obtained results of 76.10 $\%$ and 3.27 for the two key evaluation metrics, DSC and HD95. Experiments demonstrate that the designed network is reasonable and effective. The full implementation is available at https://github.com/LiMingrui1/TASE-UNet .

Deep learning has significantly advanced medical imaging, particularly computed tomography (CT), which is vital for diagnosing heart and cancer patients, evaluating treatments, and tracking disease progression. High-quality CT images enhance clinical decision-making, making image reconstruction a key research focus. This study develops a framework to improve CT image quality while minimizing reconstruction time. The proposed four-step medical image analysis framework includes reconstruction, preprocessing, segmentation, and image description. Initially, raw projection data undergoes reconstruction via a Radon transform to generate a sinogram, which is then used to construct a CT image of the pelvis. A convolutional neural network (CNN) ensures high-quality reconstruction. A bilateral filter reduces noise while preserving critical anatomical features. If required, a medical expert can review the image. The K-means clustering algorithm segments the preprocessed image, isolating the pelvis and removing irrelevant structures. Finally, the FuseCap model generates an automated textual description to assist radiologists. The framework's effectiveness is evaluated using peak signal-to-noise ratio (PSNR), normalized mean square error (NMSE), and structural similarity index measure (SSIM). The achieved values-PSNR 30.784, NMSE 0.032, and SSIM 0.877-demonstrate superior performance compared to existing methods. The proposed framework reconstructs high-quality CT images from raw projection data, integrating segmentation and automated descriptions to provide a decision-support tool for medical experts. By enhancing image clarity, segmenting outputs, and providing descriptive insights, this research aims to reduce the workload of frontline medical professionals and improve diagnostic efficiency.

This preliminary study investigates the temporal dynamics of multisensory integration in early to mid-adulthood. Five regions of interest (ROIs) were identified, and integration times from 0 to 500 ms were analyzed. The impact of temporal asynchrony on audio-visual integration was assessed through behavioral analysis. Brain topography-based age-related differences in multisensory processing, particularly in the middle-aged group, were observed. Early integration consistently occurs between 200 and 325 ms across age groups. Audio stimuli integrate slower than visual stimuli, with AV integration times falling in between. Delayed integration is observed in audio-leading conditions (A50V), while faster integration occurs in visual-leading conditions (V50A). ERP-based channel selection significantly enhances age group classification accuracy. The random forest classifier achieves 98.3% accuracy using a small set of 13 selected channels during the A50V task. This optimized channel selection improves the ergonomics of EEG-based age group classification and simplifies the clustering process. The study demonstrates the effectiveness of using minimal electrodes and straightforward features for multisensory integration tasks in early to mid-adulthood.

Endovascular aneurysm repair (EVAR) is associated with favorable short-term outcomes; however, its long-term durability can be enhanced through effective decision-making tools. Currently, most clinical decision-making relies on pre-operative CT imaging that does not fully account for vascular deformation or endovascular device behavior. To address this limitation, we present a high-fidelity virtual EVAR model designed to predict procedural outcomes and optimize treatment planning. We used only patient imaging data to reconstruct tissue structures, preserving a non-invasive workflow. Finite element simulations captured the crimping of stent grafts (SGs), navigation of endovascular devices, and SG implantation. Deformation and stored energy were continuously tracked at each procedural step. We validated the model's results against post-operative CT data using an image fusion technique. Compared to the patient's post-operative data, the model showed strong alignment, with a mean modified Hausdorff distance of 1.71 ± 1.40 mm between the simulated and actual lumen centerlines. Additionally, the 1.65 ± 1.13 mm lumen radius error further supports the model's validity. This high-fidelity, automated, and cost-effective framework can serve as a complementary tool for current EVAR pre-planning practices, potentially improving device selection, streamlining navigation roadmaps, reducing complications, and ultimately enhancing patient outcomes.

