Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A novel approach to classification of ECG arrhythmia types with latent ODEs

Angelina Yan, Matt L. Sampson, Peter Melchior

TL;DR

This work tackles the challenge of arrhythmia detection from wearable ECG data by introducing an end-to-end pipeline that uses a path-minimized latent ODE to encode high-frequency ECG waveforms into compact latent vectors, which are then classified by a gradient boosted decision tree. The model is trained on the MIT-BIH Arrhythmia dataset at 360 Hz and evaluated under downsampled conditions of 90 Hz and 45 Hz, showing minimal degradation in macro-averaged AUC-ROC across frequencies (0.984, 0.978, 0.976). The approach leverages SMOTE balancing and mode aggregation over multiple latent samples to handle class imbalance and produce robust predictions. The results suggest that high-fidelity arrhythmia classification can be achieved with lower-frequency wearable data, enabling smaller, longer-lasting devices for continuous cardiac monitoring, though limitations include dataset size and minority-class performance. Overall, the work demonstrates that latent-ODE encodings can preserve essential morphology for reliable downstream classification even when signal fidelity is reduced, which has practical implications for proactive health monitoring.

Abstract

12-lead ECGs with high sampling frequency are the clinical gold standard for arrhythmia detection, but their short-term, spot-check nature often misses intermittent events. Wearable ECGs enable long-term monitoring but suffer from irregular, lower sampling frequencies due to battery constraints, making morphology analysis challenging. We present an end-to-end classification pipeline to address these issues. We train a latent ODE to model continuous ECG waveforms and create robust feature vectors from high-frequency single-channel signals. We construct three latent vectors per waveform via downsampling the initial 360 Hz ECG to 90 Hz and 45 Hz. We then use a gradient boosted tree to classify these vectors and test robustness across frequencies. Performance shows minimal degradation, with macro-averaged AUC-ROC values of 0.984, 0.978, and 0.976 at 360 Hz, 90 Hz, and 45 Hz, respectively, suggesting a way to sidestep the trade-off between signal fidelity and battery life. This enables smaller wearables, promoting long-term monitoring of cardiac health.

A novel approach to classification of ECG arrhythmia types with latent ODEs

TL;DR

This work tackles the challenge of arrhythmia detection from wearable ECG data by introducing an end-to-end pipeline that uses a path-minimized latent ODE to encode high-frequency ECG waveforms into compact latent vectors, which are then classified by a gradient boosted decision tree. The model is trained on the MIT-BIH Arrhythmia dataset at 360 Hz and evaluated under downsampled conditions of 90 Hz and 45 Hz, showing minimal degradation in macro-averaged AUC-ROC across frequencies (0.984, 0.978, 0.976). The approach leverages SMOTE balancing and mode aggregation over multiple latent samples to handle class imbalance and produce robust predictions. The results suggest that high-fidelity arrhythmia classification can be achieved with lower-frequency wearable data, enabling smaller, longer-lasting devices for continuous cardiac monitoring, though limitations include dataset size and minority-class performance. Overall, the work demonstrates that latent-ODE encodings can preserve essential morphology for reliable downstream classification even when signal fidelity is reduced, which has practical implications for proactive health monitoring.

Abstract

12-lead ECGs with high sampling frequency are the clinical gold standard for arrhythmia detection, but their short-term, spot-check nature often misses intermittent events. Wearable ECGs enable long-term monitoring but suffer from irregular, lower sampling frequencies due to battery constraints, making morphology analysis challenging. We present an end-to-end classification pipeline to address these issues. We train a latent ODE to model continuous ECG waveforms and create robust feature vectors from high-frequency single-channel signals. We construct three latent vectors per waveform via downsampling the initial 360 Hz ECG to 90 Hz and 45 Hz. We then use a gradient boosted tree to classify these vectors and test robustness across frequencies. Performance shows minimal degradation, with macro-averaged AUC-ROC values of 0.984, 0.978, and 0.976 at 360 Hz, 90 Hz, and 45 Hz, respectively, suggesting a way to sidestep the trade-off between signal fidelity and battery life. This enables smaller wearables, promoting long-term monitoring of cardiac health.

Paper Structure

This paper contains 16 sections, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Training data
  4. Path-minimized latent ODEs
  5. Gradient boosted decision tree
  6. Full classification pipeline
  7. Experiments and Results
  8. Latent ODE performance
  9. Classification results
  10. Discussion
  11. Limitations and future work
  12. Conclusion
  13. Model details
  14. Latent ODE details
  15. Boosted classifier details
  16. ...and 1 more sections

Figures (3)

  • Figure 1: a) Reconstruction of a randomly sampled ECG timeseries with the latent ODE prediction in red and the true signal in black. b) UMAP of the latent feature vectors from the test set of the BIH-MIT data, with colors indicating the arrhythmia class. c) AUC-ROC curves from the GBDT classifier based on latent ODE encodings of ECG curves sampled at 360, 90, and 45 Hz.
  • Figure 2: Normalized confusion matrices from the GBDT on the ECG data at samping frequencies of 360 Hz, 90 Hz, and 45 Hz from left to right, respectively.
  • Figure 3: Additional reconstructions of randomly sampled ECG timeseries with the latent ODE prediction in red and the true signal in black.