Physionet Challenge Database Research Articles

Analysis of heart sound anomalies using ensemble learning

Phonocardiogram (PCG) signal analysis is a common method for evaluating the condition of the heart and detecting possible anomalies such as cardiovascular diseases. This work concentrates on the Physionet challenge database that stores PCG recordings for more than one thousand subjects, including healthy and pathological records. A complete methodology is provided to analyze and classify PCG data. The PCG signals are first filtered and segmented into different parts, then analyzed by applying a feature extraction process, followed by classifying the signal as that of a healthy or unhealthy person. The extracted optimal features subset includes statistical components, such as the mean and standard deviation from different parts of the signal, in addition to wavelet-based features. The classification mainly relies on bagging and boosting algorithms, as well as adequately preparing the data in order to yield an enhanced ensemble classifier. The work further provides the approach to combine multiple classification models to improve accuracy. The effect of segmenting the different beats of the PCG on classification scores is also addressed, and the results are shown to be of high precision with an accuracy score of 86.6% on the hidden test, in comparison with recent and best performing literature that achieved 86%. Also, the proposed methodology is extended to the Pascal heart sounds challenge with accurate results that exceed previous works confirming the performance and robustness of the work since it can be applied to multiple databases and sources. The work further provides important insights regarding analyzing PCG recordings and discusses future work possibilities.

Detection of sleep apnea from heart beat interval and ECG derived respiration signals using sliding mode singular spectrum analysis

The heartbeat interval (HBI) signal (RR-time series), and electrocardiogram (ECG) derived respiration (EDR) signal quantify the information about the cardiopulmonary activity, and monitoring these two signals simultaneously will provide more information for the sleep apnea detection. This paper proposes a novel approach to detect sleep apnea using both HBI and EDR signals. The approach consists of the decomposition of both HBI and EDR signals into reconstructed components (RCs) or modes using a data-driven signal processing approach namely, the sliding mode singular spectrum analysis (SM-SSA), extraction of features from each RC, and the use of classifier for the detection of sleep apnea. The features such as the mean and the standard deviation values are extracted from the instantaneous amplitude (IA) and instantaneous frequency (IF) of each RC of both HBI and EDR signals. The classifiers, such as the stacked autoencoder based deep neural network (SAE-DNN), and support vector machine (SVM) are considered to classify normal and apnea episodes using the statistical features obtained from the RCs of HBI and EDR signals. The proposed approach is evaluated using different public databases such as apnea-ECG database, University College Dublin (UCD) database, and Physionet challenge database, respectively. The results demonstrate that the combination of the statistical features and SVM classifier has the sensitivity and specificity values of 82.45% and 79.72%, respectively using the 10-fold cross-validation based selection of training and test instances from the apnea-ECG database. Moreover, for subject-specific cross-validation, the proposed method has an average sensitivity and specificity values of 62.87%, and 81.53%, respectively. The proposed method has achieved the accuracy values of 94.3%, and 72% for per-recording based classification of sleep apnea and normal classes using signals from apnea-ECG and UCD databases.

SleepNet: automated sleep analysis via dense convolutional neural network using physiological time series

Objective: In this work, a dense recurrent convolutional neural network (DRCNN) was constructed to detect sleep disorders including arousal, apnea and hypopnea using polysomnography (PSG) measurement channels provided in the 2018 PhysioNet Challenge database. Approach: Our model structure is composed of multiple dense convolutional units (DCU) followed by a bidirectional long-short term memory (LSTM) layer followed by a softmax output layer. The sleep events, including sleep stages, arousal regions and multiple types of apnea and hypopnea, are manually annotated by experts, which enables us to train our proposed network using a multi-task learning mechanism. Three binary cross-entropy loss functions, corresponding to sleep/wake, target arousal and apnea-hypopnea/normal detection tasks, are summed up to generate our overall network loss function that is optimized using the Adam method. Our model performance was evaluated using two metrics: the area under the precision-recall curve (AUPRC) and the area under the receiver operating characteristic curve (AUROC). To measure our model generalization, 4-fold cross-validation was also performed. For training, our model was applied to full night recording data. Main results: Finally, the average AUPRC and AUROC values associated with the arousal detection task were 0.505 and 0.922, respectively, on our testing dataset. An ensemble of four models trained on different data folds improved the AUPRC and AUROC to 0.543 and 0.931, respectively. Significance: Our proposed algorithm achieved the first place in the official stage of the 2018 PhysioNet Challenge for detecting sleep arousals with an AUPRC of 0.54 on the blind testing dataset.

Classifying Heart Sounds Using Images of Motifs, MFCC and Temporal Features.

