Body surface potential (BSP) mapping (BSPM) from reduced lead sparse electrocardiogram (ECG) signals is of high clinical relevance. Cost-efficient and non-invasive cardiac activity localization of myocardial infarction scars and arrhythmic sources are a few of its potential applications. However, most BSP reconstruction algorithms’ stability suffers from the underlying ill-posed inverse problem and poor morphological fidelity. Hence, model-constrained regularization is generally employed to incorporate physiological knowledge about the spatio-temporal BSP dynamics. In this treatise, we leverage the recent generative adversarial networks (GAN) paradigm and propose a multi-head attention-based pix2pix GAN architecture with an integrated structural similarity metric for high-fidelity BSPM from sparse ECG sensing. The attention mechanism ensures morphological saliency, while the structural similarity-based loss function regularizes the ill-posed reconstruction during model training and preserves the reconstructed BSP’s intricate fiducial morphology, which is critical for subsequent cardiac activity localization. The proposed BSPM framework has been tested on measured and synthetically generated BSP data utilizing a forward electrophysiology pipeline. The morphologically preserved BSP generated from the proposed model can potentially lead to improved, cost-efficient, and non-invasive cardiac activity monitoring.

The aim of this paper is to create a personalized computational model replicating the cardiorespiratory functionality and modulations of endurance athletes, and evaluate changes in the endurance metrics during dynamic running conditions. A wearable electrocardiogram (ECG) sensor alongside an accelerometer and GPS module within a mobile phone, were utilized to acquire data from marathon runners during weekly training runs. This data served as input for a personalized computational model, termed as 'cardiac digital twin', specifically developed for the runner's heart. Dynamic breathing rate (BR) was computed throughout the run. BR in conjunction with the model-generated cardiac parameters enabled the computation of several clinically relevant cardiorespiratory metrics linked to endurance. Information on intrinsic cardiorespiratory parameters like change in ventilation-perfusion, oxygen uptake efficiency slope, cardiac output, ejection fraction etc., during running could provide a holistic understanding of the cardiorespiratory health of the athlete. It could help runners better understand how their bodies perform under various conditions like high humidity, altitude variation, etc., identify potential health risks, and tailor their training accordingly.

Recent progress in large-scale electrocardiogram (ECG) analysis has yielded cardiologist-level performance, albeit on uni-labeled datasets. However, in the real-world, ECG recordings often have markers of multiple labels like those in the PhysioNet Challenge 2020 dataset (PNC20DB). In the proposed classification strategy, we employ a novel class-distribution rebalancing approach, followed by a hybrid deep learning architecture to manage such multi-class multi-label (MCML) and class-imbalanced datasets. In the hybrid architecture, we pass 12-lead ECG data through a convolution neural network (CNN) and only Lead II ECG data through a residual network (ResNet). We also include domain features as a third processing unit. Finally, we employ k-means on the window-level predictions generated by the concatenation layer to obtain the final class-prediction. The proposed classification model is evaluated on the PNC20DB consisting of 27-class, 12-lead, MCML ECG recordings. The proposed architecture beats the leading entries of PhysioNet Challenge 2020 (PNC2020) and recent state-of-the-art techniques in a five-fold cross-validation method. It generates median AUROC and AUPRC of 94.1% and 56.2%, respectively. It also yields a median PNC2020 challenge metric of 60.4.

In recent years stricter regulation of data privacy and protection has accelerated research in synthetic ECG timeseries generation. With the increasing prevalence of Atrial Fibrillation (AF), its mathematical modelling remains an open research problem due to its unique and complex interval and morphological characteristics. The present research work presents an R-peak to R-peak (RR) interval distribution-aware integrated Hidden Markov Model-Van der Pol (HMM-VdP) oscillator based AF electrocardiogram (ECG) generator model. The approach consists of two parts: A HMM that is trained on RR intervals obtained from real-world AF ECG signals. The output of the trained HMM model is utilized to drive a coupled VdP oscillator model that is modified to generate synthetic ECG time-series with AF-like morphology. A final post-processing algorithm is applied on the output of the HMM-VdP oscillator model to get the synthetic AF time-series. The generated synthetic AF time series is evaluated comparatively with existing state-of-the-art dynamic AF models through qualitative (time-series morphology and Poincaré plot analysis) and quantitative (proposed metric) measures. The proposed AF model has shown significant improvement over the existing dynamic models. The proposed AF model can be utilized for AF detector evaluation, clinical training purpose.

