Machine learning based detection of T-wave alternans in real ambulatory conditions

BACKGROUND AND OBJECTIVE

T-wave alternans (TWA) is a fluctuation in the repolarization morphology of the ECG. It is associated with cardiac instability and sudden cardiac death risk. Diverse methods have been proposed for TWA analysis. However, TWA detection in ambulatory settings remains a challenge due to the absence of standardized evaluation metrics and detection thresholds.

METHODS

In this work we use traditional TWA analysis signal processing-based methods for feature extraction, and two machine learning (ML) methods, namely, K-nearest-neighbor (KNN) and random forest (RF), for TWA detection, addressing hyper-parameter tuning and feature selection. The final goal is the detection in ambulatory recordings of short, non-sustained and sparse TWA events.

RESULTS

We train ML methods to detect a wide variety of alternant voltage from 20 to 100 μV, i.e., ranging from non-visible micro-alternans to TWA of higher amplitudes, to recognize a wide range in concordance to risk stratification. In classification, RF outperforms significantly the recall in comparison with the signal processing methods, at the expense of a small lost in precision. Despite ambulatory detection stands for an imbalanced category context, the trained ML systems always outperform signal processing methods.

CONCLUSIONS

We propose a comprehensive integration of multiple variables inspired by TWA signal processing methods to fed learning-based methods. ML models consistently outperform the best signal processing methods, yielding superior recall scores.

Use of machine learning and Poincaré density grid in the diagnosis of sinus node dysfunction caused by sinoatrial conduction block in dogs

BACKGROUND

Sinus node dysfunction because of abnormal impulse generation or sinoatrial conduction block causes bradycardia that can be difficult to differentiate from high parasympathetic/low sympathetic modulation (HP/LSM).

HYPOTHESIS

Beat-to-beat relationships of sinus node dysfunction are quantifiably distinguishable by Poincaré plots, machine learning, and 3-dimensional density grid analysis. Moreover, computer modeling establishes sinoatrial conduction block as a mechanism.

ANIMALS

Three groups of dogs were studied with a diagnosis of: (1) balanced autonomic modulation (n = 26), (2) HP/LSM (n = 26), and (3) sinus node dysfunction (n = 21).

METHODS

Heart rate parameters and Poincaré plot data were determined [median (25%-75%)]. Recordings were randomly assigned to training or testing. Supervised machine learning of the training data was evaluated with the testing data. The computer model included impulse rate, exit block probability, and HP/LSM.

RESULTS

Confusion matrices illustrated the effectiveness in diagnosing by both machine learning and Poincaré density grid. Sinus pauses >2 s differentiated (P < .0001) HP/LSM (2340; 583-3947 s) from sinus node dysfunction (8503; 7078-10 050 s), but average heart rate did not. The shortest linear intervals were longer with sinus node dysfunction (315; 278-323 ms) vs HP/LSM (260; 251-292 ms; P = .008), but the longest linear intervals were shorter with sinus node dysfunction (620; 565-698 ms) vs HP/LSM (843; 799-888 ms; P < .0001).

CONCLUSIONS

Number and duration of pauses, not heart rate, differentiated sinus node dysfunction from HP/LSM. Machine learning and Poincaré density grid can accurately identify sinus node dysfunction. Computer modeling supports sinoatrial conduction block as a mechanism of sinus node dysfunction.

Arrhythmia classification for non-experts using infinite impulse response (IIR)-filter-based machine learning and deep learning models of the electrocardiogram

Arrhythmias are a leading cause of cardiovascular morbidity and mortality. Portable electrocardiogram (ECG) monitors have been used for decades to monitor patients with arrhythmias. These monitors provide real-time data on cardiac activity to identify irregular heartbeats. However, rhythm monitoring and wave detection, especially in the 12-lead ECG, make it difficult to interpret the ECG analysis by correlating it with the condition of the patient. Moreover, even experienced practitioners find ECG analysis challenging. All of this is due to the noise in ECG readings and the frequencies at which the noise occurs. The primary objective of this research is to remove noise and extract features from ECG signals using the proposed infinite impulse response (IIR) filter to improve ECG quality, which can be better understood by non-experts. For this purpose, this study used ECG signal data from the Massachusetts Institute of Technology Beth Israel Hospital (MIT-BIH) database. This allows the acquired data to be easily evaluated using machine learning (ML) and deep learning (DL) models and classified as rhythms. To achieve accurate results, we applied hyperparameter (HP)-tuning for ML classifiers and fine-tuning (FT) for DL models. This study also examined the categorization of arrhythmias using different filters and the changes in accuracy. As a result, when all models were evaluated, DenseNet-121 without FT achieved 99% accuracy, while FT showed better results with 99.97% accuracy.

