IEEE Transactions on Biomedical Engineering最新文献

Background: A wrist-worn wearable device for acquiring limb and chest ECG leads (wECG) may constitute a promising approach to detection of acute myocardial infarction (AMI). However, it remains to be demonstrated whether the information conveyed by the wECG is sufficient for AMI detection.

Objective: To explore explainable machine learning models for detecting AMI using the wECG.

Methods: Two types of machine learning models are explored: a convolutional neural network (CNN) using the raw ECG as input and a gradient-boosting decision tree (GBDT) using clinically informative features. 123 participants were included, divided into patients with AMI, patients with other cardiovascular diseases, and healthy individuals. A wrist-worn device equipped with three biopotential electrodes was used to acquire two ECG leads with a single touch: limb lead I and another lead involving a specific body site, i.e., either the V3 or V5 electrode positions, or the abdomen.

Results: The best performance on the test dataset is obtained using models that incorporate all four leads. The CNN model performs slightly better than the GBDT model, with a sensitivity of 0.77 and specificity of 0.75 compared to 0.77 and 0.72, respectively. When distinguishing between AMI and healthy participants, the specificity increases to 0.94 for the CNN model and 0.90 for the GBDT model. Feature importance analysis shows that the GBDT model primarily relies on the J point, while the CNN model primarily relies on the QRS complex.

Conclusions: wECG-based AMI detection shows considerable promise in out-of-hospital settings. However, caution is needed as CNN explanations rarely agree with the ECG intervals typically analyzed in clinical practice.

This study explores the feasibility of using an RGB camera to estimate the bilirubin level of neonates with an emphasis on applications within the Neonatal Intensive Care Unit (NICU), aiming to provide a non-contact, real-time, and continuous monitoring solution for neonatal jaundice. We investigated two fundamental models for camera-based bilirubin level monitoring: blood perfusion (AC component) based and skin reflectance (DC component) based. The blood perfusion model used the ratio of AC components in the blue and green channels, while the skin reflectance model employed the ratio of DC components in these two channels. Videos of 68 neonates in the NICU were recorded using an RGB camera and custom-built dual-wavelength light sources (460 nm and 570 nm). Clinical results showed that the blood perfusion based method negatively correlated with bilirubin concentration, contrary to our modeling and expectation, likely due to the interference of concentration in arterial blood. In contrast, the skin reflectance model demonstrated an expected strong negative correlation between DC ratio and bilirubin (i.e., r=-0.652 and p <0.005) and better consistency with the reference of transcutaneous bilirubin meter (agreement limits range = -5.72 mg/dL to 4.06 mg/dL) in intermittent bilirubin level estimation experiments. Additionally, camera-based continuous bilirubin level monitoring of resting neonates shows high potential (MAE = 4.57 mg/dL) in the NICU.

Decoding motor imagery based on electroencephalography (EEG) is limited by high data noise and high model computational complexity. Starting from EEGNet, this study achieves high-accuracy decoding through three steps. First, frequency domain analysis was performed to reveal the frequency modeling patterns of deep learning models. Utilizing prior knowledge from brain science regarding the key frequency bands for motor imagery, we adjusted the convolution kernels and pooling sizes of EEGNet to focus on effective frequency bands. Subsequently, a residual network was introduced to preserve high-frequency detailed features. Finally, temporal convolution modules were used to deeply capture temporal dependencies, significantly enhancing feature discriminability. Experiments were conducted on the BCI Competition IV 2a and 2b datasets. The 2a dataset includes multi-channel data with 22 channels, while the 2b dataset contains low-channel data with only 3 channels, reflecting significant scenario differences. Our method achieved average classification accuracies of 86.23% and 86.75% respectively, surpassing advanced models like EEG-Conformer and EEG-TransNet. Meanwhile, the Multiply-accumulate operations (MACs) were 27.16 M, a reduction of over 50% compared to the comparison models, and the Forward/Backward Pass Size was 14.33 MB. This significantly reduced computational complexity and memory footprint. This paper employs the simplest and most fundamental techniques in its design, highlighting the critical role of brain science knowledge in model development. The proposed method demonstrates broad application potential.