Clinician decision support system (CDSS) is used to assist physicians in personalized disease identification and patient monitoring assessments. In this paper, we propose a Clinical decision system to predict the cardiac recovery score of a subject in post-exercise conditions, employing a hybrid approach using a computational cardiac model and wearable data. We have simulated several cardiac parameters from a previously developed cardiovascular model, using subject specific electrocardiogram (ECG) recorded during exercise recovery phase. These cardiac parameters, when used along with features derived from the wearable ECG data and metadata information to derive the post-exercise recovery score, demonstrate an improvement in capturing the stochastic nature of underlying cardiac conditions among individuals under stress. Generated recovery scores were in agreement with clinical findings for a limited set of patient data.

Cardiac computational models that replicate cardiac functionalities like a ‘digital twin’, can play important roles in serving clinical and research requirements, ranging from presurgical planning to predictive analysis. Personalization of such digital twins are non-trivial and computationally exhaustive due to the uncertainties of fitting mechanistic models to clinical measurements for individual patients. In this paper, we propose a method to personalize the hemodynamics functionality of a cardiac digital twin using a particle swarm optimization (PSO) framework that tunes the cardiac chamber properties based on subject-specific echocardiogram (Echo) and electrocardiogram (ECG) data. Parameters derived from ECG like information related to time instances of pumping action in cardiac chambers and Echo parameters like left ventricle end systolic and diastolic diameters and volumes are used to personalize the cardiac chamber parameters of an existing lumped cardiac hemodynamics model. Using this strategy, personalized hemodynamics parameters are generated for a healthy and two diseased subjects, suffering from cardiac Amyloidosis and Grade I Diastolic dysfunction. The proposed method of non-invasive modality-based (Echo and ECG) personalizing cardiac functionality can play a critical role in facilitating the circulatory hemodynamics model for clinical settings and aid in enabling precision medicine applications.

The bio-acoustic information contained within heart sound signals are utilized by physicians world-wide for auscultation purpose. However, the heart sounds are inherently susceptible to noise contamination. Various sources of noises like lung sound, coughing, sneezing, and other background noises are involved in such contamination. Such corruption of the heart sound signal often leads to inconclusive or false diagnosis. To address this issue, we have proposed a novel U-Net based deep neural network architecture for denoising of phonocardiogram (PCG) signal in this paper. For the design, development and validation of the proposed architecture, a novel approach of synthesizing real-world noise corrupted PCG signals have been proposed. For the purpose, an open-access real-world noise sample dataset and an open-access PCG dataset has been utilized. The performance of the proposed denoising methodology has been evaluated on the synthesized noisy PCG dataset. The performance of the proposed algorithm has been compared with existing state-of-the-art (SoA) denoising algorithms qualitatively and quantitatively. The proposed denoising technique has shown improvement in performance as comparison to the SoAs.

Body Surface Potential Map (BSPM) is an augmented version of 12-lead Electrocardiogram (ECG) with an increased number of electrodes that provides high density spatial information of the cardiac potential on the torso surface for source localization of cardiac abnormalities. A total reconstruction of BSPM is a challenging task. In this paper, we propose a novel Generative Adversarial Network (GAN) architecture to reconstruct 65-lead BSP from standard 12-lead ECG. We present Time-Series GAN (TSGAN), a specially designed modified pix2pix GAN for an accurate reconstruction of time-series BSP data. Further, we propose certain regularization terms in the generator loss function to preserve the key morphological properties of the generated waveform which is a major contribution of this work. The proposed architecture outperforms a Variational Autoencoder (VAE) and a baseline GAN on publicly available dataset in reconstructing 65-lead BSP with morphological preservation.

Single-lead Electrocardiogram (ECG) can be easily measured by a commercial smartwatch or a dedicated wearable device. The waveforms are often susceptible to background noise and motion artifacts introducing errors in disease interpretation. An effective yet light-weight de-noising of ECG is an open area of research. In this paper, we propose a novel convolutional autoencoder structure considering a number of regularization terms like sparsity constraint, contractive regularization and L2 norm for ECG de-noising. The deep learning model is duly optimized to efficiently run on low-power edge devices. The proposed approach is evaluated on a simulated and a real-world single-lead ECG database recorded from normal subjects as well as patients having Atrial Fibrillation (AF) and other kinds of abnormal heart rhythms. A thorough comparison is performed with a number of related signal processing and deep learning based prior approaches. Experimental results show that the proposed autoencoder yields the least Root Mean Square Error (RMSE) in reconstruction of clean signals from input ECG corrupted due to addition of noise. Our approach is also able to preserve the relevant morphological properties in the reconstructed ECG data for successful detection of AF and other abnormal rhythms. The optimized model is deployed on a low-power single-board computer for real-time noise cleaning.

