Affiliations: [a] School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, Jiangsu, China
| [b] Tuberculosis and Respiratory Department, Wuhan Jinyintan Hospital, Wuhan, Hubei, China
| [c] Department of Radiotherapy, Affiliated Hospital, Jiangnan University, Wuxi, Jiangsu, China
Correspondence:
[*]
Corresponding author. Pengjiang Qian, School of Artificial Intelligence and Computer Science, Jiangnan University, 1800 Lihu Avenue, Wuxi, Jiangsu 214122, China. E-mail qianpjiang@jiangnan.edu.cn.
Abstract: Epilepsy is a common brain disease, caused by abnormal discharge of human brain neurons, resulting in brain dysfunction syndrome. Although epilepsy does not have much impact on patients in the short term, but long-term frequent seizures can lead to physical and mental impact of patients. At present, the method used to detect epilepsy is to make a comprehensive judgment by EEG examination combined with clinical symptoms. With the application of AI technology, some advanced algorithms have been used to assist medical diagnosis. In this trend, we use extreme learning machine to observe and detect patients with epilepsy. ELM has the characteristics of high efficiency and high precision, so it is often used in regression and classification problems. However, in the face of different data sets, ELM structure is not enough to achieve good performance. This is caused by the uneven distribution of data in different data sets. To solve this problem, we add the transfer learning module to the basic ELM structure. The purpose of adding transfer learning is to divide the disordered data in the domain space and construct a data set suitable for ELM learning. Specifically, the raw data are mapped to high-dimensional space by kernel method through domain adaptive method. Secondly, in high-dimensional space, the distance between different domains should be reduced appropriately. Finally, ELM method is used to analyze and predict the changed data set. In the whole algorithm process, due to the characteristics of ELM updating weight, only a certain amount of hidden nodes are needed, and the training process is very fast. At the same time, after adding the transfer learning function module, the accuracy of ELM is also satisfactory. In this paper, the epilepsy data of patients were used for comparative experiments. The experimental results show that the method can maintain high efficiency and satisfactory accuracy.
Keywords: Extreme learning machine, domain adaptation, signal classification, Epileptic EEG
