IEEE Transactions on Biomedical Engineering最新文献

Objective: The Pressure Reactivity Index (PRx) is a common metric for assessing cerebral autoregulation in neurocritical care. This study aimed to enhance the clinical utility of PRx by developing a personalized PRx algorithm (pPRx) and identifying ideal hyperparameters.

Methods: Algorithmic errors were quantified using simulated data and multimodal monitoring data from traumatic brain injury patients from the Track-TBI dataset. Using linear regression, heart rate was identified as a potential cause of PRx error. The pPRx method was developed by reparameterizing PRx averaging to heartbeats. Ideal hyperparameters for the standard PRx algorithm were identified that minimized algorithmic errors.

Results: PRx was sensitive to hyperparameters and patient variability. Errors were related to patient heart rates. By parameterizing PRx to heartbeats, the pPRx methodology significantly reduced noise and sensitivity to both patient variability and hyperparameter selection. In the standard PRx algorithm, averaging windows of 10 seconds and correlation windows of 40 samples resulted in the lowest overall error.

Conclusion: Personalized PRx enhances the robustness and accuracy of cerebral autoregulation estimation by addressing patient- and hyperparameter-sensitivity. This improvement is crucial for reliable clinical decision-making in neurocritical care.

Significance: Robust estimation of cerebral autoregulation would be beneficial for identifying precision medicine targets and improving outcomes for neurocritical care patients. We systematically increased the robustness of PRx to make it more consistent across patient populations.

Human body communication, particularly of the Electro-Quasistatic variety, has gained traction among low-power wireless circuit designers due to its benefits in terms of power and physical security compared to conventional electromagnetic or radio frequency communication. Yet, applying its theory and practice has thus far been limited to a single person, without a clear understanding of how the channel behaves when multiple individuals or human bodies are added. To the author's knowledge, this work analyzes the limit of the quasistatic approximation in a multiple human body scenario for the first time. It demonstrates how the approximation changes with the length scale. Furthermore, the channel gain for varying numbers of human bodies is measured to be -35 dB (one human body), -41 dB (two humans), and -44 dB (three humans) for a ground-connected transmitter. A complete bio-physical circuit model is generated to accurately predict the voltage received on the quasi-static structure depending on how many human bodies are connected serially. The impact of the work can aid designers in realizing applications in secure key exchange, authentication, and music sharing using electro-quasistatic human body communication techniques.

The frequency dependence of backscattered radiofrequency (RF) signals produced by ultrasound scanners carries rich information related to the tissue microstructure (i.e., scatterer size, attenuation). This information can be sue to classify tissues based on microstructural changes associated to disease onset and progression. Conventional convolutional neural networks (CNNs) can learn this information directly from radio-frequency (RF) data, but they often struggle to achieve adequate frequency selectivity. This increases model complexity and convergence time, and limits generalization. To overcome these challenges, SincNet, originally developed for speech processing, was adapted to classify RF data based on differences in frequency properties. Rather than learning every filter coefficient, SincNet only learns each filter's low frequency and bandwidth, dramatically reducing the number of parameters and improving frequency resolution. For model interpretability, a Gradient-Weighted Filter Contribution is introduced, which highlights the importance of spectral bands. The approach was validated on three datasets: simulated data with different scatterer sizes, experimental phantom data, and in vivo data of rats which were fed a methionine and choline- deficient diet to develop liver steatosis, inflammation, and fibrosis. The modified SincNet consistently achieved the best results in material/tissue classifications.

Motor imitation impairments are commonly reported in individuals with autism spectrum conditions (ASCs), suggesting that motor imitation could be used as a phenotype for addressing autism heterogeneity. Traditional methods for assessing motor imitation are subjective and labor-intensive, and require extensive human training. Modern Computerized Assessment of Motor Imitation (CAMI) methods, such as CAMI-3D for motion capture data and CAMI-2D for video data, are less subjective. However, they rely on labor-intensive data normalization and cleaning techniques, and human annotations for algorithm training. To address these challenges, we propose CAMI-2DNet, a scalable and interpretable deep learning-based approach to motor imitation assessment in video data, which eliminates the need for ad hoc normalization, cleaning and annotation. CAMI-2DNet uses an encoder-decoder architecture to map a video to a motion representation that is disentangled from nuisance factors such as body shape and camera views. To learn a disentangled representation, we employ synthetic data generated by motion retargeting of virtual characters through the reshuffling of motion, body shape, and camera views, as well as real participant data. To automatically assess how well an individual imitates an actor, we compute a similarity score between their motion encodings, and use it to discriminate individuals with ASCs from neurotypical (NT) individuals. Our comparative analysis demonstrates that CAMI-2DNet has a strong correlation with human scores while outperforming CAMI-2D in discriminating ASC vs NT children. Moreover, CAMI-2DNet performs comparably to CAMI-3D while offering greater practicality by operating directly on video data and without the need for ad hoc normalization and human annotations.

