Physiological Measurement

BackgroundCardiorespiratory deterioration due to sepsis is a leading cause of morbidity and mortality for extremely premature infants with very low birth weight (VLBW, birthweight <1500g). Abnormal heart rate (HR) patterns precede the clinical diagnosis of late-onset sepsis in this population. Decades ago, clinicians recognized a pattern of reduced HR variability and increased HR decelerations in electrocardiogram tracings of septic preterm infants. A predictive logistic regression model was developed from this finding using mathematical algorithms that detect this signature of illness. Display of this model as the fold increase in risk of imminent sepsis reduced mortality in a large randomized trial. Here, we sought to determine if machine learning or deep learning approaches would identify this uncommon but distinctive signature of sepsis in VLBW infants. MethodsWe studied VLBW infants admitted from 2012 to 2021 to a regional Level IV NICU. We collected one-hour HR time series data from bedside monitoring sampled at 0.5 Hz (n=300 HR values per series) throughout the NICU admission. First, we applied the principles of highly comparative time series analysis (HCTSA) to generate many mathematical time series features and combined them in a machine learning model. Next, we used deep learning in the form of a convolutional neural network on the raw data to learn the HR features. The output was a set of HR records determined by HCTSA or deep learning to be at high risk for imminent sepsis. ResultsWe analyzed data from 566 infants with 61 episodes of sepsis. HCTSA and deep learning models predicted sepsis with high out-of-sample validation metrics. The riskiest records determined by both approaches demonstrated the previously identified HR signatures-reduced variability and increased decelerations. ConclusionsWe tested the ability of unguided machine learning approaches to detect the novel HR signature of sepsis in VLBW infants previously identified by human experts. Our main finding is that the computerized approach returned the same result - it identified heart rate characteristics of reduced variability and transient decelerations. We conclude that unguided machine learning can be as effective as human experts in identifying even a very rare phenotype in clinical data.

Pulse arrival time (PAT) is known to be correlated with blood pressure. Although PAT can be measured using electrocardiography (ECG), photoplethysmography (PPG), and other signals commonly available in clinical settings, recent literature has noted that devices recording these waveforms are often subject to many hardware-specific factors related to digital filtering, clock synchronization, temporal resolution, and latency. These factors can introduce relative timing errors between the ECG and PPG signals, resulting in a situation where traditional approaches for PAT measurement will not work as intended. In this work, we propose a methodology that accounts for these confounding factors and generates precise measurements of PAT using standard bedside monitoring equipment. This technique involves using heart rate variability to match heartbeats across waveforms and experimentally profiling the timing systems of bedside medical devices to correct various timing-related artifacts. To improve the precision of the resulting PAT measurements, we model temporal uncertainties stemming from the finite temporal resolution of the waveform samples. We apply this approach to a dataset with roughly 1.6 million hours of continuous ECG and PPG data from over 10,000 unique patients at a pediatric intensive care unit (ICU). After demonstrating that the observed timing artifacts are consistent across the entire dataset, we show that accounting for them results in more reasonable distributions of PAT measurements across age groups. It is our hope that this work will spur discussion around the standardization of PAT measurement using routinely collected signals in a clinical environment.

