Skip Navigation
Navigation Skipped
Graph of hand position versus time. Explanation to the side.


This graph displays the relationship between the actual real time movements of the lever, which was controlled by the monkey (black line) and the two different types of predictions (linear prediction –redline and ANN prediction-blue line). Over a period of ten seconds the predicted hand position and actual hand position were fairly accurate. This comparison is also made in the following graph.
Graph of correlation versus time. Explanation to the side. The correlation graph shown here demonstrates the relationship between the two types of predictions and their accuracy. On a correlation graph there is the amount of time (X axis) compared to the correlation constant (y-axis). If there was a correlation of 1 this would mean the accuracy was perfect, there was a perfect correlation between actual and predicted. If it was a 0 correlation that would mean there was no relationship, and the movements were completely random. In this case it is observed that it takes a few minutes for the predictions to go from random to the middle range, where it planes out.
Graph displaying real time prediction of hand movement with respect to time.  Explanation to the side. This graph shows that real time movements of the monkey and studying the neurons can be used to make predictions and that those predictions can be used to control a local robot (blue) and a remote robot (pink). Both of these predictions are fairly accurate when compared to the observed movements of the monkey (black).
Graphical representation of the monkey's movements while reaching for the piece of food.  Explanation to the side. This study involved a complex sequence of 3-D hand movements in food-reaching tasks. This task can be divided into (1) reach for food (2) grasp food (3) move food to mouth (4) return to starting position. This movement is shown in the representation above. The dispersion (the total area the all the movements were confined to) had dimensions of 7.0 x 7.5 x 6.0 cm (315 cm3).
Graph of the observed movement of the monkey's hand versus the predicted movement.  Explanation to the side. In this figure, the movement produced by the test subject (the monkey) that was observed is shown in black, and the predicted movement is shown in red. The predictions get better when more data is collected and the outliers are excluded.
Linear model fit for neuron-dropping analysis.  Explanation to the side. The authors also predicted the number of neurons that would need to be recorded in order to obtain a nearly perfect fit between predicted movements (based on brain recordings) and actual movements. To do this, the prediction of movement was re-run several times with one neuron dropped from the analysis each time. For example, when all ~100 neurons recorded in monkey 1 were included, the fit between actual predicted movement was represented by a correlation coefficiend (R2) value of about 0.6. As fewer neurons were used, the correlation coefficient is lowered. When examining the relation between number of neurons used in the analysis and the correlation coefficient, it was found to follow a hyperbolic function. This function then allowed researchers to determine that in order to obtain a 0.90 correlation coefficient, approximately 500 neurons would needed to be employed.
Neuron-dropping curves for each cortical area.  Explanation to the side. This graph demonstrates the number of neurons from different regions of the brain that are utilized in controlling the movement of the robotic arm and what the correlation coefficient for each number of neurons at every location. By this graph one can see that the PMd (left dorsal premotor cortex) has steepest slope, indicating that these neurons are used primarily in controlling movement of the robotic arm. On the other hand, the ipsi MI (ispsilateral motor cortex) has the lowest hyperbola, indicating that these neurons are not as much involved controlling the robotic arm. Ipsi MI contain neurons on the part of the brain that is the same side as the arm doing the movement (normally movement is controlled by the side of the brain opposite to the side it’s controlling).
Estimated neurons needed for linear model fit.  Explanation to the side. This graph simply displays the estimated number of neurons from the different part of the brain that would be needed to obtain a correlation coefficient of 0.90. PMd would need the least amount (approximately 480) neurons to obtain a 90% correlation between neuron and robotic arm movement while iMI would need the most (1,195 neurons) to obtain a correlation coefficient of 0.90. All neurons together would require approximately 500 neurons to obtain the 0.90 coefficient.
Graphs are from Nature, Vol. 408, 361-365. Used with permission from Miguel A. L. Nicolelis
Skip bottom navigational bar and return to top of page.
Group 22 - Home page - Last updated: December 7, 2004
Email webmaster -