Physiological measurement最新文献

Objective.Early diagnosis of Chagas disease plays a vital role in enabling timely treatment and reducing the likelihood of underlying severe cardiovascular complications. Electrocardiogram (ECG) signals contain vital information that reflects cardiac disease progression, motivating the use of advanced signal processing and machine intelligence for accurate diagnosis of Chagas disease. This work presents an automated system for Chagas disease detection using conventional 12-lead ECG recordings.Approach.The method begins with preprocessing by standardizing the ECG sampling frequency and detecting QRS complexes. Then, four categories of features are extracted: (a) wavelet scattering transform coefficients from limb lead II and chest leads V1 and V3; (b) statistical descriptors from heart rate variability; (c) statistical features across all leads; and (d) patient metadata. The computed diagnostic feature vector is used as an input to the random under-sampling boosting classifier for its ability to handle class imbalance for binary classification between Chagas and non-Chagas cases.Main results.The proposed framework was evaluated on the PhysioNet/Computing in Cardiology (CinC) Challenge 2025 dataset. Evaluation on the hidden test data set yielded an accuracy of 90.53%,F1 Chagas = 10.73%, AUROC = 63.67%, AUPRC = 11.96%and Challenge Score = 20.5%under the team name Medics.Significance.The findings of this work highlight the potential of signal processing and machine learning-based analysis of ECG for scalable, non-invasive, and cost-effective early detection of Chagas disease, supporting improved clinical decision-making and preventive healthcare strategies.

Objective.We investigate the application of federated learning (FL) across heterogeneous, non-independent and identically distributed (non-IID) sleep data. We evaluate three algorithms-federated stochastic gradient descent, federated averaging, and federated proximal (FedProx)-in a realistic setting where non-IID characteristics arise from distinct sensor configurations, varying acquisition protocols, and diverse patient populations across independent sleep cohort datasets.Approach.We employ a dual-layered evaluation framework. First, we systematically analyze the impact of local training epochs (E={1,30}) and aggregation schemes (weightedandunweighted) on model convergence. Second, we introduce and adapt a generalized sub-sampling strategy designed to mitigate model drift caused by heterogeneous data distribution and volume imbalances across participating clients. To ensure robust external generalization, our evaluation utilizes six independent databases in a leave-one-database-out cross-validation scheme.Main results.Our analysis has evidenced that increasing the number of local training epochs adversely affects performance across all evaluated federated schemes. This confirms that extended local training exacerbates client drift, hindering global convergence. Furthermore,weightedaggregation consistently under-performsunweightedapproaches, suggesting that disproportionate client contributions bias the global data representation. While the inclusion of a proximal term partially mitigates this instability by constraining local updates, the proposedsub-samplingstrategy proves most effective. This approach yields consistent generalization results across all algorithms and minimizes performance downgrading, while significantly reducing computational overhead.Significance.This work addresses critical privacy concerns in centralized automated sleep staging by validating FL in realistic, multi-center scenarios. We provide evidence that decentralized strategies can achieve performance comparable to centralized methods, effectively overcoming data silos. Ultimately, this approach enables robust collaborative training while strictly maintaining data privacy-a fundamental requirement for widespread clinical implementation.

Objective.The standard 12-lead electrocardiogram (ECG) remains essential for cardiac diagnosis but requires ten physical electrodes, limiting long-term and wearable monitoring applications. We developed an anatomically grounded and physiologically interpretable framework to reconstruct the complete 12-lead ECG from four synthetic chest-torso electrodes derived using geometric vector principles and cardiac territorial anatomy.Approach.The 12 standard leads were partitioned into four physiologically coherent clusters representing septal/anterior, apical-lateral, inferior, and high-lateral depolarization vectors. Synthetic electrodes were constructed as weighted linear combinations of standard leads guided by frontal- and horizontal-plane vector geometry. A hybrid convolutional neural network-Transformer architecture mapped these four synthetic inputs to full 12-lead waveforms. The model was trained on 21 786 recordings from the PTB-XL dataset and externally validated on 500 recordings from the Chapman-Shaoxing dataset. Performance was evaluated using coefficient of determination (R²), Pearson correlation, root mean square error (RMSE), diagnostic concordance analysis, ablation testing, and noise robustness assessment.Main results.On the internal test set, the model achieved meanR²= 0.878 ± 0.070, Pearson correlationρ= 0.939 ± 0.030, and RMSE = 0.071 ± 0.030 mV. External validation demonstrated only 5% performance degradation. Waveform component preservation exceeded 94%, ST-segment correlation reached 0.964, and overall diagnostic concordance was 0.883, indicating preservation of approximately 88% of clinically relevant information. Reconstruction errors were symmetrically distributed around zero with minimal bias (0.001 mV) and maintained robustness at signal-to-noise ratios ⩾ 10 dB.Significance.This anatomically explainable reconstruction framework demonstrates the algorithmic feasibility of compact four-electrode ECG systems while preserving high diagnostic fidelity. By grounding electrode design in cardiac vector anatomy and validating performance across datasets, the approach provides a physiologically interpretable foundation for future wearable and ambulatory ECG reconstruction systems, establishing a reconstruction ceiling prior to hardware implementation.

