Efficient Epileptic Seizure Detection Method Based on EEG Images: The Reduced Descriptor Patterns

Abstract

Detecting epileptic seizures through Electroencephalography (EEG) is a crucial but challenging task due to the time-consuming and error-prone nature of visual analysis in long-term EEG recordings. To overcome these limitations, numerous approaches have been proposed. In this work, we present a novel automated approach to seizure detection. Our approach is evaluated using rhythmicity spectrograms extracted from the publicly available CHB-MIT Scalp EEG database. For each channel in this database, both ictal and nonictal segments were converted separately into rhythmic spectrograms by Short-Time Fourier Transform (STFT). Image-based seizure detection is a new technique that utilizes visual representations of EEG data, which can potentially provide complementary interpretations to traditional signal-based analysis. The integration of image-based techniques can improve the accuracy and robustness of seizure detection algorithms, enabling more confident diagnostics and timely treatments for people with epilepsy. This study focuses on a novel seizure detection approach that employs image feature extraction as the main method. By offering this innovative approach, the research aims to improve the effectiveness and credibility of seizure detection Algorithms. The main innovation of this study is the introduction of a novel methodology inspired by the Decimal Descriptor Pattern (DDP) feature extraction technique. DDP involves the representation of images by decimal codes ranging from 0 to 10. The proposed approach, called Reduced Descriptor Pattern (RDP), is characterized by its emphasis on consolidating feature sets to only 5 codes, thus optimizing the process while conserving descriptive integrity. The study compared the Reduced Descriptor (RDP) and the GrayLevel Co-occurrence Matrix (GLCM) methods for feature extraction. A Support Vector Machine (SVM) classifier was utilized to differentiate between seizure and non-seizure EEG 2D images. Our proposed method, when combined with the SVM classifier, achieved excellent results in comparison to the well-known GLCM feature extraction method and DDP. We conducted a comparative study among the RDP, DDP, and GLCM methods. The results demonstrate exceptional accuracy, highlighting the efficacy of our proposed method in seizure detection. Thus, our approach provides a valuable tool for expediting seizure diagnosis and reducing reliance on subjective visual evaluations.