IEEE Transactions on Biomedical Engineering最新文献

Objective: Tissue oxygen saturation (${StO}_{2}$) is clinically useful for assessing local perfusion, but many NIRS systems are expensive and bulky. We developed PhasorDetect, a handheld multispectral continuous-wave NIRS device designed for real-time, baseline-relative ${StO}_{2}$ monitoring.

Methods: PhasorDetect (8 wavelengths, 525-940 nm) was evaluated against the Hutchinson InSpectra during vascular occlusion tests in healthy participants (n = 14; both arms). InSpectra was placed over the thenar eminence, and PhasorDetect over the distal forearm. We assessed three complementary approaches: an analytical MBLL-based estimator, a baseline-drop regressor for continuous baseline-relative estimation, and an early-alert classifier for detecting baseline-relative declines. Model evaluation used arm-wise cross-validation to prevent leakage.

Results: Continuous ${StO}_{2}$ regression (analytical and learning-based) matched the direction of ${StO}_{2}$ changes but did not consistently match drop magnitude across participants, consistent with limitations of coaxial reflectance measurements. We therefore reframed monitoring as baseline-relative early alerting and evaluated classifiers for $ge$10% and $ge$20% ${StO}_{2}$ drops. This approach yielded the most effective performance among the tested models.

Conclusion: PhasorDetect reliably tracks within-subject oxygenation dynamics during occlusion and reperfusion, supporting baseline-relative monitoring rather than absolute use.

Significance: A low-cost, portable multispectral NIRS platform can enable trend monitoring and early warning of perfusion compromise in settings in which conventional devices are impractical.

Objective: Conventional analysis approaches of evoked EEG and MEG typically rely on assumptions of independence or uncorrelatedness, fixed temporal windows, and predefined regions of interest to extract neural responses. However, cortical activity is spatially and temporally overlapping and interactive, meaning that independence or uncorrelatedness cannot be assumed. As a result, these methods often fail to account for overlapping cortical activity and individual variability, limiting the accuracy and interpretability of the results. To enhance the validity of research and clinical inferences, we aimed to reliably isolate spatiotemporally localized neural evoked responses.

Methods: We developed Spatiotemporal Event Response ENcoding ("SEREN"), an algorithm that automatically identifies spatiotemporally localized evoked response components by leveraging the spatiotemporal density properties of post-synaptic currents using Gaussian kernels in time and space. SEREN can extract individual evoked response components for both averaged and single-trial data and operates in both sensor and source space.

Results: We demonstrate SEREN's effectiveness on auditory and visual-evoked MEG data as well as on simulated datasets. Additionally, we show that SEREN can be calibrated for robust single trial monitoring in noisy EEG systems using transcranial magnetic stimulation-evoked potentials, including simulated in real-time applications.

Conclusion: SEREN reliably isolates cortical evoked responses, overcoming limitations of conventional analysis approaches that do not account for inter-response overlaps or individualization.

Significance: By improving the precision of neural response extraction, SEREN provides a powerful tool for advancing the analysis of neural dynamics and improving the validity of research and clinical applications.

Objective: Accurate and non-invasive mapping of motor unit (MU) territories is essential for linking motoneuron activity with muscle contraction. Ultrafast ultrasound (UUS) enables high-resolution mechanical imaging of MUs; however, existing methods show limited spatiotemporal consistency of MU territories and lack sufficient validation. This study aims to develop a UUS-based approach to extract MU twitch areas with high spatiotemporal precision.

Methods: We propose P2P-R2: an automated two-step framework that integrates global intensity and temporal similarity features of muscular twitches to extract and refine MU twitch areas from spike-triggered averaged (STA) UUS data. To generate the STA data and provide validation, dual-probe UUS images and intramuscular EMG signals were concurrently recorded. To benchmark the proposed framework, multiple feature extraction strategies, including intensity-based, similarity-based, and previously published methods, were implemented and compared using spatial and temporal evaluation metrics.

Results: P2P-R2 significantly outperformed all single-feature and existing methods, achieving higher within-region twitch consistency ($R^{2}_{s}$ = 0.96 $pm$ 0.01) and between-probe twitch agreement ($R$ = 0.88 $pm$ 0.26) than Naive STA ($R^{2}_{s}$ = 0.84 $pm$ 0.19, $R$= -0.03 $pm$ 0.62). It also reduced centroid-to-electrode distance (10.36mm $pm$ 6.35mm) and improved spatial agreement (RoA = 0.09 $pm$ 0.10). Furthermore, P2P-R2 captured complex MU activity patterns, including twisting, splits, and asynchronous motion.

Conclusion and significance: P2P-R2 enables precise and robust MU twitch area extraction across both spatial and temporal domains. Its fully automated, source-agnostic design supports transition to fully non-invasive applications in neuromuscular diagnostics, motor unit tracking, and human-machine interfaces.

