[An autoencoder model based on one-dimensional neural network for epileptic EEG anomaly detection].

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Context:

• Epilepsy diagnosis relies heavily on analyzing electroencephalogram (EEG) signals, which can be complex and time-consuming.
• Automated detection of epileptic seizures from EEG is crucial for timely intervention and patient management.
• Current deep learning models for EEG anomaly detection face challenges in efficiency and accuracy.

Purpose:

• To develop an efficient autoencoder model utilizing a one-dimensional convolutional neural network (1DCNN) for the feature extraction of epileptic EEG anomalies.
• To train the autoencoder to learn the normal patterns of EEG data in a low-dimensional feature space.
• To evaluate the performance of the proposed 1DCNN-autoencoder (1DCNN-AE) model on public EEG datasets.

Summary:

• The 1DCNN-AE model effectively captures local information from normal EEG signals to learn representations.
• An anomaly score is derived from the input-output difference, with thresholds determined by ROC curve analysis.
• The model demonstrated superior performance compared to LSTM-VAE and GRU-VAE models on the CHB-MIT and TUH EEG datasets, achieving high detection rates and AUC scores with significantly lower computational cost (FLOPs).

Impact:

• Provides an effective and computationally efficient method for detecting epileptic seizures using EEG.
• Offers a potential tool for improving the accuracy and speed of epilepsy diagnosis.
• Contributes to the advancement of deep learning applications in clinical neuroscience and medical diagnostics.