Ein Shoka Athar A, Alkinani Monagi H, El-Sherbeny A S, El-Sayed Ayman, Dessouky Mohamed M
Department of Computer Science and Engineering, Faculty of Electronic Engineering, Menoufia University, Menouf, Egypt.
Department of Computer Science and Artificial Intelligence, College of Computer Science and Engineering, University of Jeddah, Jeddah, Saudi Arabia.
Brain Inform. 2021 Feb 12;8(1):1. doi: 10.1186/s40708-021-00123-7.
Seizure is an abnormal electrical activity of the brain. Neurologists can diagnose the seizure using several methods such as neurological examination, blood tests, computerized tomography (CT), magnetic resonance imaging (MRI) and electroencephalogram (EEG). Medical data, such as the EEG signal, usually includes a number of features and attributes that do not contains important information. This paper proposes an automatic seizure classification system based on extracting the most significant EEG features for seizure diagnosis. The proposed algorithm consists of five steps. The first step is the channel selection to minimize dimensionality by selecting the most affected channels using the variance parameter. The second step is the feature extraction to extract the most relevant features, 11 features, from the selected channels. The third step is to average the 11 features extracted from each channel. Next, the fourth step is the classification of the average features using the classification step. Finally, cross-validation and testing the proposed algorithm by dividing the dataset into training and testing sets. This paper presents a comparative study of seven classifiers. These classifiers were tested using two different methods: random case testing and continuous case testing. In the random case process, the KNN classifier had greater precision, specificity, positive predictability than the other classifiers. Still, the ensemble classifier had a higher sensitivity and a lower miss-rate (2.3%) than the other classifiers. For the continuous case test method, the ensemble classifier had higher metric parameters than the other classifiers. In addition, the ensemble classifier was able to detect all seizure cases without any mistake.
癫痫发作是大脑的一种异常电活动。神经科医生可以使用多种方法来诊断癫痫发作，如神经学检查、血液检查、计算机断层扫描（CT）、磁共振成像（MRI）和脑电图（EEG）。医学数据，如EEG信号，通常包含许多不包含重要信息的特征和属性。本文提出了一种基于提取最显著EEG特征进行癫痫诊断的自动癫痫分类系统。所提出的算法包括五个步骤。第一步是通道选择，通过使用方差参数选择受影响最大的通道来最小化维度。第二步是特征提取，从选定的通道中提取最相关的11个特征。第三步是对从每个通道提取的11个特征求平均值。接下来，第四步是使用分类步骤对平均特征进行分类。最后，通过将数据集划分为训练集和测试集来进行交叉验证和测试所提出的算法。本文对七个分类器进行了比较研究。这些分类器使用两种不同的方法进行测试：随机病例测试和连续病例测试。在随机病例过程中，KNN分类器比其他分类器具有更高的精度、特异性和阳性预测率。不过，集成分类器比其他分类器具有更高的灵敏度和更低的漏诊率（2.3%）。对于连续病例测试方法，集成分类器比其他分类器具有更高的度量参数。此外，集成分类器能够毫无错误地检测出所有癫痫发作病例。
