Detection of Alzheimer's disease using nonlinear features of ERP signal
- OLCIAS
- Sep 1, 2023
- 2 min read
Talk Title: Detection of Alzheimer's disease using nonlinear features of ERP signal
Author: Elias Mazrooei Rad Biomedical Engineering Department, Khavaran Institute of Higher Education, Mashhad Iran Email: elias_mazrooei@yahoo.com Volume: Brain, Mind and AI 2023_CN41_AA04
International Live Conference on Brain, Mind and Artificial Intelligence Date: September 2, 2023 Time: 8 PM Indian Standard Time Location: Online Conference Theme: Discover the Future of Brain Science and AI Integration
Abstract
Non-linear dynamic analysis has been widely used for physiological signals, and the interesting point is the analysis of the EEG signal of an Alzheimer's patient by this method. The relationship between the non-linear characteristics of the EEG signal of Alzheimer's patients is associated with a decrease in the complexity of the EEG pattern and a decrease in the functional connections of different areas of the brain cortex in these patients. The basic assumption in this processing method is that the EEG signal itself has non-linear characteristics and the interaction of neurons is non-linear and chaotic. The level of this disease should be diagnosed according to the relationship of this disease with different characteristics in the brain signal. First, with appropriate pre-processing, non-linear properties such as phase diagram, correlation dimension, entropy and Lyapunov view have been extracted and used for classification of the channelized neural network. The use of deep learning methods, including the channelized neural network, can have more appropriate and accurate results among other classification methods. In the case of using a CNN network with non-linear features of the brain signal, in the stimulation mode, the accuracy of the results is 95% in healthy people, 92.5% in mild patients, and 97.5% in severe patients, in the recall mode, the accuracy of the results in healthy people 75% and in mild patients it is 72.5% and in severe patients it is 87.5%. As the severity of the disease increases, the entropy feature, which indicates the amount of disorder and signal changes, decreased. Because with the increase in severity of the disease and the loss of interactions and neurons of the nervous system, the amount of signal changes decreases. The examined periods are eyes closed, eyes open, recall and stimulation, and among these 4 periods, the stimulation period was the best period for recording the brain signal because it is more effective to check the speed of the stimulation response to diagnose Alzheimer's disease. If the classifier separates the groups based on linear and non-reversible methods by extracting inappropriate features, correct results will not be produced. On the other hand, the use of the channelized neural network with the feature extraction and group separation approach has been able to achieve the highest accuracy and precision in this study. Keywords: Alzheimer's disease, ERP signal, nonlinear features, convolutional neural network. Biography The individual is a PhD holder in Medical Engineering and a dedicated researcher specializing in Alzheimer's disease diagnosis through the integration of brain signal methodologies and medical imaging, aimed at benefiting both healthcare professionals and patients through the utilization of advanced deep learning techniques. They currently hold the position of a faculty member and serve as the Research Director at the Khavaran Institute of Higher Education.