Cardiovascular disease is the leading cause of death in the world, and its early detection is a key to improving long-term health outcomes. The auscultation of the heart is still an important method in the medical process because it is very simple and cheap. To detect possible heart anomalies at an early stage, an automatic method enabling cardiac health low-cost screening for the general population would be highly valuable. By analyzing the phonocardiogram signals, it is possible to perform cardiac diagnosis and find possible anomalies at an early-term. Therefore, the development of intelligent and automated analysis tools of the phonocardiogram is very relevant. In this work, we use simultaneously collected electrocardiograms and phonocardiograms from the Physionet Challenge database with the main objective of determining whether a phonocardiogram corresponds to a "normal" or "abnormal" physiological state. Our main contribution is the methodological combination of time domain features and frequency domain features of phonocardiogram signals to improve cardiac disease automatic classification. This novel approach is developed using both features. First, the phonocardiogram signals are segmented with an algorithm based on a logistic regression hidden semi-Markov model, which uses electrocardiogram signals as a reference. Then, two groups of features from the time and frequency domain are extracted from the phonocardiogram segments. One group is based on motifs and the other on Mel-frequency cepstral coefficients. After that, we combine these features into a two-dimensional time-frequency heat map representation. Lastly, a binary classifier is applied to both groups of features to learn a model that discriminates between normal and abnormal phonocardiogram signals. In the experiments, three classification algorithms are used: Support Vector Machines, Convolutional Neural Network, and Random Forest. The best results are achieved when both time and Mel-frequency cepstral coefficients features are considered using a Support Vector Machines with a radial kernel.

Classification of short single-lead electrocardiograms (ECGs) for atrial fibrillation detection using piecewise linear spline and XGBoost

Objective: Detection of atrial fibrillation is important for risk stratification of stroke. We developed a novel methodology to classify electrocardiograms (ECGs) to normal, atrial fibrillation and other cardiac dysrhythmias as defined by the PhysioNet Challenge 2017. Approach: More specifically, we used piecewise linear splines for the feature selection and a gradient boosting algorithm for the classifier. In the algorithm, the ECG waveform is fitted by a piecewise linear spline, and morphological features relating to the piecewise linear spline coefficients are extracted. XGBoost is used to classify the morphological coefficients and heart rate variability features. Main results: The performance of the algorithm was evaluated by the PhysioNet Challenge database (3658 ECGs classified by experts). Our algorithm achieved an average F1 score of 81% for a 10-fold cross-validation and also achieved 81% for F1 score on the independent testing set. This score is similar to the top 9th score (81%) in the official phase of the PhysioNet Challenge 2017. Significance: Our algorithm presents a good performance on multi-label short ECG classification with selected morphological features.

An automated ECG signal quality assessment method for unsupervised diagnostic systems

In this paper, the authors present an automated method for quality assessment of electrocardiogram (ECG) signal. Our proposed method not only detects and classifies the ECG noises but also localizes the ECG noises which can play a crucial role in extracting reliable clinical features for ECG analysis systems. The proposed method is based on three stages: Wavelet decomposition of ECG signal into sub-bands; simultaneous ECG signal and noise reconstruction; extraction of temporal features such as maximum absolute amplitude, zerocrossings, kurtosis and autocorrelation function for detection, localization and classification of ECG noises including flat line (FL), time-varying noise or pause (TVN), baseline wander (BW), abrupt change (AB), power line interference (PLI), muscle artifacts (MA) and additive white Gaussian noise (AWGN). The proposed method is tested and validated against manually annotated ECG signals corrupted with aforementioned noises taken from MIT-BIH arrhythmia database, Physionet challenge database, and real-time recorded ECG signals. Comparative detection and classification results depict the superior performance of the proposed method over state of art methods. Detection results show that our method can achieve an average sensitivity (Se), average specificity (Sp) and accuracy (A) of 99.61%, 98.51%, 99.49% respectively. Also, the method achieves a Se of 98.18%, and Sp of 94.97% for real-time recorded ECG signals. The method has an average timing error of 0.14s in localizing the noise segments. Further, classification results demonstrate that the proposed method achieves an average sensitivity (Se), average positive predictivity (PP) and classification accuracy (Ac) of 98.53%, 98.89%, 97.50% respectively.

Fetal electrocardiogram estimation using polynomial eigenvalue decomposition

In this paper, we propose the application of polynomial matrix eigenvalue decomposition (PEVD) to the problem of fetal electrocardiogra (ECG) extraction from real ECG recordings obtained from abdominal leads. We model the fetal ECG extraction problem as a broadband sensor array signal processing problem in order to account for the broadband nature of the ECG noise present in the recordings. An algorithm for providing an approximate PEVD is used in order to estimate the broadband noise subspace. Suppression of the broadband noise and maternal ECG is achieved by carrying out an orthonormal projection of the recordings onto the estimated fetal subspace. The proposed scheme was evaluated with multichannel synthetic ECG signals and real ECG data from the PhysioNet Challenge database and is shown to perform favorably as compared to prior-art methods. Results indicate that our method is more robust than the prior-art ones for the task of fetal ECG estimation.

Automatic ECG quality scoring methodology: mimicking human annotators

An algorithm to determine the quality of electrocardiograms (ECGs) can enable inexperienced nurses and paramedics to record ECGs of sufficient diagnostic quality. Previously, we proposed an algorithm for determining if ECG recordings are of acceptable quality, which was entered in the PhysioNet Challenge 2011. In the present work, we propose an improved two-step algorithm, which first rejects ECGs with macroscopic errors (signal absent, large voltage shifts or saturation) and subsequently quantifies the noise (baseline, powerline or muscular noise) on a continuous scale. The performance of the improved algorithm was evaluated using the PhysioNet Challenge database (1500 ECGs rated by humans for signal quality). We achieved a classification accuracy of 92.3% on the training set and 90.0% on the test set. The improved algorithm is capable of detecting ECGs with macroscopic errors and giving the user a score of the overall quality. This allows the user to assess the degree of noise and decide if it is acceptable depending on the purpose of the recording.