Wearable cardioverter defibrillator (WCD) is a life saving, wearable, noninvasive therapeutic device that prevents fatal ventricular arrhythmic propagation that leads to sudden cardiac death (SCD). WCD are frequently prescribed to patients deemed to be at high arrhythmic risk but the underlying pathology is potentially reversible or to those who are awaiting an implantable cardioverter-defibrillator. WCD is programmed to detect appropriate arrhythmic events and generate high energy shock capable of depolarizing the myocardium and thus re-initiating the sinus rhythm. WCD guidelines dictate very high reliability and accuracy to deliver timely and optimal therapy. Computational model-based process validation can verify device performance and benchmark the device setting to suit personalized requirements. In this article, we present a computational pipeline for WCD validation, both in terms of shock classification and shock optimization. For classification, we propose a convolutional neural network-"Long Short Term Memory network (LSTM) full form" (Convolutional neural network- Long short term memory network (CNN-LSTM)) based deep neural architecture for classifying shockable rhythms like Ventricular Fibrillation (VF), Ventricular Tachycardia (VT) vs. other kinds of non-shockable rhythms. The proposed architecture has been evaluated on two open access ECG databases and the classification accuracy achieved is in adherence to American Heart Association standards for WCD. The computational model developed to study optimal electrotherapy response is an in-silico cardiac model integrating cardiac hemodynamics functionality and a 3D volume conductor model encompassing biophysical simulation to compute the effect of shock voltage on myocardial potential distribution. Defibrillation efficacy is simulated for different shocking electrode configurations to assess the best defibrillator outcome with minimal myocardial damage. While the biophysical simulation provides the field distribution through Finite Element Modeling during defibrillation, the hemodynamic module captures the changes in left ventricle functionality during an arrhythmic event. The developed computational model, apart from acting as a device validation test-bed, can also be used for the design and development of personalized WCD vests depending on subject-specific anatomy and pathology.

Objective: Atrial fibrillation (AF) and other types of abnormal heart rhythm are related to multiple fatal cardiovascular diseases that affect the quality of human life. Hence the development of an automated robust method that can reliably detect AF, in addition to other non-sinus and sinus rhythms, would be a valuable addition to medicine. The present study focuses on developing an algorithm for the classification of short, single-lead electrocardiogram (ECG) recordings into normal, AF, other abnormal rhythms and noisy classes. Approach: The proposed classification framework presents a two-layer, three-node architecture comprising binary classifiers. PQRST markers are detected on each ECG recording, followed by noise removal using a spectrogram power based novel adaptive thresholding scheme. Next, a feature pool comprising time, frequency, morphological and statistical domain ECG features is extracted for the classification task. At each node of the classification framework, suitable feature subsets, identified through feature ranking and dimension reduction, are selected for use. Adaptive boosting is selected as the classifier for the present case. The training data comprises 8528 ECG recordings provided under the PhysioNet 2017 Challenge. F1 scores averaged across the three non-noisy classes are taken as the performance metric. Main result: The final five-fold cross-validation score achieved by the proposed framework on the training data has high accuracy with low variance (0.8254 +/- 0.0043). Significance: Further, the proposed algorithm has achieved joint first place in the PhysioNet/Computing in Cardiology Challenge 2017 with a score of 0.83 computed on a hidden test dataset.

Feature subset selection and identification of appropriate classification method plays an important role to optimize the predictive performance of supervised machine learning system. Current literature makes isolated attempts to optimize the feature selection and classifier identification. However, feature set has an intrinsic relationship with classification technique and together they form a `model' for classification task. In this paper, we propose AutoModeling that finds optimal learning model and jointly optimize the feature and hypothesis space to maximize performance measure objective function. It is an automated framework of selecting the ensemble model {selected feature subset, selected classifier} from a given superset of features and classifiers learned from given training dataset in a computational efficient manner. We introduce novel relax-greedy search with our proposed patience function as a wrapper feature selection that maximizes the predictive performance and eliminates the classical nesting effect. We perform extensive experimentations on different types of publicly available datasets and AutoModeling demonstrates superior performance over relevant state-of-the-art methods, expert-driven manual methods and deep neural networks.

This paper presents an automated approach for interpretable feature recommendation for solving signal data analytics problems. The method has been tested by performing experiments on datasets in the domain of prognostics where interpretation of features is considered very important. The proposed approach is based on Wide Learning architecture and provides means for interpretation of the recommended features. It is to be noted that such an interpretation is not available with feature learning approaches like Deep Learning (such as Convolutional Neural Network) or feature transformation approaches like Principal Component Analysis. Results show that the feature recommendation and interpretation techniques are quite effective for the problems at hand in terms of performance and drastic reduction in time to develop a solution. It is further shown by an example, how this human-in-loop interpretation system can be used as a prescriptive system.