Hippocampal ripple-replay events are typically identified using a two-step process that at each time point uses past and future data to determine whether an event is occurring. This prevents researchers from identifying these events in real time for closed-loop experiments. It also prevents the identification of periods of non-local representation that are not accompanied by large changes in the spectral content of the local field potentials (LFPs). In this work, we present a new state-space model framework that is able to detect concurrent changes in the rhythmic structure of LFPs with non-local activity in place cells to identify ripple-replay events in a causal manner. The model combines latent factors related to neural oscillations, represented space, and switches between coding properties to simultaneously explain the spiking activity from multiple units and the rhythmic content of LFPs recorded from multiple sources. The model is temporally causal, meaning that estimates of the switching state can be made at each instant using only past information from the spikes and LFPs, or can be combined with future data to refine those estimates. We applied this model framework to simulated and real hippocampal data to demonstrate its performance in identifying ripple-replay events.

Objective: Dynamic causal influences between brain regions are crucial for understanding the temporal variation and fluctuation of the interaction in human brain. However, recent causal discovery approaches often focus on fixed causality under directed acyclic graph constraints, and do not infer the dynamic and fluctuating nature of causality, which commonly exists in the brain.

Methods: We propose a causality learning framework with evolving distribution for non-stationary and non-linear systems. Based on this framework, a time-reversal enhanced dynamic causality distribution learning (TRDCDL) model is constructed, which integrates spatio-temporal information to identify evolving distributional sparse interactions in data.

Results: TRDCDL is validated in two synthetic models, which show the accuracy in learning both linear and non-linear causality within synthetic data. We further apply TRDCDL to the Alzheimer's Disease Neuroimaging Initiative dataset and infer dynamic effective connectivity networks (dECNs) among two stages of mild cognitive impairment (MCI).

Conclusion: The results reveal significant differences in dECNs between brain regions across the these stages, indicating that dECNs can serve as reliable neuromarkers for distinguishing different stages of MCI.

Significance: Significant reductions in dynamic causal influences within the default mode network and bilateral limbic network, along with few increased connectivity, reflect neurodegeneration and changing patterns of dECNs as MCI progresses.

Objective: Enhancing the reliability of myoelectric controllers that decode motor intent is a pressing challenge in the field of bionic prosthetics. State-of-the-art research has mostly focused on Supervised Learning (SL) techniques to tackle this problem. However, obtaining high-quality labeled data that accurately represents muscle activity during daily usage remains difficult. We investigate the potential of Reinforcement Learning (RL) to further improve the decoding of human motion intent by incorporating usage-based data.

Methods: The starting point of our method is a SL control policy, pretrained on a static recording of electromyographic (EMG) ground truth data. We then apply RL to fine-tune the pretrained classifier with dynamic EMG data obtained during interaction with a game environment developed for this work. We conducted real-time experiments to evaluate our approach and achieved significant improvements in human-in-the-loop performance.

Results: The method effectively predicts simultaneous finger movements, leading to a two-fold increase in decoding accuracy during gameplay and a 39% improvement in a separate motion test.

Conclusion: By employing RL and incorporating usage-based EMG data during fine-tuning, our method achieves significant improvements in accuracy and robustness.

Significance: These results showcase the potential of RL for enhancing the reliability of myoelectric controllers, which is of particular importance for advanced bionic limbs.

This study aims to enhance brain-computer interface (BCI) applications for individuals with motor impairments by comparing the effectiveness of noninvasive tripolar concentric ring electrode electroencephalography (tEEG) with conventional electroencephalography (EEG) technology. The goal is to determine which EEG technology is more effective in measuring and decoding different grasp-related neural signals. The approach involves experimenting on ten healthy participants who performed two distinct reach-and-grasp movements: power grasp and precision grasp, with a no-movement condition serving as the baseline. Our research compares EEG and tEEG in decoding grasping movements, focusing on signal-to-noise ratio (SNR), spatial resolution, and wavelet time-frequency analysis. Additionally, our study involved extracting and analyzing statistical features from the wavelet coefficients, and both binary and multiclass classification methods were employed. Four machine learning algorithms-Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), and Linear Discriminant Analysis (LDA)-were used to evaluate the decoding accuracies. Our results indicated that tEEG demonstrated higher quality performance compared to conventional EEG in various aspects. This included a higher signal-to-noise ratio and improved spatial resolution. Furthermore, wavelet time-frequency analyses validated these findings, with tEEG exhibiting increased power spectra, thus providing a more detailed and informative representation of neural dynamics. The use of tEEG led to significant improvements in decoding accuracy for differentiating grasp movement types. Specifically, tEEG achieved around 90.00% accuracy in binary and 75.97% for multiclass classification. These results exceed those from conventional EEG, which recorded a maximum of 77.85% and 61.27% in similar tasks, respectively.