A Novel Approach to Fetal ECG Extraction Using Temporal Convolutional Encoder-Decoder Network (TCED-Net)

To extract weak fetal ECG signals from the mixed ECG signal on the mother's abdominal wall, providing a basis for accurately estimating fetal heart rate and analyzing fetal ECG morphology. First, based on the relationship between the maternal chest ECG signal and the maternal ECG component in the abdominal signal, the temporal convolutional encoder-decoder network (TCED-Net) model is trained to fit the nonlinear transmission of the maternal ECG signal from the chest to the abdominal wall. Then, the maternal chest ECG signal is nonlinearly transformed to estimate the maternal ECG component in the abdominal mixed signal. Finally, the estimated maternal ECG component is subtracted from the abdominal mixed signal to obtain the fetal ECG component. The simulation results on the FECGSYN dataset show that the proposed approach achieves the best performance in F1 score, mean square error (MSE), and quality signal-to-noise ratio (qSNR) (98.94%, 0.18, and 8.30, respectively). On the NI-FECG dataset, although the fetal ECG component is small in energy in the mixed signal, this method can effectively suppress the maternal ECG component and thus extract a clearer and more accurate fetal ECG signal. Compared with existing algorithms, the proposed method can extract clearer fetal ECG signals, which has significant application value for effective fetal health monitoring during pregnancy.

Performance of Off-the-Shelf Machine Learning Architectures and Biases in Detection of Low Left Ventricular Ejection Fraction

Artificial intelligence - machine learning (AI-ML) is a computational technique that has been demonstrated to be able to extract meaningful clinical information from diagnostic data that are not available using either human interpretation or more simple analysis methods. Recent developments have shown that AI-ML approaches applied to ECGs can accurately predict different patient characteristics and pathologies not detectable by expert physician readers. There is an extensive body of literature surrounding the use of AI-ML in other fields, which has given rise to an array of predefined open-source AI-ML architectures which can be translated to new problems in an "off-the-shelf" manner. Applying "off-the-shelf" AI-ML architectures to ECG-based datasets opens the door for rapid development and identification of previously unknown disease biomarkers. Despite the excellent opportunity, the ideal open-source AI-ML architecture for ECG related problems is not known. Furthermore, there has been limited investigation on how and when these AI-ML approaches fail and possible bias or disparities associated with particular network architectures. In this study, we aimed to: (1) determine if open-source, "off-the-shelf" AI-ML architectures could be trained to classify low LVEF from ECGs, (2) assess the accuracy of different AI-ML architectures compared to each other, and (3) to identify which, if any, patient characteristics are associated with poor AI-ML performance.

Machine learning models of 6-lead ECGs for the interpretation of left ventricular hypertrophy (LVH)

BACKGROUND

Left Ventricular Hypertrophy (LVH) is closely linked to the cardiovascular disease prognosis, and thus, timely diagnosis improves outcomes. Diagnosis is challenging due to dependency on doctor's visits and a 12‑lead ECG. In addition, the interpretation of LVH from ECGs is challenging due to variability of ECG measurements, body habitus, electrode positioning, several LVH ECG criteria and EP mechanisms. The aims of this study are to evaluate different big data-driven machine learning models for ECG LVH interpretation based on limb leads only, and to compare the performance of an ECG parameter-based statistical model with a deep learning-based model.

