A Convolutional Long Short-Term Memory-Based Neural Network for Epilepsy Detection From EEG

Abstract

Epilepsy is a severe neurological disorder characterized by recurrent seizures, which increases the risk of death three times more than normal. Currently Electroencephalography (EEG) has emerged as a highly promising technique for the diagnosis of epilepsy. The majority of current EEG-based epilepsy detection research have employed a variety of deep learning-based models, but most of the approaches suffer from poor generalizability, optimal design and performance rates. To address these issues, this study aims to develop an efficient framework based on deep spatio-temporal neural network called convolutional long short-term memory (ConvLSTM) for epilepsy detection from EEG signals. In the proposed model, firstly standard 19 channel EEG data are selected and resampled at 256Hz, and then those signals are segmented into three-second time frames. Afterward, the segmented data are fed as input to the ConvLSTM model for identifying epileptic patients from normal subjects. To generalize the proposed model, we have tested it on two different datasets with varying population sizes. We have used the five-fold cross validation and leave-one-out cross validation schemes to eliminate the experiment's biases. To further validate the proposed framework, we have carried out various ablation studies. The experimental results demonstrate that the proposed model outperforms the current state-of-theart results for the studied datasets, making it suitable for use as an automated system for the diagnosis of epilepsy.

Using Explainable Artificial Intelligence to Obtain Efficient Seizure-Detection Models Based on Electroencephalography Signals

“…Considering that the methods used in the literature have a large number of input vectors [ 10 , 11 , 18 ] (which also requires a greater number of samples for training) and make use of more robust machines such as DL [ 13 , 14 ], the proposed approach simplifies the solution by delivering a numerically equivalent performance, employing temporal feature extraction, using simpler classifiers (XGBoost), and reducing the dimensionality through an explainable technique (SHAP), enabling model interpretability.…”

“…Given the significance and intricacy of seizure detection, considerable efforts have been directed towards machine learning (ML)-based methodologies for the automated analysis of EEG signals. Supervised ML classifiers are commonly employed in this domain [ 7 , 8 , 9 , 10 , 11 , 12 ], along with deep learning (DL) techniques [ 13 , 14 ]. These approaches are geared towards developing models adept at binary classification [ 7 , 8 , 9 , 11 , 12 ] as well as multiclass classification tasks [ 15 , 16 ].…”

Epilepsy detection based on multi-head self-attention mechanism

“…They balanced EEG data through data augmentation and designed a 1D CNN for efficient detection. Moreover, CNN variants [ 9 , 10 ] and hybrid models [ 11 – 13 ] are also gradually being applied in the field of epilepsy signal detection.…”

Exploring Frequency Band-Based Biomarkers of EEG Signals for Mild Cognitive Impairment Detection

“…Among those techniques, MMSE is a manual question-and-answer test, and PET, MRI, and CT scans are expensive [1], [11]. In this study, we have used EEG data for MCI detection as it is non-invasive, cost-effective, widely available, and portable [12], [13]. Moreover, it captures brain electrical activity over time, making it valuable for assessing cognitive concerns [3], [14], [15].…”

“…EEG signal has several rhythm bands like delta (0-4 Hz), theta (4-8 Hz), alpha (8)(9)(10)(11)(12)(13)(14)(15)(16), beta (16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32) and gamma (>32 Hz) [21]. Among those rhythm bands, <0.5 Hz and >32 Hz bands are considered noise [1], [18] and can be removed from the classification process.…”