Objective: To develop a transceiver radio frequency (RF) coil optimized for high resolution small-animal imaging at 14.1 T, aimed at enhancing signal-to-noise ratio (SNR) performance.

Methods: A hybrid distributed capacitance (HDC) birdcage coil was designed, combining conventional endring lumped capacitors with distributed capacitance along the legs, implemented using double-layer copper-clad substrates. Electromagnetic (EM) simulations were employed to optimize the coil's structural parameters and capacitance values for maximum RF performance. The HDC birdcage coil's performance was evaluated against a conventional bandpass (BP) design through electromagnetic simulations, bench tests, and phantom imaging. In vivo validation was performed using mouse imaging.

Results: EM simulations demonstrated that the HDC design enhances mean $text{B}_{1}^{+}$ and $text{B}_{1}^{-}$ field strengths by 11.8% and 11.7%, respectively, relative to the conventional BP design. The HDC design also showed reduced electric field (E-field) value in phantom, with 4.2% lower mean and 11.4% lower maximum E-field value. Bench measurements revealed a superior quality factor (Q factor) for the HDC coil, with a 34.2% higher unloaded Q value compared to the conventional design. Phantom imaging confirmed a 41% SNR improvement with the HDC design. The optimized HDC coil enabled mouse brain imaging at 50 $ !!mu !!text{ m}$ resolution.

Conclusion: The proposed HDC birdcage coil demonstrated superior receiver sensitivity and Q factor compared to conventional designs, yielding significant SNR improvements in 14.1 T imaging.

Significance: The results demonstrated the feasibility of achieving enhanced coil performance through HDC design at ultra-high field strength, providing a promising approach for improving image quality in small-animal MRI applications.

Objective: Accurate, non-invasive methods for estimating joint and muscle physiological states have the potential to greatly enhance control of wearable devices during real-world ambulation. Traditional modeling approaches and current estimation methods used to predict muscle dynamics often rely on complex equipment or computationally intensive simulations and have difficulty estimating across a broad spectrum of tasks or subjects.

Methods: Our approach used deep learning (DL) models trained on kinematic inputs to estimate internal physiological states at the knee, including moment, power, velocity, and force. We assessed each model's performance against ground truth labels from both a commonly used, standard OpenSim musculoskeletal model without EMG (static optimization) and an EMG-informed method (CEINMS), across 28 different cyclic and noncyclic tasks.

Results: EMG provided no benefit for joint moment/power estimation (e.g., biological moment), but was critical for estimating muscle states. Models trained with EMG-informed labels but without EMG as an input to the DL system significantly outperformed models trained without EMG (e.g., 33.7% improvement for muscle moment estimation) (p < 0.05). Models that included EMG-informed labels and EMG as a model input demonstrated even higher performance (49.7% improvement for muscle moment estimation) (p < 0.05), but require the availability of EMG during model deployment, which may be impractical.

Conclusion/significance: While EMG information is not necessary for estimating joint level states, there is a clear benefit during muscle level state estimation. Our results demonstrate excellent tracking of these states with EMG included only during training, highlighting the practicality of real-time deployment of this approach.

Objective: To develop an effective method for phase correction of magnetic resonance spectroscopic imaging (MRSI) data.

Methods: In many MRSI applications, it is desirable to generate absorption-mode spectra, which requires correction of phase errors in the measured MRSI data. Conventional phase correction methods are sensitive to measurement noise and baseline distortion, often resulting in distorted absorption-mode spectra from MRSI data with low-SNR and long acquisition dead time. This paper proposed a novel model-based method for improved phase correction of MRSI data. The proposed method determined the zeroth-order phase and acquisition dead time using a Lorentzian-based spectral model and performed signal extrapolation using a generalized series model. Absorption-mode spectra were then generated from the phase-corrected and extrapolated MRSI data.

Results: The proposed method was evaluated using both simulated data and experimental data acquired from human subjects in multi-nuclei (³¹P, ²H, and ¹H) MRSI experiments. Simulation results demonstrated improved parameter estimation accuracy by the proposed method under various noise levels and dead times. The proposed method also consistently generated high-quality absorption-mode spectra with minimal spectral distortions from experimental data. The proposed method was compared with state-of-the-art methods (including the entropy method and LCModel method) and showed more robust phase correction performance with less spectral distortions.

Conclusion: This paper introduced a novel method for phase correction of MRSI data. Results from simulated and in vivo data demonstrated that high-quality absorption-mode spectra could be obtained using the proposed method.

Significance: This method will provide a useful tool for processing MRSI data.