IEEE Transactions on Biomedical Engineering最新文献

Objective: The interpretation of electrocardiogram (ECG) signals is vital for diagnosis of cardiac conditions. Traditional methods rely on expert knowledge, which is time consuming, costly and potentially misses subtle features. AI has shown promise in ECG interpretation, but clinically desired model explainability is often lacking in literature.

Methods: We introduce an explainable AI method for ECG classification by partitioning the variational autoencoder (VAE) latent space into a label-specific and a non-label-specific subset. By optimizing both subsets for signal reconstruction and one subset also for prediction while constraining the other from learning label-specific information with an adversarial network, the latent space is disentangled in a supervised manner. This latent space is leveraged to create enhanced visualizations for ECG feature interpretation by means of attribute manipulation. As a proof of concept, we predict the left ventricular function (LVF), a critical prognostic determinant in cardiac disease, from the ECG.

Results: Our study demonstrates the effective segregation of LVF-specific information within a single dimension of the VAE latent space, without compromising classification performance. We show that the proposed model improves state-of-the-art VAE methods (AUC 0.832 vs. 0.790, F1 0.688 vs. 0.640) in prediction and performs comparable to ground truth LVF (concordance 0.72 vs.0.72) in predicting survival.

Conclusion: The model facilitates the interpretation of LVF predictions by providing visual context to ECG signals, offering a general explainable and predictive AI method.

Significance: Our explainable AI model can potentially reduce time and expertise required for ECG analysis.

With the advancement of neuroscience and computer science, electroencephalography (EEG) has drawn increasing attention as a promising modality for biometric identification, owing to its universality, permanence, and security. However, existing studies have pointed out that maintaining stable and temporally robust inter-individual features remains a major challenge in EEG-based identification. Therefore, developing effective methods for cross-time EEG-based identity recognition is essential for achieving reliable and practical biometric systems. In this study, we propose a novel EEG-based identification framework grounded in symmetric positive definite (SPD) manifolds. Specifically, we utilize the spatial covariance matrices of EEG signals to represent individual differences and introduce an enhanced feature extraction method (E-SPD-M) that simultaneously captures temporal, spatial, and spectral characteristics. These matrices are embedded into the Riemannian manifold to construct a discriminative representation space. For each subject, we build a personalized classification model and integrate their outputs to achieve accurate identification. Furthermore, we construct a comprehensive multi-task, cross-time EEG dataset and validate our approach on both our dataset and a publicly available longitudinal EEG dataset (M3CV). Experimental results demonstrate that our method achieves superior cross-time identification performance. Overall, this work offers a novel pathway for improving EEG-based biometric algorithms and extending the application of Riemannian geometry in the field.

Objective: Recent advances in electroencephalography (EEG) technology present new opportunities for mobile neuroimaging and real-world human neuroscience studies. However, EEG is sensitive to many sources of artifacts, and it can be especially difficult to remove high amplitude motion artifacts.

Methods: We demonstrate a novel method using generalized eigen decomposition (GED) for artifact removal, validated this approach using semi-simulated and real artifactual EEG collected during walking and jogging, and showed the feasibility of using cleaned ambulatory EEG data for brain microstate analysis.

Results: We found that GED is effective even in ultra-low SNR (0.1 - 5) conditions, achieving a correlation of 0.93 and RMSE of 1.43 μV in recovering ground truth activity using semi-simulated data, and increased the number of brain components by 10.9 and 11.8 for real data. GED showed superior performance compared with artifact subspace reconstruction (ASR) and independent component analysis (ICA) methods on semi-simulated data in very low SNR regimes. Using cleaned data, we were able to extract canonical EEG microstates across all tasks and sessions for examining task-related modulation in microstate duration, occurrence, and time coverage. We observed increased duration, occurrence and time coverage of microstates A and duration of microstate B, and decreased occurrence and time coverage of microstate D during motion compared with rest, corresponding to heightened alertness and increased visual processing.

Conclusion: These findings not only validate our artifact removal approach but also open new avenues for investigating neural dynamics during naturalistic human behavior.

