Unsupervised feature learning for electrocardiogram data using the convolutional variational autoencoder

Abstract

Most existing electrocardiogram (ECG) feature extraction methods rely on rule-based approaches. It is difficult to manually define all ECG features. We propose an unsupervised feature learning method using a convolutional variational autoencoder (CVAE) that can extract ECG features with unlabeled data. We used 596,000 ECG samples from 1,278 patients archived in biosignal databases from intensive care units to train the CVAE. Three external datasets were used for feature validation using two approaches. First, we explored the features without an additional training process. Clustering, latent space exploration, and anomaly detection were conducted. We confirmed that CVAE features reflected the various types of ECG rhythms. Second, we applied CVAE features to new tasks as input data and CVAE weights to weight initialization for different models for transfer learning for the classification of 12 types of arrhythmias. The f1-score for arrhythmia classification with extreme gradient boosting was 0.86 using CVAE features only. The f1-score of the model in which weights were initialized with the CVAE encoder was 5% better than that obtained with random initialization. Unsupervised feature learning with CVAE can extract the characteristics of various types of ECGs and can be an alternative to the feature extraction method for ECGs.

Improving explainability of deep neural network-based electrocardiogram interpretation using variational auto-encoders

“…3,5,10,24 Other studies investigated the use (variational) auto-encoders 12-lead ECGs in smaller and showed that VAEs can be useful for compression of ECGs, data augmentation, clustering, and feature generation. [25][26][27][28] Interestingly, Kuznetsov et al 28 also deterthat factors are needed to encode a single or beat Our work makes the latent space of a VAE (i.e. FactorECG) clinically useful and explainable to physicians, by linking the ECG factors with known ECG measurements and diagnostic statements (Figures 4 and Tables 2 and 3), (ii) providing sive visualizations offline (Figure 2) and using an tool (https:// decoder.ecgx.ai), and (iii) showing that the ECG factors have equate predictive power in various downstream tasks.…”

Deep learning analysis of mid-infrared microscopic imaging data for the diagnosis and classification of human lymphomas

“…NNs have already been implemented in medicine and used in various studies in the fields of radiology, cardiology, neurology, and pathology [30][31][32][33][34]. But several caveats prevent such techniques from being widely used.…”

“…One unsupervised technique is the Convolutional Autoencoder (CAE), which is trained with unlabelled data [29,30]. Such NNs were developed to work with high-dimensional data and have already been implemented to solve medical problems, such as radiology, cardiology, neurology, and even pathology [30][31][32][33][34]. In brief, the input layer of the CAE compresses the data and creates a code, which is then used to reconstruct them in the output layer [30].…”

Exploring patient medication adherence and data mining methods in clinical big data: A contemporary review

“…Jang et al 313 . deployed a Convolutional Variational Autoencoder (CVAE) for unsupervised ECG feature learning from an expansive collection of 596,000 samples.…”

“…innovated an "autoencoder feature selector" by melding autoencoder and group lasso regression, targeting the extraction of salient linear and nonlinear data features. Jang et al313 deployed a Convolutional Variational Autoencoder (CVAE) for unsupervised ECG feature learning from an expansive collection of 596,000 samples. This method astutely captures vital clinical attributes from the Asian University Medical Center's ICU patients, even without annotations.…”