BackgroundElectronic Health Records(EHR) are very crucial for Clinical Decision Support Systems and for proper care to be delivered to ICU heart failure patients, there is often missing data due to monitoring device errors thus the need for robust imputation methodologies. ObjectiveTo compare and evaluate three different methodologies for imputing missing data for heart failure patients from the MIMIC-III database: Denoising Autoencoder (DAE), Self-Attention Imputation for Time Series (SAITS), and Multiple Imputation by Chained Equations (MICE) with LightGBM. MethodsAnalysis of 14,090 ICU admissions for patients with heart failure was performed using data from the MIMIC-III database. Features were selected based off of clinical relevance, and 19 clinical features were selected through a combination of Random Forest analysis, correlation analysis, and Mutual Information. The introduction of artificial missing values of 20%, 30%, and 50% was applied to the data set, and then 3 imputation methodologies were evaluated with the DAE, SAITS, and MICE+LightGBM. The performance of each imputation methodology was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Normalized Root Mean Square Error (NRMSE). ResultsBoth DAE and SAITS had superior performance on the imputation of missing values across all percentages of missing values. At 20% missingness, DAE had mean MAE = 0.004967, RMSE = 0.005217, and NRMSE = 3.260893 while SAITS had mean MAE = 0.005461, RMSE = 0.005797, and NRMSE = 3.244695; thus MICE+LightGBM resulted in a higher number of errors. At 50% missingness, the SAITS methodology demonstrated the best performance followed by DAE and MICE+LightGBM methods demonstrated decreased performance. The deep learning methodologies maintained a consistent level of accuracy between the clinical variables measured. ConclusionsOur analysis indicates that deep learning-based imputation methodologies significantly outperform traditional methodologies for imputing missing values in ICU heart failure data thus supporting the implementation of these methodologies into Clinical Decision Support Systems for heart failure patients.

We detect seizures in newborn infants using a novel method derived from triple correlation, which integrates spatial and temporal structure in neonatal electroencephalograms (EEGs). Triple correlation natively encompasses analogues to a variety of lower-order approaches (auto-correlation, cross-correlation) in addition to introducing higher-order signals, so we hypothesized that our approach would both effectively detect and differentiate notoriously difficult-to-detect and heterogeneous neonatal seizures. Indeed, our method in its simplest form performs comparably well to a current standard of care, amplitude-integrated EEG (aEEG), and by some measures outperforms aEEG, suggesting at a minimum that a combination of triple correlation and aEEG could produce a more effective first-line bedside detector. Moreover, we find that the triple correlation seizure-signal varies between patients, with 1) differences in dominance of either within or between channel correlations and 2) differing levels of higher order structure. We hope that our approach will provide a fertile field for future work in distinguishing and detecting seizures.

PurposeWhile bowel sound auscultation represents a key component of abdominal examination, its utility is limited because bowel sounds (BS) are intermittent, variable, and influenced by factors such as diet and digestive state. This renders it challenging to use them for a quantitative assessment of gastrointestinal health. MethodsBS signals were recorded from 84 subjects (39 patients and 45 healthy controls) using an acoustic SonicGuard sensor and categorized into four patterns. Metadata on physiological parameters were collected to examine their influence on BS characteristics and the differences between healthy and patient BS patterns. ResultsBowel sound patterns are significantly influenced by meal timing, caffeine consumption, and medication intake. Significant differences between healthy and patient groups were also observed in sound count, duration, energy, and waveform shape. These differences were mirrored in the performance of machine learning models finetuned for BS patterns classification, with performance depending on the group used for training and evaluation. ConclusionBS patterns present a promising quantitative indicators of gas-trointestinal health when analyzed alongside relevant physiological parameters.

In the early stages of atrial fibrillation (AF), most cases are paroxysmal (pAF), making identification only possible with continuous and prolonged monitoring. With the advent of wearables, smartwatches equipped with photoplethysmographic (PPG) sensors are an ideal approach for continuous monitoring of pAF. There have been numerous studies demonstrating successful capture of pAF events, especially using deep learning. However, deep learning requires a large amount of data and independent testing on diverse datasets, to ensure the generalizability of the model, and most prior studies did not meet these requirements. Moreover, most prior studies using wearable-based PPG sensor data collection were limited either to controlled environments, to minimize motion artifacts, or to short duration data collection. Most importantly, frequent premature atrial and ventricular contractions (PAC/PVC) can confound most AF detection algorithms. This has not been well studied, largely due to limited datasets containing these rhythms. Note that the recent deep learning models show 97% AF detection accuracy, and the sensitivity of the current state-of-the-art technique for PAC/PVC detection is only 75% on minimally motion artifact corrupted PPG data. Our study aims to address the above limitations using a recently completed NIH-funded Pulsewatch clinical trial which collected smartwatch PPG data over two weeks from 106 subjects. For our approach, we used multi-modal data which included 1D PPG, accelerometer, and heart rate data. We used a computationally efficient 1D bi-directional Gated Recurrent Unit (1D-Bi-GRU) deep learning model to detect three classes: normal sinus rhythm, AF, and PAC/PVC. Our proposed 1D-Bi-GRU models performance was compared with two other deep learning models that have reported some of the highest performance metrics, in prior work. For three-arrhythmia-classification, testing data for all deep learning models consisted of using independent data and subjects from the training data, and further evaluations were performed using two independent datasets that were not part of the training dataset. Our multimodal model achieved an unprecedented 83% sensitivity for PAC/PVC detection while maintaining a high accuracy of 97.31% for AF detection. Our model was computationally more efficient (14 times more efficient and 2.7 times faster) and outperformed the best state-of-the-art model by 20.81% for PAC/PVC sensitivity and 2.55% for AF accuracy. We also tested our models on two independent PPG datasets collected with a different smartwatch and a fingertip PPG sensor. Our three-arrhythmia-classification results show high macro-averaged area under the receiver operating characteristic curve values of 96.22%, and 94.17% for two independent datasets, demonstrating better generalizability of the proposed model.