Over the last decade, 3D ultrafast ultrasound imaging has been used in different applications including Elastic Tensor Imaging (ETI) based on 3D Shear Wave Elastography and Backscatter Tensor Imaging (BTI). BTI and ETI can provide important biomechanical and structural parameters of fibrous soft tissues such as the skeletal muscles or the myocardium. However, 3D ultrafast imaging requires 2D transducers arrays with a large number of elements, which mainly limits their use to laboratory research settings. This study aims to develop a clinically transposable ultrasound system combining 3D-ETI and BTI to characterize anisotropic tissues. A low channel count system with 128 channels based on a vantage system and a dedicated matrix transducer driven at 2.5MHz was developed. The performance of the approach was demonstrated on anisotropic and fibrous phantoms. In-vivo feasibility was performed on the brachii biceps of 4 healthy volunteers at controlled contractions levels using weights held in the hand. Using this approach, we could investigate the functional change of muscle stiffness during contraction (shear wave speed from 3.2±0.20m/s to 6.6±0.58m/s, and elastic fractional anisotropy from 0.26±0.04 to 0.49±0.07). Structural characterization was performed with BTI, fiber organization and coherence fractional anisotropy remained constant with contraction (0.27±0.05). This novel device enables non-invasive characterization of anisotropic tissues, discerning stress and structural anisotropy in promising applications in musculoskeletal and myocardial pathologies.

Objective: Young children and infants, especially newborns, are highly susceptible to seizures, which, if undetected and untreated, can lead to severe long-term neurological consequences. Early detection typically requires continuous electroencephalography (cEEG) monitoring in hospital settings, involving costly equipment and highly trained specialists. This study presents a low-cost, active dry-contact electrode-based, adjustable electroencephalography (EEG) headset, combined with an explainable deep learning model for seizure detection from reduced-montage EEG, and a multimodal artifact removal algorithm to enhance signal quality.

Methods: EEG signals were acquired via active electrodes and processed through a custom-designed analog front end for filtering and digitization. The adjustable headset was fabricated using three-dimensional printing and laser cutting to accommodate varying head sizes. The deep learning model was trained to detect neonatal seizures in real time, and a dedicated multimodal algorithm was implemented for artifact removal while preserving seizure-relevant information. System performance was evaluated in a representative clinical setting on a pediatric patient with absence seizures, with simultaneous recordings obtained from the proposed device and a commercial wet-electrode cEEG system for comparison.

Results: Signals from the proposed system exhibited a correlation coefficient exceeding 0.8 with those from the commercial device. Signal-to-noise ratio analysis indicated noise mitigation performance comparable to the commercial system. The deep learning model achieved accuracy and recall improvements of 2.76% and 16.33%, respectively, over state-of-the-art approaches. The artifact removal algorithm effectively identified and eliminated noise while preserving seizure-related EEG features.

Teleultrasound is a promising application of telemedicine, enabling remote diagnostic imaging through teleoperated systems. While previous reviews have addressed various aspects such as clinical applications, autonomy levels, or assistive functionalities, a systematic analysis of teleoperation architectures-the core enabler of remote manipulation and feedback-remains lacking. This paper presents a comprehensive review of thirty-nine teleultrasound systems, focusing specifically on teleoperation architectures. Each system is examined in terms of leader and follower design, control strategies, force feedback implementation, time delay handling, communication infrastructure, and anatomical targets. The aim of this review is the evaluation of system transparency and stability, analyzed not only in relation to devices' mechanical design and control implementation, but specifically from a comprehensive architectural point of view. Also, the evaluation critically takes into account the realism of the test conditions. Through this lens, the review shows how architectural choices critically shape the performance and safety of remote ultrasound interactions. Open challenges and future directions are also discussed, offering a reference framework for researchers and developers working on next-generation remote ultrasound technologies.

Objective: The accurate reconstruction of extended cortical activity from M/EEG data is a difficult, ill-conditioned problem. This work proposes to model distributed sources through spectral graph wavelets on the cortical surface, and addresses resulting numerical optimization problems. The objective is accurate localization, especially for extended sources, together with quantitatively relevant amplitude and time course.

