Automatic detection of the spike-and-wave discharges in absence epilepsy for humans and rats using deep learning. (July 2022)

• Record Type: Journal Article
• Authors: Baser, Oguzhan; Yavuz, Melis; Ugurlu, Kutay; Onat, Filiz; Demirel, Berken Utku
• Is Part Of: Biomedical signal processing and control. Volume 76(2022)
• Publication Date: 2022-07
• Subjects: Electroencephalography (EEG) -- Spike-and-wave (SWD) -- Absence epilepsy -- Power spectral density -- Deep learning; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103726
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: This work proposed methods to distinguish the SWDs more efficiently and overcome the limitations of traditional machine learning methods which limits the performance and overshadows algorithms' learning capability for the detection of epileptic EEG signals. Performance evaluation on publicly available TUSZ and one using our collected data. We introduce an efficient convolutional neural network architecture that can be used for both human and animal models to detect SWD using only two channels of EEG signals. The proposed methodology outperforms or reaches the state-of-the-art works on widely-used public TUSZ dataset for the absence epilepsy detection in the human model. On average, it increases the detection rate of SWD patterns while decreasing the number of false detection. Abstract: Automatic detection of spike-and-wave discharges (SWDs) of absence seizures, is a highly time-consuming process requiring trained technicians or neurologists to categorize thousands of non-overlapping epochs of electroencephalography (EEG) data by visually inspecting several interconnections among different channels. This paper aims to develop an algorithm for a non-invasive real-time detection of SWDs in the EEG recordings of humans with absence epilepsy and a genetic model of absence epilepsy. We develop a SWD detection framework using Convolutional Neural Networks. Our approach utilizes the nature of EEG signals; as the brain signals are dynamics in discrete time, we found that it is more efficient and useful to represent the signal's power as a function of frequency and time using Thomson's multitaper power spectral density estimation analysis. Our experiments show that the developed method classified SWDs in humans and rats with high diagnostic performance similar to that of the trained neurologists while using fewer channels, proving that the proposed algorithm can be applied to different domains where the main focus is the detection of SWDs. Although there are different methods to detect SWDs in humans and animals, we showed the need for efficient and more accurate SWD detection. The proposed method, characterized by low computational and memory requirements using non-invasive EEG techniques with fewer channels, offers an efficient multi-purposed deep learning framework to be implemented in wearable or portable devices for accurate real-time detection of SWD patterns in EEG signals. Eventually, the proposed method is a step towards detecting seizures and closed-loop seizure interventions.