Atrial Fibrillation (AFib) and Atrial Flutter (AFlut) are prevalent irregular heart rhythms that poses significant risks, particularly for the elderly. While automated detection systems show promise, misdiagnoses are common due to symptom similarities. This study investigates the differentiation of AFib from AFlut using standard 12-lead ECGs from the PhysioNet CinC Challenge 2021 (CinC2021) databases, along with data from a private database. We employed both one dimensional-based (1D) and image-based (2D) Deep Learning models, comparing different 1D and 2D Convolutional Neural Network (CNN) architectures for classification. For 1D models, LiteVGG-11 demonstrated the highest performed, achieving an accuracy (Acc) of 77.91 ({+/-}1.73%), area under the receiver operating characteristic curve (AUROC) of 87.17 ({+/-}1.29%), F1 score of 76.59 ({+/-}1.90%), specificity (Spe) of 71.69 ({+/-}4.73%), and sensitivity (Se) of 86.53 ({+/-}5.33%). On the other hand, for 2D models the EfficientNet-B2 outperformed other architectures, with an Acc of 75.20 ({+/-}3.38%), AUROC of 85.50 ({+/-}1.14%), F1 of 71.59 ({+/-}3.66%), Spe of 74.76 ({+/-}13.85%) and Se of 75.74 ({+/-}13.85%). Our findings indicate that distinguishing between AFib and AFlut is non-trivial, with 1D signals exhibiting superior performance compared to their 2D counterparts. Furthermore, its noteworthy that the performance of our models on the CinC2021 databases was considerably lower than on our private dataset.

Devices that record single-lead ECGs, such as smartwatches and handheld ECG recorders, hold promise for detecting undiagnosed atrial fibrillation (AF). Accurately extracting RR-intervals from telehealth ECGs is key for heart rhythm assessment. The aim of this study was to develop an algo-rithm to extract RR-intervals from telehealth ECGs, and assess whether the extracted RR-intervals are accurate and therefore suitable for analysis. Two datasets of 30-second handheld ECGs were used: TELE ECG Database (250 ECGs) and SAFER ECG dataset (507 ECGs). One of three high-performance primary QRS detectors, selected based on previous evidence, was used to detect QRS complexes and extract RR-intervals. These detec-tions were compared to those from a secondary QRS detector to assess accu-racy. All pairs of 3 primary and 18 secondary QRS detectors were tested. Ac-curacy was quantified using mean absolute error (MAE) and the proportion of time RR-intervals were assessed as accurate (coverage). Best performance was achieved using unsw and nk as primary and secondary detectors, with MAEs of 19.8ms and 16.3ms, and coverages of 89% on TELE and SAFER respectively. Using a single detector alone produced higher MAEs (23.8ms and 43.9ms on TELE; 38.2ms and 41.7ms on SAFER). Accuracy was similar between AF and non-AF, but reduced on low-quality signals (50.8 vs. 7.7ms, p<0.001). In conclusion, the recommended algorithm produced more accu-rate RR-intervals than using a single QRS detector, maintaining accuracy during AF, although accuracy was reduced on low-quality signals. HighlightsO_LIAlgorithm extracts RR-intervals from ECGs and assesses their accuracy C_LIO_LIAlgorithm was developed using two datasets of ECGs collected using different devices C_LIO_LIThe impacts of arrhythmia and noise on algorithm performance were assessed C_LIO_LIThe algorithm uses a pair of openly available QRS detection algorithms C_LI