Approach: Unknown sources are expanded on a system of spectral graph wavelets (SGW) defined on the cortical surface. Unknown wavelet coefficients are estimated using either variational or Bayesian formulations, involving priors that favor extended sources through sparsity in the wavelet domain: sparsity-inducing regularization, or sparse Bayesian learning (SBL). These approaches are tested and compared with concurrent approaches on real (open-access) data and numerical simulations. The quality of reconstructions is assessed using complementary metrics.

Results: SGW-based approaches are able to identify accurately extended sources. The combination with SBL is particularly attractive, as it doesn't involve hyperparameter tuning and automatically adapts to the signal to noise ratio (SNR). It yields accurate and robust results with respect to all considered metrics, and performs remarkably well in terms of depth bias.

Conclusion: This paper demonstrates the usefulness of spectral graph cortical wavelets for reconstructing cortical activity from M/EEG data, especially when coupling spatial wavelets with SBL.

Significance: Being able to identify localization, depth, amplitude and time course of brain activity from M/EEG data is important in clinical applications such as epilepsy, as it can improve the detection of potential sources of seizures.

Photonics and Optoelectronics are becoming increasingly important for use in implantable devices. In animal trials, although not desirable, failure can lead to the direct replacement of either the component or the test subject. In human implementations, however, the longevity of implantable devices typically needs to exceed 5 years and, in some cases, decades. Traditional hermetic metal packages, per definition, are impervious to water vapour. However, such packaging is unsuitable for structures which are millimetre sized or less. Brain probes encompassing optical micro-emitters must, therefore, use protective passivation/encapsulation layers such as silicon oxynitrides/silicone. However, such protection is prone to electrolytic failure driven by the LED-driving voltages. In this paper, we describe an electrical driving methodology which can improve device lifetime of encapsulated devices by balancing time-averaged electric fields to zero. We have tested the method on commercial optrodes to demonstrate platform independence. We show that, with this method, the time to failure can be increased by over two orders of magnitude.

Objective: Graph signal processing (GSP) has enabled new approaches for jointly analyzing graphs and graph signals. Various classical operations, such as the Fourier transform and filtering, have been extended to this setting, along with more advanced constructs including Slepian functions. The latter provides a basis for bandlimited graph signals that are maximally concentrated in a given subgraph. Here, we propose a novel approach that introduces complex values to encode several subgraphs, enabling a richer analysis of how graph signals are expressed.

Methods: The motivating application from neuroscience is to jointly analyze brain graphs obtained from diffusion-weighted magnetic resonance imaging (MRI), with brain graph signals from functional MRI. The brain activity measured by the latter is constrained by the underlying brain graph. Complex-valued graph Slepians constructed with prior knowledge from well-known task-positive and -negative functional networks can then reflect how activity is dynamically reorganizing.

Results: The feasibility of the approach is demonstrated using synthetic data first, and then applied to data from the Human Connectome Project, revealing patterns of brain network interactions. Results are currently limited to two subgraphs, but future work will explore more extensive graph configurations.

Conclusion: Slepian functions offer new ways to decode graph signals lying on top of a graph structure.

Significance: This confirms that the proposed method provides a new representation for studying brain activity constrained by the brain's structural connectivity.

Obtaining pixel-level expert annotations is expensive and labor-intensive in the medical imaging field, especially for multi-modality imaging data like MR. Most conventional cross-modality segmentation methods rely on unsupervised domain adaptation to achieve efficient cross-domain segmentation. However, these methods are often hindered by discrepancies between the source and target domains. In this paper, we propose a new scheme for cross-modality segmentation based on foundation models, which uses spatial consistency across multiple modalities and is not affected by discrepancies between the source and target domains. This scheme allows us to use annotated data from one imaging modality to train a network capable of performing accurate segmentation on other target imaging modalities, without the need for labels or registration processes. Specifically, we propose using a SAM-based model that uses segmentation results from one imaging modality as pseudo labels and prompts to guide training and testing in the target imaging modality. Moreover, we introduce consistency-based prompt tuning and hybrid representation learning to address potential unregistered issues and noisy label problems that may arise in cross-modality segmentation. We conducted extensive validation experiments on two internal datasets and one public dataset, including liver lesion segmentation and liver segmentation. Our method demonstrates significant improvement compared to current state-of-the-art approaches.