Objective: Sleep spindles, characteristic waveforms of N2 sleep in EEG, are associated with various neural processes such as cognitive function. However, their identification relies on visual inspection by experts-a time-consuming, labor-intensive, and low inter-rater consistency process that impedes cutting edge spindle research.

Methods: We introduce S5, an automatic method for sleep spindle detection employing a novel encoder-decoder architecture for time-series segmentation. A two-stage training paradigm, comprising task-agnostic pre-training followed by downstream fine tuning, ensures high-precision identification.

Results: S5 demonstrates robust and competitive performance on two public datasets. On the multi-expert annotated MODA dataset, our method outperforms the average human expert. We further conducted an exploratory analysis on a large-scale unlabeled dataset of over 7,000 recordings as a physiological sanity check.

Significance: S5 offers a precise and efficient solution for automating spindle detection, thereby accelerating related research. An accompanying graphical toolbox makes our method accessible for simple and intuitive analysis.

In recent years, artificial neural networks have been effectively used to improve the target recognition performance of steady-state visual evoked potential (SSVEP) based Brain-Computer interfaces (BCIs). However, these models require the collection of a large number of calibration trials from users, which typically results in a poor user experience. When fewer calibration trials are acquired this leads to insufficient training of model parameters and weak recognition performance. To tackle these issues, this study proposes a two-branch multi-scale convolutional correlation network (TBMSCCN) in which a correlation network framework is introduced to reduce the model training parameters and prior knowledge of the SSVEP is used to enhance the model representation ability and convergence. First, a multi-scale temporal convolution module is designed to learn local temporal dependencies in a parallel two-branch feature extraction module. Next, a contrastive loss function is constructed in the latent feature space, which can guide the model to learn the intra-class consistent features while speeding up model convergence. Finally, a group convolution module is used as a decision layer to reduce the network parameters, while learning distinguishability features between targets and non-targets. Our offline tests on two public datasets show that proposed TBMSCCN method outperforms TRCA, eTRCA, DNN, Conv-CA and Bi-SiamCA in individual calibration scenarios, which can achieve an average information transform rates (ITRs) of 378.03 ± 139.18 bit/min and 198.92 ± 111.27 bit/min on the "Benchmark" dataset and the "Beta" dataset respectively. Additionally, proposed TBMSCCN method outperform FBCCA, ttCCA, EEGNet, and TST-CFSR in calibration-free scenarios. Furthermore, an online Chinese spelling experiment confirmed the real-world effectiveness of the proposed method. The proposed model has the characteristics of low parameter and strong robustness, which can facilitate the practical engineering application of SSVEP-Based-BCI system. The code is available at https://github.com/xinjieHe123/TBMSCCN.

Objective: Diffusion Tensor Imaging (DTI) is a special magnetic resonance imaging (MRI) technique. Most of the existing research on DTI data primarily focuses either on Structural Connectivity (SC) networks derived from DTI or on DTI-derived metrics like Fractional Anisotropy, Mean Diffusivity, $lambda _{1}$, $lambda _{2}$, and $lambda _{3}$. This may lead to the neglect of potential complementary information provided by different graphs, thereby preventing the improvement of classification performance. In this study, we propose a graph neural network framework based on a dual graph strategy using DTI data for the diagnosis of ASD.

Methods: Specifically, we have done the following: 1) To address the challenges of small datasets and class imbalance, we employed data augmentation techniques (including replication of minority class samples and the mixup method) to enhance data diversity and representativeness. 2) We combined a threshold-based real physical connectivity adjacency matrix with a local microstructure adjacency matrix learned from node features to mitigate the limitations of relying on single structural information. 3) We designed a Multi-Layer Pooling Fusion (MLPF) method to capture multi-layered and richer feature representations.

Results: Our proposed method was evaluated on 198 subjects and the experimental results showed that our proposed method outperformed multiple existing methods in five-fold cross-validation, achieving 75.24% accuracy and 73.12% AUC.

Conclusion: DTI is crucial for analyzing connectivity abnormalities in ASD. Our proposed method enables more efficient, objective, and reliable diagnosis of ASD.

Significance: This work provides a valuable reference framework for utilizing DTI data in research on neurological disorders.

Accurate assessment of patients with disorders of consciousness (DoC) remains a major clinical challenge due to the limitations of behavior-based evaluations and task-dependent neurophysiological paradigms. Whole-night polysomnography (PSG), a passive and noninvasive monitoring tool, offers unique potential for revealing residual brain function during sleep. In this study, we propose a temporal-dynamic feature extraction and aggregation framework for PSG analysis to enable machine learning-based diagnosis and prognosis in DoC patients. Whole-night EEG/EOG signals were segmented into non-overlapping 30-second epochs, from which time-domain, spectral, and nonlinear complexity features were extracted. To obtain a unified and compact representation of variable-length feature sequences, two aggregation strategies were applied: stage- wise averaging based on sleep staging and clustering-based grouping via unsupervised learning. A two-stage feature selection pipeline further reduced dimensionality while preserving discriminative power and interpretability. Classifiers trained on the aggregated features achieved strong performance in distinguishing minimally conscious state (MCS) from vegetative state (VS), with AUC values exceeding 0.84, and demonstrated robust predictive ability for long-term recovery outcomes (AUC=0.79). These findings highlight the diagnostic and prognostic value of whole-night PSG and support the development of fully automated, task-free assessment tools for DoC.

