| Objective Working memory(WM) is one of important cognitive functions. Behavioral studies have found that the behavioral error rate increases significantly when WM load reaches the limit. What is the neural mechanism of the above phenomenon? In this paper, we calculated the information transfer of multichannel electroencephalographs(EEGs) recorded from 16 healthy subjects when they performed a Sternberg working memory task with different memory loads to study how the information transfer efficiency, causal flow and information flow of the EEG causal network change as the WM load increases. It is expected to provide an insight of the brain network mechanism underlying the behavioral error rate increase significantly when the WM load reaches the limit.Methods1. Behavioral analysis: Reaction time and accuracy were calculated across all the subjects for load 1, 3, 5, 7 and 9, respectively.2. EEG data: 16-channel EEGs were recorded from 16 healthy subjects(age ranging from 22 to 32 years) while they performed a Sternberg WM paradigm with WM load 1, 3, 5, 7 or 9, respectively. The data came from the 16 healthy subjects and 240 trials for each WM load.3. EEGs preprocessing: Removal of the environmental noise caused by baseline drift, frequency interference and the artifacts, such as eye movement from the original recording EEGs.4. EEGs time-frequency analysis: Calculate the spatial distribution of all channels,extract the critical channel of the EEGs. Apply the short-time Fourier transform to calculate time-frequency distribution of the critical channel and analyze the principal frequency band in working memory, extract the principal frequency band component of the EEGs.5. Directed transform function(DTF) of critical frequency band(θ component in EEGs): Apply the multi-variable Granger causality analysis to calculate the DTFij of all channels from j to i and DTF of the θ component for WM load 1, 3, 5,7 and 9, respectively.6. The global efficiency Eglob of network: Based on EEG network constructed by DTFij, calculate the Eglob of the network information transfer for WM load 1, 3, 5,7 and 9, respectively.7. Weighted causal flow: Apply the weighted DTFij to calculate the causal flow of each channel.8. Information flow: Employ the DTFij to calculate the information flow from channel j to i. Extract the strongest 5% of the DTFij show the size and direction of information flow in the graph.Results1. Behavioral results Reaction time(RT): The RTs for load 1, 3, 5, 7 and 9 were 469.98±20.47 ms,536.51±17.89 ms, 580.66±24.83 ms, 622.24±18.25 ms and 759.43±32.34 ms,respectively. The RTs increased significantly from WM load 7 to 9(p<0.01).Accuracy: The mean accuracy values for load 1, 3, 5, 7 and 9 were99.72±0.25%, 99.04±0.51%, 98.35±0.49%, 97.19±0.45% and 93.34±0.91%,respectively. The accuracy decreased significantly from WM load 7 to 9(p<0.001).2. EEGs time-frequency distribution The power of EEGs was concentrated in the retention period during the working memory task. In the retention period, the power was concentrated in theθ band, and the power was maximum at channel Fz.3. Global efficiency(Eglob) of EEG causal network Based on the EEG causal network constructed by DTFij, the global efficiency(Eglob) values for WM load 1, 3, 5, 7 and 9 were 0.0203±0.0007, 0.0216±0.0007,0.0230±0.0003, 0.0232±0.0005 and 0.0203±0.0010, respectively. The Eglob decreased significantly from WM load 7 to 9(p<0.05).4. Causal flow The causal flow of channel Fz for WM load 1, 3, 5, 7 and 9 were0.2567±0.0581, 0.4361±0.0872, 0.5676±0.1055, 0.9092±0.1313 and0.5715±0.1239, respectively. The sum of positive causal flow for WM load 1, 3, 5,7 and 9 were 1.5057±0.0836, 1.6098±0.0856, 1.8664±0.1083, 2.0112±0.1202 and1.6861±0.1486, respectively. The causal flow decreased significantly from WM load 7 to 9(p<0.05).5. Information flow The maximum information flow was flow from channel Fz to others. The information flow decreased significantly from WM load 7 to 9(p<0.01).Conclusions1. The RT increases and the accuracy decreases significantly when the WM load increases from 7 to 9, indicating that the load 7 is the limit load in WM.2. θ band is the critical frequency band in WM.3. The global efficiency Eglob of the EEG causal network increases with loads and decreases significantly in load 9, which shows the information transfer efficiency decreased mechanism when the WM load reaches the limit.4. The channels with positive causal flow are causal source. The sum of the positive causal flow increase with loads and decrease significantly in load 9, which shows the causal flow decrease mechanism when the WM load reaches the limit.5. The normalized of the sum of the strongest 5% of the DTFij in the DTF matrix increase with loads and decrease significantly in load 9, which shows the information flow decrease mechanism when the WM load reaches the limit. |
PDF Full Text Request