Accurate simulation of dynamic biological phenomena, such as tissue response and disease progression, is crucial in biomedical research and diagnostics. Traditional GPU-based simulation frameworks, typically static CUDA^® environments, struggle with dynamically evolving parameters, limiting flexibility and clinical applicability. We introduce Barracuda, an open-source, lightweight, header-only, Turing-complete virtual machine designed for seamless integration into GPU environments. Barracuda enables real-time parameter perturbations through an expressive instruction set and operations library, implemented in a compact C/CUDA library. A dedicated high-level programming language and Rust-based compiler enhance accessibility, allowing straightforward integration into biomedical simulation workflows. Benchmark validations, including Rule 110 cellular automaton and Mandelbrot computations, confirm Barracuda's versatility and computational completeness. In magnetic resonance imaging (MRI) simulations, Barracuda allows for the dynamic recalculation of critical parameters, such as $T_1$ relaxation times and temperature-induced off-resonance frequencies. Although it introduces computational overhead compared to static kernels, Barracuda significantly improves simulation accuracy by enabling dynamic modeling of key biological processes. Barracuda's modular architecture supports incremental integration, providing valuable flexibility for biomedical research and rapid prototyping. Future developments aim to optimize performance and expand domain-specific instruction sets, reinforcing Barracuda's role in bridging static GPU programming and dynamic simulation requirements.

Middle-aged people generally experience greater work pressure but higher health risks. However, the existing EEG-based cognitive load monitoring research has paid less attention to this segment of the population. We investigated high temporal resolution decoding of cognitive load from EEG signals in middle-aged individuals during inhibition and updating tasks. In this paper, we employed publicly available EEG data from Multi-Source Interference Task (MSIT) and Sternberg Memory Task (STMT) paradigms to examine variations in brain activation modes and cognitive load under low and high cognitive demands. This analysis was conducted using time courses of event-related potential (ERP) scalp maps. To validate the effect of the method, we conducted multivariate pattern recognition and statistics analysis. The point-by-point classification accuracy sequences obtained from decoding were assessed for significance above chance levels using one-tailed t-tests, with corrections for multiple comparisons made via the false discovery rate (FDR) method. After comparative analysis, we found that the decoder was more effective in categorizing different tasks, while the MSIT was better than STMT's in categorizing cognitive loads. In addition, we also analyzed the spatio-temporal properties of brain activation under different conditions, which is instrumental in developing more powerful classifiers. Additionally, group-level statistical comparisons were performed to explore how AD risk may influence cognitive load decodings. The study results show that this program is feasible and can be used in the future to monitor the workload of high-risk job operators in real time and longitudinal observation in medical diagnostics.

Effective Electroencephalogram (EEG) signal processing necessitates the mitigation of physiological artifacts. While deep learning frameworks have demonstrated superior performance over traditional methods for this task, their high complexity and computational demands hinder deployment on resource-constrained platforms. In this work, denoising network called EEGPARnet is proposed to address this limitation. The proposed architecture integrates transformer encoders equipped with temporal and spectral attention modules and a Gated Recurrent Unit (GRU)-based decoder. This fusion enables the model to learn time-frequency long-range similarities, facilitating efficient feature extraction and a reduced number of trainable parameters. Experimental validation of the proposed model on the EEGDenoiseNet dataset revealed an average temporal relative root mean square error ([Formula: see text]) of 0.289, spectral relative root mean square error ([Formula: see text]) of 0.312, and a correlation coefficient (CC) of 0.942 for ocular artifact removal. For muscular artifact removal, the proposed method achieved competitive results against state-of-the-art techniques, with mean [Formula: see text], [Formula: see text], and CC values of 0.458, 0.428, and 0.855, respectively. Compared to state-of-the-art model, the proposed EEGPARnet demonstrated a significant reductions in computational complexity with [Formula: see text] fewer trainable parameters, [Formula: see text] less FLOPS, and [Formula: see text] smaller storage, making it a step closer towards deployment on resource-constrained devices for real-time EEG denoising without compromising performance.