Objective: Abnormal coupling of elbow flexors with shoulder abductors-the "flexor synergy"-is a common post stroke motor impairment that interferes with upper extremity function. Previous studies have shown that practicing elbow extension while loading shoulder abductors can improve independent joint control. However, these approaches often involve expensive and bulky equipment, limiting their use in clinic. SepaRRo is a semi-passive rehabilitation robot that uses brakes to generate training forces, reducing its cost relative to existing systems. SepaRRo can also load the shoulder abductors during horizontal planar reaching, suggesting that it could target the flexor synergy. However, it is unclear if training with a semi-passive robot can produce out-of-flexor synergy kinematic adaptations.

Methods: Chronic stroke survivors (n = 15) with upper extremity impairment participated in a randomized, crossover-design experiment where they reached for a functional target with their more-impaired limb in two conditions: SepaRRo resisting their motion to the target (Resistance), and SepaRRo generating an additional lateromedial force to load their shoulder abductors (Steering). For each condition, we measured changes in reaching kinematics with motion capture equipment.

Results: Following the Steering condition, participants demonstrated significantly greater shoulder abduction than the Pre-test and the Resistance condition (p0.112). Participants reduced their el bow extension following the Resistance condition (p = 0.018).

Conclusion: Steering facilitated out-of-synergy adaptations that were not present following simple resistance.

Significance: Conventional training methods may facilitate post-stroke synergies and impede recovery, while SepaRRo's steering forces may lead to improvements in independent joint control in stroke survivors.

Non-Cartesian k-space sampling in MRI is widely used, yet images reconstructed on scanners with preliminary corrections (e.g. off-resonance) often exhibit residual artifacts (e.g. ringing and streaking) that can compromise interpretation. We propose a zero-shot residual artifact suppression method that operates directly on scanner-reconstructed images without requiring labeled data, pre-training, or an explicit degradation model. The method builds on a decoder-style generative prior and incorporates a fixed blur-kernel operator that reshapes the network's inductive bias without introducing additional learnable parameters. We formulate the procedure as an optimization problem by minimizing a data-fidelity objective between the network output and the corrupted input image. We evaluate the method on simulated data and demonstrate improved image quality over conventional baselines, while remaining competitive with supervised comparisons under acceleration factors up to R = 4. Across these settings, relative to the artifact-corrupted input, SSIM improves by up to 38% and PSNR increases by up to 10.64 dB. In in vivo experiments, the proposed method consistently attenuates residual aliasing-like artifacts, indicating reproducible performance across acquisitions. Overall, the proposed framework offers a practical and general-purpose post-processing strategy for artifact suppression in non-Cartesian MRI, with applicability across diverse sampling patterns and imaging settings.

The advancement of intelligent surgery has imposed greater requirements on the precision and real-time performance of pulmonary minimally invasive surgical navigation. However, existing intraoperative navigation techniques, including optical tracking, X-ray imaging, and magnetic resonance imaging (MRI), have inherent limitations such as inadequate real-time performance, complicated workflows, strong equipment dependency, and restricted visual fields. These constraints hinder the ability of interventional surgeries to provide continuous and stable three-dimensional coordinate feedback in deep, non-line-of-sight environments. Therefore, this study proposes an electric current field source reconstruction method for determining the terminal coordinates of surgical actuators. An electric current is injected from the tip of the surgical instrument, creating an electric field within the human tissue. The potential measured by surface electrodes are then used to reconstruct the current source coordinates, enabling real-time and active sensing of the surgical probe coordinates. A mathematical model for electric current field-based coordinate positioning was developed, involving analyses of the forward and inverse problems as well as coordinate reconstruction. Random single-point positioning simulations were conducted, and a 16 + 1-electrodes experimental platform was constructed for coordinates navigation tests to evaluate positioning and navigation performance. In addition, dynamic positioning experiments of multiple physiological tissues were carried out to assess the robustness and anti-interference capability of the proposed method. Experimental results indicate that the positioning error remains within 2 mm under single-point, linear, and curved trajectory conditions, satisfying the precision requirements for intraoperative navigation. This method significantly improves the accuracy and safety of surgical positioning and navigation, thereby holding substantial engineering significance and clinical value for the advancement of intelligent surgical systems.