Objective: The lack of high-quality annotated images and the limited transferability of task-specific models hamper the practical of AI-assisted diagnosis for brain diseases. Developing self-supervised foundation model is a promising solution to address this problem.

Methods: We establish a masked image modeling (MIM)-based foundation model of brain disease (BDFM). We construct a database named BD-15k of more than ten brain diseases for pre-training. To improve the lesion feature extraction ability of BDFM, we propose a spatial-frequency dual-domain decoder, and introduce a spatial mean masking strategy to replace the traditional masking methods.

Results: Results indicate that these improvements help BDFM outperform the baseline method in reconstructing lesion details. Extensive qualitative and quantitative experiments on three downstream tasks show that BDFM generalizes well to segmentation and classification tasks based on small annotated datasets.

Conclusion: BDFM outperforms taskspecific models trained from scratch while avoiding complex task-specific designs.

Significance: This work contributes to the advancement of medical foundation models, paving the way for more effective brain disease analysis. The source code will be made publicly available upon publication in https://github.com/zzzjjj98/BDFM.

Lateral flow immunoassays (LFIAs) provide rapid point-of-care results but lack quantitative capabilities. This study presents a platform technology integrating full-spectrum analysis (400-700 nm) with machine learning to enhance qualitative LFIAs with semi-quantitative assessment capabilities. We analyzed 241 clinical nasopharyngeal specimens using portable spectrometry to capture gold nanoparticle optical signatures from SARS-CoV-2 rapid tests as a validation model. Systematic evaluation of normalization strategies revealed T-C differential outperformed T/C ratio normalization. Signal processing through Savitzky-Golay filtering, standard normal variate transformation, and principal component analysis reduced dimensionality from 601 to 4 features while retaining 97.26% variance. Among five evaluated algorithms, random forest achieved optimal performance (R² = 0.961, RMSE = 2.235 Ct) across clinically relevant ranges (PCR Ct 10.8-35.0). Bland-Altman analysis revealed measurement uncertainty of ±4.2 Ct, indicating suitability for population surveillance rather than precise individual quantification. Feature importance analysis identified 520-570 nm as the critical spectral region, consistent with gold nanoparticle surface plasmon resonance. This platform approach demonstrates that standard LFIAs contain extractable semi-quantitative information accessible through spectral-machine learning integration. While validated using COVID-19, the framework's modular design enables adaptation to diverse analytes including biomarkers, therapeutic drugs, and environmental contaminants without fundamental architectural changes. The methodology establishes a foundation for enhanced lateral flow diagnostics, particularly valuable in resource-limited settings where rapid semi-quantitative results provide greater utility than delayed laboratory measurements.

Objective: This study aims to support early diagnosis of Alzheimer's disease and detection of amyloid accumulation by leveraging the microstructural information available in multi-shell diffusion MRI (dMRI) data, using a vision transformer-based deep learning framework.

Methods: We present a classification pipeline that employs the Swin Transformer, a hierarchical vision transformer model, on multi-shell dMRI data for the classification of Alzheimer's disease and amyloid presence. Key metrics from DTI and NODDI were extracted and projected onto 2D planes to enable transfer learning with ImageNet-pretrained models. To efficiently adapt the transformer to limited labeled neuroimaging data, we integrated Low-Rank Adaptation. We assessed the framework on diagnostic group prediction (cognitively normal, mild cognitive impairment, Alzheimer's disease dementia) and amyloid status classification.

Results: The framework achieved competitive classification results within the scope of multi-shell dMRI-based features, with the best balanced accuracy of 95.2% for distinguishing cognitively normal individuals from those with Alzheimer's disease dementia using NODDI metrics. For amyloid detection, it reached 77.2% balanced accuracy in distinguishing amyloid-positive mild cognitive impairment/Alzheimer's disease dementia subjects from amyloid-negative cognitively normal subjects, and 67.9% for identifying amyloidpositive individuals among cognitively normal subjects. Grad-CAM-based explainability analysis identified clinically relevant brain regions, including the parahippocampal gyrus and hippocampus, as key contributors to model predictions.

