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

Abstract

Aims Deep neural networks (DNNs) perform excellently in interpreting electrocardiograms (ECGs), both for conventional ECG interpretation and for novel applications such as detection of reduced ejection fraction (EF). Despite these promising developments, implementation is hampered by the lack of trustworthy techniques to explain the algorithms to clinicians. Especially, currently employed heatmap-based methods have shown to be inaccurate. Methods and results We present a novel pipeline consisting of a variational auto-encoder (VAE) to learn the underlying factors of variation of the median beat ECG morphology (the FactorECG), which are subsequently used in common and interpretable prediction models. As the ECG factors can be made explainable by generating and visualizing ECGs on both the model and individual level, the pipeline provides improved explainability over heatmap-based methods. By training on a database with 1.1 million ECGs, the VAE can compress the ECG into 21 generative ECG factors, most of which are associated with physiologically valid underlying processes. Performance of the explainable pipeline was similar to ‘black box’ DNNs in conventional ECG interpretation (AUROC 0.94 vs 0.96), detection of reduced EF (AUROC 0.90 vs 0.91) and prediction of one-year mortality (AUROC 0.76 vs 0.75). Contrary to the ‘black box’ DNNs, our pipeline provided explainability on which morphological ECG changes were important for prediction. Results were confirmed in a population-based external validation dataset. Conclusions Future studies on DNNs for ECGs should employ pipelines that are explainable to facilitate clinical implementation by gaining confidence in artificial intelligence and making it possible to identify biased models.

Electrocardiogram-based deep learning improves outcome prediction following cardiac resynchronization therapy

“…FactorECG improves upon heatmap-based attempts to make deep learning explainable, as such approaches merely highlight ‘where’ on the ECG significant features are detected but provide no information on which morphological change explains the prediction. 16 Rather, FactorECG allows for ‘quantifiable’ identification of specific ECG features, rendering physicians able to evaluate and confirm the clinical rationale of said features. This is reflected by our results that confirm the known importance of LBBB morphology and QRS AREA for the prediction of echocardiographic response.…”

“…Many clinicians regard deep learning as a ‘black box’, which limits trust in such algorithms. 16 However, our approach to make the model inherently explainable may abate this concern and increase willingness to facilitate clinical adoption of the FactorECG. Although an overall c-statistic of 0.69 leaves room for improvement, our approach is unique in its clinical practicality, with better risk stratification than QRS AREA .…”

“…For this approach, the VAE was pretrained to learn these morphological features (or underlying generative factors) of the ECG, using a data set of 1 144 331 ECGs from 251 473 consecutive patients that underwent ECG recording in the University Medical Centre Utrecht between July 1991 and August 2020. 16 Overlap in the pretraining cohort and patients included in this study was negligible at 0.04% and could not influence the results since the VAE was trained unsupervised (i.e. without any knowledge of CRT outcome).…”

“…Importantly, it has been shown before that only 21 of the 32 factors encode significant information. 16 Hence, only these 21 factors were used in subsequent models. In the training step of the current analysis, the 21 continuous FactorECG values for every ECG, as calculated by the encoder, are used in Cox and logistic regression models to perform prediction of the different endpoints ( Figure 2B ).…”

“…Although the need for very large data sets and the lack of interpretability were deemed common drawbacks of deep learning, a novel technique that uses a variational auto-encoder (VAE; the FactorECG) was recently introduced. 16 This approach enables physicians to better understand and verify the learned ECG features of deep learning algorithms, and makes the technique available to much smaller data sets.…”