BackgroundObtaining reliable rate heart estimates from waist based electrocardiograms (ECGs) poses a very challenging problem due to the presence of extreme motion artifacts. The literature reveals few, if any, attempts to apply motion artifact cancellation methods to waist based ECGs. This paper describes a new methodology for ameliorating the effects of motion artifacts in ECGs by specifically targeting ECG peaks for elimination that are determined to be correlated with accelerometer peaks. This peak space cancellation is applied to real world waist based ECGs. Algorithm SummaryThe methodology includes successive applications of a previously described pattern-based heart beat detection scheme (Temporal Pattern Search, or "TEPS"). In the first application, TEPS is applied to accelerometer signals recorded contemporaneously with ECG signals to identify high-quality accelerometer peak sequences (SA) indicative of quasi-periodic motion likely to impair identification of peaks in a corresponding ECG signal. The process then performs ECG peak detection and locates the closest in time ECG peak to each peak in an SA. The differences in time between ECG and SA peaks are clustered. If the number of elements in a cluster of peaks in an SA exceeds a threshold, the ECG peaks in that cluster are removed from further processing. After this peak removal process, further QRS detection proceeds according to TEPS. ExperimentThe above procedure was applied to data from real world experiments involving four sessions of walking and jogging on a dirt road for approximately 20-25 minutes. A compression shirt with textile electrodes served as the ground truth recording. A textile electrode based chest strap was worn around the waist to generate a single channel signal upon which to test peak space cancellation/TEPS. ResultsBoth walking and jogging heart rates were generally well tracked. In the four recordings, the percentage of segments within 10 beats/minute of reference was 96%, 99%, 92% and 96%. The percentage of segments within 5 beats/minute of reference was 86%, 90%, 82% and 78%. There was very good agreement between the RR intervals associated with the reference and waist recordings. For acceptable quality segments, the root mean square sum of successive RR interval differences (RMSSD) was calculated for both the reference and waist recordings. Next, the difference between waist and reference RMSSDs was calculated ({Delta}RMSSD). The mean {Delta}RMSSD (over acceptable segments) was 4.6 m, 5.2 ms, 5.2 ms and 6.6 ms for the four recordings. Given that only one waist ECG channel was available, and that the strap used for the waist recording was not tailored for that purpose, the proposed methodology shows promise for waist based sinus rhythm QRS detection.

ObjectiveThe widely used bandpass filter (BPF)-based algorithm for recovering sympathetic nerve activity (SNA) from the skin sympathetic nerve activity (SKNA-I) signal, recorded via electrocardiogram electrodes or subcutaneous sympathetic nerve activity (SCNA-I) in a lead I setup, has limitations. It excludes spectral information outside the BPF range and may retain artifacts, such as cardiac activity or pacemaker interference, in the recovered SNA (rSNA) signal. This study aims to develop an algorithm that recovers the full spectral SNA information as comprehensively as possible for evaluating the autonomic nervous system (ANS). MethodsWe propose a novel algorithm, S3 (SNA from Shrink and Subtraction), which integrates the optimal shrinkage algorithm (eOptShrink) with the template subtraction (TS) method. The performance of S3 was evaluated against other algorithms using semi-real simulated SKNA-I data, a human SKNA-I database including subjects with pacemakers or atrial fibrillation, and a mouse SCNA-I database. ResultsThe S3 algorithm demonstrated numerical efficiency and outperformed existing approaches, including traditional TS, BPF and other methods, in both time and frequency domains. Notably, in addition to the traditional 500-1000Hz spectral band, S3 effectively recovers spectral information across the 50-300Hz and 300-500Hz frequency bands, making it suitable for homecare ANS evaluation. All quantitative results are supported by the rSNA tracing for visual inspections. ConclusionS3 accurately recovers the full-spectrum SNA. SignificanceBy enabling the exploration of the entire SNA spectrum, S3 offers a promising tool for ANS evaluation and applications in homecare environments.

