Detection and classification of adult epilepsy using hybrid deep learning approach

Abstract

The electroencephalogram (EEG) has emerged over the past few decades as one of the key tools used by clinicians to detect seizures and other neurological abnormalities of the human brain. The proper diagnosis of epilepsy is crucial due to its distinctive nature and the subsequent negative effects of epileptic seizures on patients. The classification of minimally pre-processed, raw multichannel EEG signal recordings is the foundation of this article’s unique method for identifying seizures in pre-adult patients. The new method makes use of the automatic feature learning capabilities of a three-dimensional deep convolution auto-encoder (3D-DCAE) associated with a neural network-based classifier to build an integrated framework that endures training in a supervised manner to attain the highest level of classification precision among brain state signals, both ictal and interictal. A pair of models were created and evaluated for testing and assessing our method, utilizing three distinct EEG data section lengths, and a tenfold cross-validation procedure. Based on five evaluation criteria, the labelled hybrid convolutional auto-encoder (LHCAE) model, which utilizes a classifier based on bidirectional long short-term memory (Bi-LSTM) and an EEG segment length of 4 s, had the best efficiency. This proposed model has 99.08 ± 0.54% accuracy, 99.21 ± 0.50% sensitivity, 99.11 ± 0.57% specificity, 99.09 ± 0.55% precision, and an F1-score of 99.16 ± 0.58%, according to the publicly available Children’s Hospital Boston (CHB) dataset. Based on the obtained outcomes, the proposed seizure classification model outperforms the other state-of-the-art method’s performance in the same dataset.

Multimodal Biomedical Image Segmentation using Multi-Dimensional U-Convolutional Neural Network

“…The VGFR Spectrogram Detector achieved an impressive classification accuracy of 88.1%, while the Voice Impairment Classifier attained a remarkable 89.15% accuracy. Saravanan et al [26] leveraging the automated feature extraction capabilities of a three-dimensional deep convolutional autoencoder (3D-DCAE), a novel method has been developed that integrates a neural network-based classifier to construct a unified framework capable of supervised training, achieving the pinnacle of classification accuracy for both ictal and interictal brain state signals. To thoroughly evaluate our method, two distinct models were meticulously crafted and assessed, employing three separate EEG data section lengths and a rigorous tenfold cross-validation procedure.…”

Evolutionary transfer optimization-based approach for automated ictal pattern recognition using brain signals

“…Consequently, many researchers opt to exclusively utilize deep learning (DL)-based methods. While these methods perform optimally when trained with sufficiently large annotated/synthetic datasets ( Pascual et al, 2020 ; Srinivasan et al, 2023 ; Dash et al, 2024 ), practical applications often encounter scarcity of such datasets and/or face challenges with the “black box” nature of the model. This opacity seldom instills confidence in clinicians to adopt new computer-aided diagnosis (CAD) systems.…”