Objective. Self-supervised learning (SSL) has enabled vision transformers (ViTs) to learn robust representations from large-scale natural image datasets, enhancing their generalization across domains. In retinal imaging, foundation models pretrained on either natural or ophthalmic data have shown promise, but the benefits of in-domain pretraining remain uncertain.Approach. To investigate this, we benchmark six SSL-pretrained ViTs on seven digital fundus image (DFI) datasets totaling 70 000 expert-annotated images for the task of moderate-to-late age-related macular degeneration (AMD) identification.Main results. Our results show that DINOv2, pretrained on natural images, shows similar performance than domain-specific models. These findings highlight the value of foundation models in improving AMD identification, and challenge the assumption that in-domain pretraining is necessary.Significance. We present our model AMDNet, which performs state-of-the-art out-of-domain AUROCs on six public datasets. Furthermore, we release BRAMD, an open-access dataset (n = 587) of DFIs with AMD labels from Brazil. Project page:www.aimlab-technion.com/lirot-ai.

Objective.In mechanically ventilated patients, inhomogeneity of air volume distribution in the lungs can lead to lung collapse and overdistention, increasing the risk of ventilator-induced lung injury. This study aims to estimate the degree of lung collapse (DoLC) from electrical impedance tomography (EIT) images without relying on lung segmentation.Approach.Traditional DoLC assessment based on the global inhomogeneity index is limited by lung segmentation. To address this limitation, a deep learning framework is proposed to directly estimate DoLC from EIT images. The model was trained on synthetic datasets simulating various lung conditions.Main results.The proposed method achieved high accuracy, with errors within 0%-5% in numerical and phantom tests across heterogeneous simulated lung conditions.Significance.The proposed framework enables segmentation-free estimation of DoLC from EIT images.

Objective.To develop a comprehensive physiological dataset for assessing internal and external stress and to propose robust automated stress recognition methods based on photoplethysmographic (PPG) signals.Approach.We established the Internal and External Stress Dataset (IESD), comprising PPG signals from 107 participants subjected to four distinct stress-inducing paradigms. Exploratory analyses revealed significant differences in heart rate variability (HRV) across these paradigms, underscoring the necessity for advanced methods capable of differentiating various stress types. To address this, we introduced a transfer learning-based inter-paradigm stress recognition model utilizing a domain adversarial neural network combined with maximum mean discrepancy for robust feature extraction.Main results.Analysis identified significant differences between internal and external stress, as well as among different external paradigms. Our proposed model demonstrated superior accuracy in recognizing homologous stress compared to heterologous stress within the same target domain, achieving accuracies of 73.86% (TSST to ST) and 60.41% (TSST to VWT). Moreover, the deep feature extraction significantly improved recognition performance and robustness across both intra- and inter-paradigm contexts.Significance.This study provides a valuable dataset and advanced methodology to enhance automated stress detection capabilities, effectively differentiating internal and external stress. The application of deep learning significantly improves recognition accuracy, offering promising prospects for future research and practical applications in stress monitoring.

Objective: Electrical impedance tomography (EIT) is a powerful imaging technique for assessing regional ventilation, but its analysis remains challenging due to the diversity of input formats, acquisition protocols, and research objectives. This work aims to simplify and standardize EIT data analysis through the development of a modular, user-friendly software platform. Approach: We developed the Parametric EIT Analysis Software (PEAS), a modular platform for EIT data analysis based on configurable, template-driven workflows. The software supports both raw voltage data with integrated image reconstruction and pre-reconstructed images, provides temporal detectors for breathing cycles and maneuvers, and reusable analysis components. These functionalities are accessed through a graphical user interface that enables interactive workflow configuration and execution. Main results: The implemented framework supports multiple vendor-specific data formats, including both raw voltage recordings and reconstructed image data. It provides automated detection of breathing cycles and respiratory maneuvers, as well as over 40 generic building blocks that can be combined into customized analysis pipelines. Typical workflows execute within seconds on standard hardware, enabling interactive use. A questionnaire-based user study indicated that the software is easy to learn and operate. Significance: By providing a standardized, extensible, and user-friendly environment for EIT data analysis, PEAS lowers the technical barrier to applying EIT in both research and clinical practice. This platform supports reproducibility, interoperability, and wider adoption of EIT for physiological monitoring and diagnostic applications. By offering a standardized yet extensible environment for EIT data analysis, PEAS reduces technical barriers in both research and clinical contexts. The platform promotes reproducibility, interoperability, and broader adoption of EIT for physiological monitoring and diagnostic applications.