Efficient patient monitoring on the medical-surgical wards is crucial to prevent significant in-hospital adverse events. Standard episodic inpatient assessment of vital signs can potentially miss changes in health status and delay recognition of elevated risk. To reduce the likelihood of this delayed recognition of risk, we developed a wearable-based deep learning model, using only 9 inputs, to identify the onset of deterioration earlier than traditional early warning systems. We showed this model could generalize to produce clinical alerts ahead of a broad class of significant adverse clinical outcomes, including rapid response team (RRT) interventions, unplanned intensive care unit (ICU) transfers, intubations, cardiac arrests, and in-hospital deaths. Using data from 888 adult non-ICU inpatient visits in four hospitals in New York and employing two different clinical grade wearable biosensors (4-8% data missingness, excluding SpO2), as part of a quality initiative, we trained a recurrent neural network (RNN) to predict both MEWS alerts and adverse clinical outcomes. Using multiple stages of validation, we showed in our retrospective, time-sequence duration optimized, prospective validation the RNN model was able to predict both periods of elevated MEWS scores (ROC AUC 0.89 +/- 0.3, PR AUC 0.58 +/- 0.14) and adverse clinical outcomes (accuracy: 81.8% on 11 events) up to an average of 17 hours in advance. Our results show that our wearable based RNN alert system outperforms traditional episodic clinical support tools in detecting early onset of inpatient deterioration; enabling timely interventions that can improve outcomes and reduce hospital costs for patients in early stages of deterioration.

AO_SCPLOWBSTRACTC_SCPLOWThe electrocardiogram (ECG) is an almost universally accessible diagnostic tool for heart disease. An ECG is measured by using an electrocardiograph, and todays electrocardiographs use built-in software to interpret the ECGs automatically after they are recorded. However, these algorithms show limited performance, and therefore clinicians usually have to manually interpret the ECG, regardless of whether an algorithm has interpreted the ECG or not. Manual interpretation of the ECG can be time-consuming and require specific skills. Therefore, a better algorithm is clearly needed to make correct ECG interpretations more accessible and time efficient. Algorithms based on artificial intelligence have shown promising performance in many fields, including ECG interpretation, over the last few years and might represent an alternative to manual ECG interpretation. In this study, we used a dataset with 88253 12-lead ECGs from multiple databases, annotated with SNOMED-CT codes by medical experts. We employed a supervised convolutional neural network with an Inception architecture to classify 30 of the most frequent annotated diagnoses in the dataset. Each patient could have more than one diagnosis, which makes this a multi-label classification. We compared the Inception models performance while applying different preprocessing methods on the ECGs and different model settings during 10-folded cross-validation. We compared the models classification performance using binary cross-entropy (BCE) loss and double soft F1 loss. Furthermore, we compared the classification performance when downsampling the original sampling rate of the input ECG. Finally, we trained 30 interpretable linear models to provide class activation maps to explain the relative importance of each sample in the ECG with respect to the 30 diagnoses considered in this study. Due to the heavily imbalanced class distribution in our dataset, we placed the most emphasis on the F1 score when evaluating the performance of the models. Our results show that the best performance in terms of F1-score was seen when the Inception model used double soft F1 as the loss function and ECGs downsampled to 75Hz. This model achieved an F1 score of 0.420 {+/-} 0.017, accuracy = 0.954 {+/-} 0.002, and an AUROC score of 0.832 {+/-} 0.019. An aggregation of the generated saliency maps, achieved using Local Interpretable Model-Agnostic Explanations (LIME), showed that the Inception model paid the most attention to the limb leads and the augmented leads and less importance to the precordial leads. One of the more significant contributions that emerge from this study is the use of aggregated saliency maps to obtain ECG lead importance for different diagnoses. In addition, we emphasized the relevance of evaluating different loss functions, and in this specific case, we found double soft F1 loss to be slightly better than BCE. Finally, we found it somewhat surprising that downsampling the ECG led to higher performance compared to the original 500Hz sampling rate. These findings contribute in several ways to our understanding of the artificial intelligence-based interpretation of ECGs, but further studies should be carried out to validate these findings in other datasets from other patient cohorts.