METHODS AND DATA

The first two models are binary class Random Forest (RF) models, an ensemble learning method which constructs many decision trees at training time and predicts the class chosen by the greatest number of trees at inference time. One random forest is trained using the following five features: lead aVL R-wave amplitude, lead I, II, aVL ST segment amplitude, and QRS duration. The second RF model uses 54 features across all limb leads, including the five features used by the smaller model. The second type of model is a multi-class deep neural network (DNN) which takes median beats of 6 limb leads arranged in Cabrera sequence as input. The signal preprocessing included forming median beats, filtering with a 40-Hz lowpass filter, and down-sampling to 125 Hz. The DNN models consist of 1 lead-formation convolutional layer, 5 downsampling convolutional resnet blocks with skip connections, and 3 fully connected layers. The training dataset consisted of 1 million 10-s 12‑lead ECGs, and an independent test dataset consisted of 250,000 10-s ECGs from the Mayo Clinic.

RESULTS

The five-parameter RF model has the prediction performance of Area Under the Receiver-Operator Curve (AUC) 0.78, and the larger RF model had AUC of 0.83. The DNN model for ECG LVH detection achieves AUC 0.92 using only the limb leads, compared to an AUC of 0.98 for the full 12‑lead DNN.

CONCLUSION

The study shows that machine learning models trained only on limb leads achieve promising results with potential to add clinical value to early detection mechanisms. We also observe that the RF model splits parameters by thresholds known to be characteristic of LVH, and that the DNN model can automatically detect morphology differences from 6 limb lead ECGs. This will be meaningful for expanding the capabilities of potential electrical LVH detection in mobile 6‑lead ECG devices.

Machine Learning approach for TWA detection relying on ensemble data design

Background and objective

T-wave alternans (TWA) is a fluctuation of the ST-T complex of the surface electrocardiogram (ECG) on an every-other-beat basis. It has been shown to be clinically helpful for sudden cardiac death stratification, though the lack of a gold standard to benchmark detection methods limits its application and impairs the development of alternative techniques. In this work, a novel approach based on machine learning for TWA detection is proposed. Additionally, a complete experimental setup is presented for TWA detection methods benchmarking.

Methods

The proposed experimental setup is based on the use of open-source databases to enable experiment replication and the use of real ECG signals with added TWA episodes. Also, intra-patient overfitting and class imbalance have been carefully avoided. The Spectral Method (SM), the Modified Moving Average Method (MMA), and the Time Domain Method (TM) are used to obtain input features to the Machine Learning (ML) algorithms, namely, K Nearest Neighbor, Decision Trees, Random Forest, Support Vector Machine and Multi-Layer Perceptron.

Results

There were not found large differences in the performance of the different ML algorithms. Decision Trees showed the best overall performance (accuracy 0.88 ± 0.04 , precision 0.89 ± 0.05 , Recall 0.90 ± 0.05 , F1 score 0.89 ± 0.03 ). Compared to the SM (accuracy 0.79, precision 0.93, Recall 0.64, F1 score 0.76) there was an improvement in every metric except for the precision.

Conclusions

In this work, a realistic database to test the presence of TWA using ML algorithms was assembled. The ML algorithms overall outperformed the SM used as a gold standard. Learning from data to identify alternans elicits a substantial detection growth at the expense of a small increment of the false alarm.

Golden Standard or Obsolete Method? Review of ECG Applications in Clinical and Experimental Context

Cardiovascular system and its functions under both physiological and pathophysiological conditions have been studied for centuries. One of the most important steps in the cardiovascular research was the possibility to record cardiac electrical activity. Since then, numerous modifications and improvements have been introduced; however, an electrocardiogram still represents a golden standard in this field. This paper overviews possibilities of ECG recordings in research and clinical practice, deals with advantages and disadvantages of various approaches, and summarizes possibilities of advanced data analysis. Special emphasis is given to state-of-the-art deep learning techniques intensely expanded in a wide range of clinical applications and offering promising prospects in experimental branches. Since, according to the World Health Organization, cardiovascular diseases are the main cause of death worldwide, studying electrical activity of the heart is still of high importance for both experimental and clinical cardiology.

The Application of Computer Techniques to ECG Interpretation

It is over 120 years since Einthoven introduced the electrocardiogram [...]