Conclusion/significance: This study demonstrates the promise of diffusion MRI and transformer-based architectures for early detection of Alzheimer's disease and amyloid pathology, supporting biomarker-driven diagnostics in data-limited biomedical settings.

Objective: Ultra-Low-Field Magnetic Resonance Imaging (ULF MRI) offers low cost and portability but suffers from electromagnetic interference (EMI) in unshielded environments. This study developed a deep learning-based active EMI suppression method to overcome these limitations.

Methods: Using a 68mT ULF MRI system, human body coupling was identified as a primary EMI pathway. We proposed EMIC-Net, a U-Net architecture incorporating Transformer and hybrid attention mechanisms, to learn the data-driven nonlinear mapping from sensing coil signals to radio-frequency (RF) receiver coil interference. Acquired data underwent phase and gain compensation prior to model training. The model's efficacy was validated through in vivo human brain imaging, comparing its performance with EDITER and standard CNN methods, and by assessing the impact of varying EMI coil numbers and training data volumes.

Results: EMIC-Net effectively suppressed complex dynamic EMI. It restored image SNR from 2.35 dB to 17.63 dB, with significant PSNR and SSIM improvements. Image quality neared shielded acquisitions and surpassed comparative methods. Three EMI coils provided optimal balance, and the model showed data efficiency, requiring a small dataset for effective training.

Conclusion: The EMIC-Net method accurately predicts and efficiently removes EMI in unshielded ULF MRI, offering superior performance and practicality.

Significance: This research promotes portable, low-cost ULF MRI for primary healthcare and bedside diagnosis. It also offers insights for mitigating complex EMI issues in other RF sensing domains.

Objective: Accurately predicting glucose levels is essential for effectively managing type 1 diabetes (T1D), a chronic condition in which the body cannot produce insulin. Although deep learning approaches have shown promise, their training requires extensive datasets that capture a wide range of physiological and behavioral variations. However, obtaining such datasets can be challenging and impractical, especially when their collection demands significant patient effort. To overcome this limitation, we propose a data augmentation strategy that leverages digital twins of individuals with T1D (DT-T1D) to generate personalized synthetic data mirroring real-world glucose-insulin dynamics.

Methods: ReplayBG, an open-source tool for creating DT-T1D, was adapted to develop a two-steps strategy: first, generating DT-T1D from retrospective patient data; then, using DT-T1D with new inputs, to simulate synthetic, patient-specific data. The practical impact of this approach is demonstrated in a case study where personalized deep networks were developed to predict glucose levels. Models were trained on an open-source dataset from 12 patients, using either the original data or a combination of the original and synthetic data.

Results: Integrating synthetic data into the training process consistently enhances model performance. Moreover, models trained on synthetic data combined with only a small fraction of the original dataset achieve results comparable to those obtained from the full, unaugmented dataset.

Conclusion: Leveraging DT-T1D to generate personalized synthetic data mitigates data scarcity and enhances deep learning model performance for accurate glucose prediction.

Significance: This work highlights the potential of digital twin-driven data augmentation to tackle data scarcity and develop robust, personalized predictive models for T1D management.

Objective: The biomechanical properties of the lumbar spine is crucial for assisting the diagnosis, treatment, and prevention of spinal diseases. Traditional biomechanical analysis methods, especially the finite element analysis, require extensive computational resources, precise material property definitions, and complex meshing processes to accurately model the biomechanical behavior of the lumbar spine. While deep learning is introduced to enhance efficiency and accuracy, challenges like data dependency and lack of physical consistency remain.

Methods: We propose a novel framework that consists of a 3D generative adversarial network for data augmentation together with a dual-channel vision transformer to extract geometric and physical information. We also introduce a physics-guided mechanism into training phase, ensuring model consistency with mechanical principles.

Results: The proposed method achieved an Intersection over Union of 0.8332 and a Mean Squared Error of 0.0002. The five vertebrae of the lumbar spine are processed in 87 milliseconds, which is approximately 3000 times faster than traditional finite element methods.

Conclusion: Our framework demonstrates high accuracy and substantial computational efficiency, offering a reliable alternative to conventional biomechanical modeling.

Significance: This enables real-time lumbar spine analysis for diagnosis, surgical planning, and personalized treatment.