This paper presents an algorithm for online detection of arcing in low-voltage (230 V, 50 Hz) distribution systems. Electromagnetic radiation has been used as a feature for discrimination between arc and nonarc signals. The output of the electromagnetic radiation sensor has been analyzed in the spectral domain using the log-spectral distance metric for the detection of arcing. For validation of this method, a testbench comprising an arc generation setup has been developed in the laboratory. The algorithm has been tested against arc and arc-mimicking normal signals using the testbench to validate its effectiveness. Further, the algorithm is implemented online using xPC target (an embedded platform) for studying the feasibility and accuracy of its online implementation. A system based on this method can continuously monitor the health of a distribution network and, thus, can be very helpful in safeguarding against fire hazards due to electric arcing.

Remote cardiac health management is an important healthcare application. We have developed Heartmate that enables basic screening of cardiac health using low cost sensors or smartphone-inbuilt sensors without manual intervention. It consists of robust denoising algorithm along with effective anomaly analytics for physiological signals. Heartmate identifies and eliminates signal corruption as well as detects cardiac anomaly condition from physiological cardiac signals like heart sound or phonocardiogram (PCG) and photoplethysmogram (PPG).

Rampant power theft at the low voltage consumer end is a growing concern for the power distribution companies. This paper proposes an effective method for detection of power theft at low voltage consumer end. The proposed method is designed to reliably detect hooking in service line cable and bypassing of electric energy meter. In this method, a low magnitude, high-frequency, non-interfering signal has been injected into the power line. Two LC traps have been designed and placed on either side of the energy meter to restrict the flow of the injected component from reaching the load end. In either case of bypassing or hooking, the power of the high-frequency component will deviate from its value under normal operating condition. The proposed algorithm utilizes this fact for the power theft detection. In order to attain the purpose, power spectral density (PSD) coefficients of the acquired line current are evaluated, magnitude of the PSD coefficients corresponding to the injected frequency are identified and subsequently thresholding technique is applied on it for detection of power theft. The proposed algorithm has been validated in simulation environment as well as with real-world data. This algorithm can play a significant role in arresting power theft.

Phonocardiogram (PCG) records heart sound and murmurs, which contains significant information of cardiac health. Analysis of PCG signal has the potential to detect abnormal cardiac condition. However, the presence of noise and motion artifacts in PCG hinders the accuracy of clinical event detection. Thus, noise detection and elimination are crucial to ensure accurate clinical analysis. In this paper, we present a robust denoising technique, Proclean that precisely detects the noisy PCG signal through pattern recognition, and statistical learning. We propose a novel self-discriminant learner that ensures to obtain distinct feature set to distinguish clean and noisy PCG signals without human-in-loop. We demonstrate that our proposed denoising leads to higher accuracy in subsequent clinical analytics for medical investigation. Our extensive experimentations with publicly available MIT-Physionet datasets show that we achieve more than 85% accuracy for noisy PCG signal detection. Further, we establish that physiological abnormality detection improves by more than 20%, when our proposed denoising mechanism is applied.

In this paper, we present a methodology for classifying normal, atrial fibrillation (AF), non-AF related other abnormal heart rhythms and noisy recordings by analysing single lead ECG signal of short duration.In a two layer binary cascaded approach proposed in our methodology, an unlabelled recording is initially classified into one of the two intermediate classes ('normal+others' and 'AF+noisy') at the first layer before actual classification at the second layer.The Physionet Challenge 2017 dataset containing more than 8500 ECG recordings are used for creation of training models and interval validation.The proposed methodology yields an average F1-score of 0.91, 0.79 and 0.77 respectively in classifying normal, AF and other rhythms on the training dataset using 5-fold cross validation.Results also show that, the said methodology, when applied on a hidden test set maintained by the challenge organisers yields F1-score values of 0.92, 0.86 and 0.74 in classifying the same.

We aim to develop a reliable and robust algorithm that accurately analyses a single short PCG recording (10-60s) from a single precordial location to determine the presence of heart abnormality for the Physionet/ Computing-in-Cardiology 2016 challenge. We extract timing information for the fundamental Heart Sounds i.e. S1 and S2 using Hidden Markov Model based Springer's improved version of Schmidt's method. These values are then used to generate statistical features set in temporal, frequency, time-frequency and wavelet domain. We choose the optimal feature set out of the pool of overall 54 features using mutual information based minimum Redundancy Maximum Relevance (mRMR) technique. In order to cope with bad signals, we also check the signal quality of the PCG signal. Signals are rejected for further normal abnormal classification when the outside/background noise has rendered them useless for processing. Then, non-linear radial basis function based Support Vector Machine (SVM) classifier along with ensemble based methods is used to train with the reduced optimal feature sets, on a balanced training set chosen from the group of all PCG datasets. Our algorithm is tested with hidden Physionet Challenge 2016 datasets and performance achieved is: Sensitivity (Se) = 0.7749, Specificity (Sp) = 0.7891 and Overall Score calculated as mean (Se, Sp) = 0.7820.