Heart rate variability, or the variation in the time interval between consecutive beats, is a non-invasive dynamic metric of the autonomic nervous system and an independent risk factor for cardiovascular death. Prior limitations of use include requirements for continuous electrocardiography and lack of reference standards. Consumer wrist-worn tracking devices using photoplethysmography now provide the unique potential of continuously measuring surrogates of sympathetic and parasympathetic activity through the analysis of interbeat intervals. Here we leverage wrist-worn trackers to present the largest, to our knowledge, analysis of heart rate variability in humans across the time, frequency, and graphical domains. We derive diurnal parasympathetic and sympathetic measures and provide scaling parameters by age, sex, and time of day. Poincare plots graphically summarize heart rate variability metrics and may detect common arrhythmias. Lastly, we observe a strong dose-dependent correlation between daily steps and optimal heart rate variability metrics. Our results provide the ability to interpret continuous heart rate variability for tens of millions of wrist-worn trackers already in use.

Epilepsy is a prevalent condition characterised by recurrent, unpredictable seizures. The diagnosis of epilepsy is by surface electroencephalography (EEG), a time-consuming and uncomfortable process for patients. The diagnosis of seizures using EEG over a brief monitoring period has variable success, dependent on patient tolerance and seizure frequency. Further, the availability of hospital resources, and hardware and software specifications inherently limit the capacity to perform long-term data collection whilst maintaining patient comfort. The application and maintenance of the standard number of electrodes restrict recording time to a maximum of approximately ten days. This limited monitoring period also results in limited data for machine learning models for seizure detection and classification. This work examines the literature on the impact of reduced electrodes on data accuracy and reliability in seizure detection. We present two electrode ranking models that demonstrate the decline in seizure detection performance associated with reducing electrodes. We assert the need for further research in electrode reduction to advance solutions toward portable, reliable devices that can simultaneously provide patient comfort, long-term monitoring and contribute to multimodal patient care solutions.

Accurate and timely documentation in the electronic health record (EHR) is essential for delivering safe and effective patient care. AI-enabled medical tools powered by automatic speech recognition (ASR) offer to streamline this process by transcribing clinical conversations directly into structured notes. However, a critical challenge in deploying these technologies at scale is their variable performance across speakers with diverse accents, which leads to transcription inaccuracies, misinterpretation, and downstream clinical risks. We measured transcription accuracy of Whisper and WhisperX on clinical texts across native and non-native English speakers and found that both models have significantly higher errors for non-native speakers. Fortunately, we found that post-processing the transcripts using GPT-4o recovers the lost accuracy. Our findings indicate that using a chained model approach, WhisperX-GPT, will enhance transcription quality significantly and reduce errors associated with accented speech. We make all code, models, and pipelines freely available.

A method to estimate myocardial action potentials (APs) from electrocardiograms (ECGs) would be an advance in ECG-based diagnosis, utilised for clinical diagnosis, assessment of potential cardiac disease risk and prediction of lethal arrhythmias. However, the ECG inverse problem, which estimates the spatial distribution of AP signals from the ECG, has been considered difficult electromagnetically. For clinical ECG analysis, timescales of collective APs, synchrony and the duration of depolarisation and repolarisation is informative. Thus, we attempted to obtain the time distribution of collective AP transitions from the ECG rather than the spatial distribution. To analyse the variance of the collective myocardial APs from the ECG, we designed a model equation using the probability densities of the Gaussian function of time-series point processes in the cardiac cycle and dipoles of collective APs in the myocardium. The equation to calculate the difference between the two cumulative distribution functions (CDFs) as the positive- and negative-epicardium potential fits well with the R and T waves. The mean, standard deviation, weights, and level of each CDFs are metrics for the variance of the AP transition state of the collective myocardial AP transition states. Clinical ECGs of myocardial ischaemia during coronary intervention showed abnormalities in the aforementioned specific elements of the tensor associated with repolarisation transition variance earlier than in conventional indicators of ischaemia. The tensor could evaluate the beat-to-beat dynamic repolarisation changes between the ventricular epi and endocardium using the Mahalanobis distance (MD). Tensor Cardiography, a method that uses CDF differences CDF as the transition of a collective myocardial AP transition, has the potential to be a new analysis tool for ECGs. Authors SummaryMyocardial action potentials (APs) which indicate electric excitation of the cells can provide important information to suggest the mechanisms of cardiac disease such as myocardial ischemia and arrhythmias. However, it has been challenging to estimate APs from electrocardiograms (ECGs). Unlike other imaging techniques like CT or MRI, the electrocardiographic inverse problem requires estimating the geometric distribution of APs from the ECG, has been considered difficult. Our approach, known as Tensor Cardiography, uses a model equation based on cumulative distribution functions (CDFs) to analyze the time series variance of collective myocardial APs from the ECG. By fitting this equation to the R and T waves, we have obtained a set of metrics that represent beat-to-beat dynamic variance of polarization and repolarization of the epi and endocardium. Our study of ECGs from myocardial ischemia during coronary intervention has demonstrated abnormalities in the tensor elements associated with repolarization, which appeared earlier and more prominently than conventional ST changes. Tensor Cardiography provides a revolutionary analysis tool for ECGs that holds enormous potential for clinical diagnosis, risk assessment, and prediction of lethal arrhythmias. Our approach shows promise as a new frontier in cardiac disease management and has significant implications for patient care.

The analysis of electrophysiological signals from the human body has become increasingly crucial, especially given the widespread adoption of wearable technologies and the growing trend of remote and online monitoring. In situations where demographic patient data is unavailable, the evaluation of such information from electrophysiological signals becomes imperative for making well-informed diagnostic and therapeutic decisions, particularly in ambulatory and urgent cases. This study underscores the significance of this necessity by utilizing intracardiac electrograms to predict patient weight. Intracardiac electrograms were recorded from 44 patients (14 female, with an average age of 59.2{+/-}11.5 years) using a 64-pole basket catheter over a duration of 60 seconds. A dataset comprising 2,816 unipolar electrogram signal segments, each lasting 4 seconds, was utilized. Weight, considered as a continuous variable, underwent discretization into k bins with uniformly distributed widths, where various values of k were experimented with. As the value of k increases, class imbalance also increases. The state-of-the-art time series classification algorithm, Minirocket, was employed alongside the popular machine learning algorithm eXtreme Gradient Boosting (XGBoost). Minirocket consistently demonstrates superior performance compared to XGBoost across all class number scenarios and across all evaluation metrics, such as accuracy, F1 score, and Area Under the Curve (AUC) values, achieving scores of approximately 0.96. Conversely, XGBoost shows signs of overfitting, particularly noticeable in scenarios with higher class imbalance. Tuning probability thresholds for classes could potentially mitigate this issue. Additionally, XGBoosts performance improves with reduced bin numbers, emphasizing the importance of balanced classes. This study provides novel insights into the predictive capabilities of these algorithms and their implications for personalized medicine and remote health monitoring.

