Key result
LSTM model analyzing 15-minute ECG signals achieves ~100% accuracy for diagnosing CHF.
Why the study?
Existing CHF diagnostic methods require high competence or rely on trial-and-error ECG feature extraction in small, balanced datasets, prompting the development of automated deep-learning algorithms.
Can deep-learning techniques using LSTM architectures accurately classify congestive heart failure from short time-step ECG signals?
Can deep-learning techniques using LSTM architectures accurately classify congestive heart failure from short time-step ECG signals?
An LSTM-based deep learning model using 15 minutes of ECG signals can accurately classify congestive heart failure, potentially aiding in preliminary diagnosis.
May aid preliminary CHF detection from brief ECGs; leaves open prospective validation before clinical use.
Congestive heart failure (CHF) is characterized by the heart's inability to pump blood adequately throughout the body without increased intracardiac pressure. Diverse approaches are used to treat CHF. These approaches, which include physical examination, echocardiography, and laboratory testing, require a high degree of competence to interpret findings and make diagnoses. Moreover, existing methods do not account for the relationships between variables and thus provide limited performance. Electrocardiogram (ECG), as a non-invasive test, may be used for CHF early diagnosis, which would require further examination to be referred. A previous study revealed a significant correlation between heart failure (HF) and ECG features. However, the method was only performed on small, balanced data; then, the features must be derived from trial and error. The current paper proposes deep-learning techniques—recurrent neural networks (RNNs) with long short-term memory (LSTM) architectures—to create a diagnostic algorithm that achieves high accuracy with limited information and automated feature extraction. The ECG signals used in this study were obtained from the public PhysioNet databases. We fine-tuned the hyperparameters of 24 LSTM models to obtain the best model. Moreover, ECG signal segmentation was compared among the first five and fifteen minutes as features. Out of the 24 LSTM models, the model with the first fifteen minutes of ECG signals (model 1) obtained the highest accuracy, sensitivity, specificity, precision, and F1-score (99.86%, 99.85%, 99.85%, 99.87%, and 99.86%, respectively). The first fifteen minutes of ECG signals performed well because the LSTM model learned an increasing number of features. In conclusion, the proposed LSTM model could give a clinician a preliminary CHF diagnosis for further medical attention. Deep learning can be a useful predictive method for increasing the number of identified CHF patients.
No takes yet. Share an insight, caveat, or question.
Darmawahyuni et al. (2020) studied Congestive heart failure. LSTM model using the first fifteen minutes of ECG signals vs. LSTM model using the first five minutes of ECG signals was evaluated on Accuracy for CHF diagnosis. An LSTM model using the first fifteen minutes of ECG signals achieved 99.86% accuracy, 99.85% sensitivity, and 99.85% specificity for diagnosing congestive heart failure.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: