DEVELOPMENT OF AN EPILEPTIC SEIZURE PREDICTION ALGORITHM USING R–R INTERVALS WITH SELF-ATTENTIVE AUTOENCODER
Abstract:
Epilepsy is a neurological disorder characterized by recurrent seizures, which can significantly impact the quality of life for affected individuals. The ability to predict epileptic seizures in advance can provide valuable insights for seizure management and personalized treatment strategies. In recent years, machine learning techniques have shown promising results in seizure prediction. This study aims to develop an algorithm for epileptic seizure prediction using R-R intervals, which are the time intervals between successive R-peaks in an electrocardiogram (ECG) signal, along with a self-attentive autoencoder.
The proposed algorithm consists of two main stages: training and prediction. In the training stage, a self-attentive autoencoder is trained using a large dataset of R-R intervals extracted from ECG signals recorded from both seizure and non-seizure periods. The self-attentive autoencoder is designed to learn meaningful representations of the R-R interval sequences by capturing both local and global dependencies within the data. This enables the model to extract relevant features that can discriminate between pre-seizure and non-seizure patterns.
Once the autoencoder is trained, the prediction stage involves feeding the R-R interval sequences into the trained model to obtain reconstructed representations. A threshold-based anomaly detection approach is then applied to identify significant deviations from the normal patterns. When the reconstructed R-R intervals exhibit anomalous behavior, indicating a potential pre-seizure state, an alarm is triggered to alert the patient or healthcare provider.
The proposed algorithm was evaluated using a dataset comprising ECG recordings from a group of patients with epilepsy. The performance of the algorithm was assessed in terms of sensitivity, specificity, and prediction accuracy. The results demonstrated that the developed algorithm achieved high sensitivity and specificity in predicting epileptic seizures, thereby showing its potential as a valuable tool for early seizure detection and intervention.
In conclusion, this study presents a novel approach for epileptic seizure prediction using R-R intervals with a self-attentive autoencoder. The proposed algorithm offers promise in accurately identifying pre-seizure states, enabling timely interventions and improved seizure management for individuals with epilepsy. Further research and validation on larger and diverse datasets are warranted to establish the clinical utility and generalizability of the algorithm.