Peripheral oxygen saturation (SpO2) and heart rate (HR) are critical physiological measures that clinicians need to observe to decide on an emergency intervention. These measures are typically determined using a contact-based pulse oximeter. This approach may pose difficulties in many cases, such as with young children, patients with burnt or sensitive skin, cognitive impairments, and those undergoing certain medical procedures or severe illnesses. Remote Photoplethysmography (rPPG) allows for unobtrusive sensing of these vital signs in a variety of settings for health monitoring systems. Several research studies have been conducted to use rPPG for this purpose; however, there is still not a commercially available, clinically validated system that overcomes the concerns highlighted in this paper. We present a state-of-the-art review of rPPG-related research conducted including related processes and techniques, such as regions of interest (ROI) selection, extracting the raw signal, pre-processing data, applying noise reduction algorithms, Fast Fourier transforms (FFT), filtering and extracting these vital signs. Further, we present a detailed, critical evaluation of available rPPG systems. Limitations and future directions have also been identified to aid rPPG researchers in further advancing this field.

BackgroundDespite its broadening indications, the implantable cardiac monitor (ICM) records a narrow, nonstandard electrocardiogram (ECG) signal which precludes morphological and functional assessments or the application of 12-lead ECG models. We hypothesize that deep learning can be used to reconstruct 12-lead ECG from a single ICM lead for continuously assessing clinical endpoints outside of rhythm detection alone. ObjectiveTo reconstruct 12-lead ECG from a single ICM lead to detect conduction, repolarization, rhythm, and cardiac functional changes in a large, diverse patient population. MethodsWe annotated 75,450 echocardiogram-ECG pairs with five disease labels a) right bundle branch block, b) left bundle branch block, c) atrial fibrillation, d) QT-prolongation and e) low left ventricular ejection fraction (LVEF) using regex-based parsing of clinician interpretations. We used perceptual loss to train a deep U-Net (ECG12-PerceptNet) to reconstruct 12-lead ECG from a simulated ICM signal. We compared the classification performance of the reconstructed 12-lead ECG against the original 12-lead and single lead ECG in an internal and external test set. Furthermore, we trained a regression model to predict the absolute LVEF using original and reconstructed 12-lead ECGs. ResultsThe reconstructed ECG approached the original 12-lead ECG in classification performance across all endpoints while significantly outperforming the single lead ECG. We show two case studies where sequential LVEF measurements were tracked using LVEF predicted with the original and reconstructed 12-lead ECG. ConclusionIn this paper, we report the ECG12-PerceptNet which reconstructs 12-lead ECG from a simulated ICM signal. This can enable continuous in-home or ambulatory monitoring of cardiac functional changes, potentially reducing hospitalizations and out-of-hospital cardiac arrest.

BackgroundRespiratory heart rate variability (RespHRV), the physiological variation in heart rate in phase with breathing, is mainly generated by central brainstem mechanisms. Its characteristics and determinants in brain-injured patients in the neuro-intensive care unit (neuro-ICU) are poorly understood. ObjectiveTo characterize RespHRV amplitude and phase in brain-injured patients compared to healthy participants, and to explore clinical variables influencing RespHRV in the neuro-ICU. MethodsWe analyzed 55 brain-injured patients (traumatic brain injury, aneurysmal subarachnoid hemorrhage, or other causes) and 31 healthy controls. ECG and respiratory signals were recorded and processed to extract cycle-by-cycle RespHRV amplitude and phase. Group differences were assessed using Mann-Whitney and Watson-Williams tests. In an additional analysis, 55 patients RespHRV amplitude and phase were modeled using generalized linear mixed-effects models to evaluate the impact of sedation, mechanical ventilation mode, vasoactive and analgesic drugs, and time, including random intercepts and slopes for subjects. ResultsCompared to controls, brain-injured patients exhibited a significantly lower RespHRV amplitude (1.04 [0.45, 1.96] vs. 6.21 [4.08, 9.34] bpm; p < 0.001) and an inverted RespHRV phase, with peak heart rate occurring during expiration rather than inspiration. Mixed-effects modeling revealed that machine-triggered ventilation and high level of sedation induced a significant reduction in RespHRV amplitude. ConclusionsBrain-injured patients demonstrate markedly impaired central generation of RespHRV, with peripheral contributors likely accounting for the remaining variability. Ventilation mode and pharmacological interventions strongly alter RespHRV. Restoration of normal RespHRV patterns may serve as a physiological marker of autonomic and brainstem recovery, warranting further investigation in longitudinal studies.