Fast processing and classification of epileptic seizures based on compressed EEG signals
Achraf Djemal,Ahmed Yahia Kallel,Cherif Ouni,Rihem El Baccouch,Dhouha Bouchaala,Fatma Kammoun Feki,Chahnez Charfi Triki,Ahmed Fakhfakh,Olfa Kanoun
DOI: https://doi.org/10.1016/j.compbiomed.2024.109346
IF: 7.7
2024-11-23
Computers in Biology and Medicine
Abstract:The diagnosis of epilepsy based on visual inspection of electroencephalogram (EEG) signals is inherently complex and prone to error, even for physicians, mainly due to the large number of signals involved and the variability between individuals. These same challenges make the development of portable epilepsy diagnostic systems for everyday use difficult. Key obstacles include the immense complexity of signal processing and the inherent ambiguity in accurately classifying disease. For these reasons, we propose in this paper the deployment of compressive sensing to condense EEG signals while preserving relevant information, allowing seizure classification based on systematically selected features of the reconstructed signals. Based on a dataset comprising EEG recordings from 13 epileptic patients with various seizure types, we explore the deployment of the discrete cosine transform (DCT) and random matrix multiplication for compression ratios ranging from 5% to 70%, balancing data reduction with signal fidelity. Following the extraction of relevant features, selection was performed based on mutual information and a correlation matrix to preserve only the most relevant features for analysis. For classification, following a comparison of adequate machine learning models, XGBoost is chosen as it realizes a classification accuracy of 98.78%. The CS method was implemented on an STM32 microcontroller and a Raspberry Pi for reconstruction and classification, to demonstrate feasibility as an embedded system. At 70% compression, significant improvements have been observed: 70% file size reduction, 84% decrease in transmission time (from 2518.532s to 400.392s), and substantial energy savings (e.g., from 11.5±0.707 mWh to 4.5±0.707 mWh for Patient 12). Thereby, the signal quality was maintained with PSNR of 16.15±3.98 and Pearson correlation coefficient of 0.68±0.15. The proposed system highlights the potential for efficient, portable, real-time epilepsy diagnosis systems that achieve precise and fully automated seizure classification.
engineering, biomedical,computer science, interdisciplinary applications,mathematical & computational biology,biology