| Objective Alzheimer's disease(AD) is a degenerative disease of the nervous system in old age, with cognitive dysfunction and memory impairment as main clinical manifestations, which serious harms the healthy of the elderly and affects survival quality. Previous studies confirmed that patients with AD had abnormal brain network connections, and the study on brain network has important theoretical significance for in-depth understanding of the pathogenesis of AD. In this paper, we studied the multi-channel EEGs frequency domain Granger causality of AD patients and healthy elder to analyze brain networks connectivity specificity and distribution of the core node and stability of the brain network of different frequency bands, providing support for the study of characteristics of AD brain network in resting state.Methods 1. Experimental data: Data were the resting eyes-closed condition 16-channel EEG of 15 AD patients and 15 healthy elder from the Department of Neurology, General Hospital of Tianjin Medical University,. 2. Data preprocessing: Selected a more stable baseline, no significant artifacts of the original EEG data, preprocessed the baseline drift correction, frequency filtering and others. 3. Frequency Analysis: Short time Fourier transform(STFT) was applied to analyze the EEG signal for each channel to access energy distribution concentrated band. Extracted the frequency components: delta[0.5-4 Hz)?theta[4-8 Hz)?alpha[8-13 Hz)?low-beta [13-20Hz)and high-beta[20-30Hz). 4. Directed transform function( DTF) of each band EGGs: Based on multivariate frequency domain Granger causality analysis, calculated causal connection matrix?network functions connection strength and network characteristic parameters of AD group and the healthy control group(HC) in each frequency band. 5. Calculate the network rich-club coefficient: A rich-club coefficient of each frequency band network was calculated, and the rich-club features of the network were also analyzed.6. The core nodes distribution: Based on betweenness centrality(BC) method, calculated the core nodes of the brain network of the AD group and HC group, and analyzed the vulnerability of the core nodes. 7. Statistical analysis: Statistical analysis was conducted with SPSS 22.0 software. We used t test for data of independent sample, measurement data was showed with the mean ± standard error(xsx ±). The comparison of the mean of causal network connection, global efficiency, clustering coefficient of two groups was done at different frequencies.Results 1. Time-frequency analysis AD group EGGs energy focused on delta band, followed by theta band; HC group EEG energy mainly distributed in alpha bands. 2. Construct causal network based on DTF matrix The DTF matrix of AD and HC groups in alpha, low-beta and high-beta bands has significant differences. DTF value in alpha band of AD group was 0.0291 ± 0.0017, DTF value of HC group was 0.0393 ± 0.0009(P<0.001); DTF value in low-beta band of AD group was 0.0251 ± 0.0014, DTF value of HC group was 0.0322 ± 0.0008(P <0.01); DTF value in high-beta band of AD group was 0.0251 ± 0.0014, DTF value of HC group was 0.0322 ± 0.0008(P<0.05). There was no significant difference among full-band, delta and theta bands. 3. Quantitative description of causal network Causal network was constructed based on DTF values, and calculated the network global efficiency and the clustering coefficient for each band of AD and HC groups. Global efficiency: The alpha, low-beta and high-beta bands have significant differences, AD group Eglob value of alpha band was 0.0421± 0.0024, HC group was 0.0532 ± 0.0012(P<0.01); AD group Eglob value of low-beta band was 0.0368 ± 0.0019, HC group was 0.0473 ± 0.0011(P<0.001); AD group Eglob value of high-beta band was 0.0346 ± 0.0018, HC group was 0.0422 ± 0.0011(P<0.01); there was no significant difference among the full-band, delta and theta bands. Clustering coefficient: The alpha and low-beta bands have significant differences,AD group C value of alpha band was 0.0237 ± 0.0015, HC group was 0.0305 ± 0.0007(P<0.01); AD group C value of low-beta band was 0.0198 ± 0.0012, HC group was 0.0244 ± 0.0009(P<0.01); there was no significant difference among the full-band,delta,theta and high-beta bands. 4. Rich-club coefficient There is situation that rich-club coefficient of AD group and HC group was greater than 1, and rich-club coefficient increases with k value illustrating that the causal network has rich-club features in AD and HC groups. 5. The core nodes of network Core nodes distribution of two groups was determined by betweenness centrality, and proceeded statistical analysis. We found that the core nodes of AD group were less than HC group, and there was no core node distribution in the fore-brain region. 6. Network stability With network attack, the vulnerability of the AD group value was 0.1251 ± 0.0055, the HC group vulnerability value was 0.1075 ± 0.0024(P<0.01),the vulnerability of the AD group was significantly greater than HC group in the alpha band, indicating that damage to the core nodes of AD had higher influence on the entire network.Conclusion Based on the resting state of EEG data, the brain networks of the AD and HC groups were constructed using the frequency domain Granger causality analysis, and analysis different characteristics on the perspective of the network connection. Calculated the two groups resting causal network core nodes distributed. 1. The causal connection strength of AD patient's brain was weak, the clustering coefficient and the global efficiency were low in alpha and beta networks indicating that AD patients have impaired brain network compared to the normal elderly. 2. There were rich-club organizations of the brain networks in both AD and HC groups, suggesting that highly degree nodes existed in AD and HC groups and highly degree nodes have a strong tendency to be mutually interconnected compared to the low degree nodes. 3. The core nodes of AD group were less than the HC group, and there was no core node distribution in the fore-brain region, prompting the connections weakened in fore-brain region of AD group. 4. The vulnerability of the AD group in alpha band was significantly higher than the HC group, indicating that damage to the core nodes of AD in alpha band had higher influence on the entire network. |
PDF Full Text Request