Figure 1.
Behavioral task and ECoG electrode locations.
A) Monkeys performed sequential right arm and hand movements, which consisted of reaching to a knob, grasping the knob with a lateral grip, pulling the knob closer, releasing the knob, and returning the hand to the home position, in a 3-D workspace. During the task, ECoG and EMG signals were recorded simultaneously. B) Schematic diagrams of ECoG electrode locations in left hemisphere. The planar-surface platinum electrode arrays were placed on the gyrus between the central sulcus (CS) and the arcuate sulcus (AS) in the primary motor area. The # indicates the location according to the column of ECoG electrodes.
Figure 2.
Example of measured signals from monkey A during the tasks.
All signals are aligned to trial onset. At the top are frequency feature values of the ECoG signals. Frequency features were resorted in time and sensorimotor rhythm bands. Frequency features of each band are ordered by channel. Below the frequency features are the 12 EMG signals recorded from wire electrodes implanted into muscles of the right forelimb. The blue traces represent original muscle activities. The red traces represent muscle activities obtained by low pass filtering (cut-off frequency: 4 Hz). Below the EMG are grip force on the knob and logical signals, indicating presence of the monkey's hand on the home button or grasping the knob.
Figure 3.
Representative example of predicted and recorded muscle activities.
Dotted lines are actual muscle activities from EMG signals measured with wire electrodes, and solid lines represent predicted muscle activities from ECoG signals of monkey A. The normalized root mean square error (nRMSE) and correlation coefficient (CC) are also shown above each panel.
Table 1.
Summary of prediction accuracies for 10-fold cross validation of monkey B.
Figure 4.
CC and nRMSE distributions for each muscle of monkey A.
The height of each blue bar is equal to the CC density of the interval (0.05). The height of each red bar is equal to the nRMSE density of the interval (0.02). The total area of the histogram is equal to the number of trials used as validation data. Each dotted line with a number shows the median of nRMSE or CC for each muscle. For visualization, we substituted zeros for all negative CC values in validation.
Figure 5.
Example of muscle activity prediction in a continuous time series from monkey B.
Dotted lines are actual muscle activities from EMG signals and solid lines are predicted muscle activities from ECoG signals over a 50 s time interval. Both lines were normalized to the ranges of actual muscle activities. The normalized root mean square error (nRMSE) and correlation coefficient (CC) are also shown.
Figure 6.
Contribution of each frequency band for EMG prediction.
Each panel shows results of multiple comparisons among the frequency bands for each location level (A: monkey A, B: monkey B). Each bar represents the mean of the median weights of each frequency band. At the top of each graph are the sum of the total weights. The # indicates the location of the ECoG electrodes. Noteworthy significant differences between weight values of frequency bands are marked with * (p<0.05) and ** (p<0.001). Other significance comparisons are omitted for visualization purposes.
Figure 7.
Simple main effect of electrode location contributing to EMG prediction.
Each panel shows results of multiple comparisons among the locations for each frequency band level (A: monkey A, B: monkey B). Each marker displays the mean of the median weights. Faded lines show non-significant frequency bands. Solid lines represent a significant difference between weight values (p<0.001) and the dotted lines represent no significance. The # indicates the location of the ECoG electrodes.