Objective. Numerous studies examine the link between health and sleep-wake patterns to understand etiology, establish preventive algorithms, or develop therapeutics. The use of actigraphy to measure physical activity (PA) and sleep is increasing, partly because of its non-invasive nature and its ability to continuously monitor PA and sleep in free-living settings. There are several actigraphy data cleaning and pre-processing methods, but there is no consensus on how to define PA metrics or standardized cleaning procedures to enable comparison across research studies. This scoping review examined existing literature on cleaning and pre-processing of actigraphy data.Approach.The PubMed (US National Library of Medicine), Scopus (Elsevier), and Web of Science: Core Collection (Clarivate Analytics) databases were searched for original studies published in English from 2017-2024. Using Covidence, two reviewers independently screened each article and collected data.Results.A total of 102 studies were included for the final analysis. Our results showed substantial heterogeneity in actigraphy devices, data cleaning and pre-processing methods, with some studies using their own algorithmic approaches to generate PA and sleep variables. While some studies used well-established algorithms like Freedson or Cole-Kripke, a large proportion either developed custom methods or did not report sufficient detail to allow replication. This variability highlights the urgent need for standardized reporting and consensus-based protocols in actigraphy data cleaning and pre-processing to allow replication and comparison of findings across studies.Significance.This scoping review is the first to differentiate, in a standardized way, betweencleaningandpre-processingpractices in actigraphy research and to quantify reporting practices across multiple device types and data processing strategies. Our findings show a critical gap in standardized reporting and offer actionable guidance for both high- and low-resource research settings.

Objective.Time-frequency (TF) analysis is used to identify oscillatory patterns in complex signals. Cardiac signals under stress conditions are highly dynamic, yet heart rate variability (HRV) is often analysed using classical methods that assume stationarity or linearity. This study applied TF analyses to beat-to-beat RR time-series data extracted from electrocardiograms of 30 healthy adults during three stress tasks: mental calculation, noise exposure, and cold pressor test.Approach.Continuous wavelet transform (CWT), and ensemble empirical mode decomposition (EEMD) were compared to the standard short-term Fourier transform (STFT). Signals were divided into anticipation, stress, and recovery periods.Main results.When analysed in 30 s windows, all three methods detected dynamic time variations in standard frequency bands (low-frequency (LF) [0.04-0.15 Hz], high-frequency (HF) [0.15-0.40 Hz]) during stress compared to baseline. Compared to SFFT, EEMD and CWT showed greater sensitivity than STFT to identify LF and HF differences. Spectrograms identified regions of interest outside standard frequency bands, where CWT provided superior temporal and frequency resolution, especially at low frequencies. While EEMD spectrograms were uninterpretable, analysis of individual EEMD modes enabled tracking instantaneous changes in both frequency and amplitude.Significance.In conclusion, CWT and EEMD proved most valuable for identifying patterns in stress-evoked HRV and providing information on autonomic nervous system activation latency, responsiveness, and adaptability.

Objective.The objective of this study was to assess the feasibility and accuracy of computed tomography (CT)-guided electrical impedance tomography (CT-guided EIT) in the quantitative detection of the spatial distribution and location of pneumothorax, hemothorax, and hemopneumothorax lesions and effective ventilation regions in pig models.Approach. Five Bama miniature pigs were used to establish models of pneumothorax, hemothorax, and hemopneumothorax by incrementally injecting air or Ringer's solution in 100 ml steps up to a total volume of 500 ml into the right pleural cavity. Synchronous EIT data and CT images were acquired at each experimental stage. EIT images were reconstructed using the GREIT algorithm with anatomical constraints derived from CT-based lung contours. Mean total boundary voltage (mTBV), pneumothorax pixel area (PPA), hemothorax pixel area (HPA), center of ventilation (CoV), Dice similarity coefficient (Dice), and centroid distance (d_c) were used for quantitative assessment. PPA, HPA, and CoV are statistically compared between EIT and CT using Spearman correlation and Bland-Altman agreement analysis.Main results.mTBV showed a strong linear correlation with injected air volume (R²= 0.968-0.994) and fluid volume (R²= 0.712-0.994). In pneumothorax models, Dice = 0.828-0.884 andd_c= 2.80-3.33. In hemothorax models, Dice = 0.850-0.874 andd_c= 2.64-3.34. PPA, HPA, and CoV derived from CT-guided EIT images correlated significantly with CT findings (Spearmanr= 0.63-0.92,p< 0.001). Ventilation distribution patterns in EIT were consistent with CT, with dorsal shifts during pneumothorax and ventral shifts during hemothorax. Bland-Altman plots showed good agreement between EIT and CT measurements.Significance.This study demonstrates the feasibility of CT-guided EIT for dynamic monitoring and quantitative evaluation of pneumothorax, hemothorax, and hemopneumothorax in pig models. Its noninvasive, radiation-free, and bedside monitoring nature makes it a promising tool for detecting pulmonary pathological accumulation during mechanical ventilation and postoperative care.