Using K-means clustering algorithm, researchers have attempted to reconstruct the functional activity of monkey neurons. The study was conducted in collaboration with the University of Florida and the Institute of Neuroscience and Biomedical Research in Spain.
Observation of biological movements
Observation of monkey neuron activation and biological movements has been explored. Neurons in the parietal and ventral premotor cortex respond to observation of goal-directed hand actions or actions observed by another monkey or human.
A neuron’s response to an observed action may vary with the kinematics of the movement, the context of the scene, and the agent of the movement. These factors can affect whether the observed motion interferes with the execution of the movement. Moreover, some aspects of the scene can filter inappropriate reactions.
A previous study found responses to expressive facial features, static faces, and vocalizations. However, the responses to non-goal-directed movements were rarely observed. This may suggest that non-biological observed movements are not used as control stimuli. Instead, cortical motion representations are optimally tuned to kinematic invariants and biological actions.
This study extends the findings of the earlier studies and explores the neural mechanisms of action observation. Specifically, it investigated the activation of mirror neurons, a hypothetical mechanism of action representation. These neurons are located in the ventral premotor cortex (VLPF) and are assumed to be activated when a monkey observes a human or another monkey’s movements. It is thought that this mirror system may help understand the mental states of other monkeys, or even facilitate communication.
To test this hypothesis, a neuron’s response to an observed grasping action was studied. The video was presented to the monkey with black shading over the first phase of the video, obscuring the object and the beginning of the forelimb movement. In the basic condition, the neuron discharged during both epochs of the video, with the strongest discharge occurring prior to the onset of the movement. When the video was obscured during the second phase, the discharge of the neuron decreased, but did not shift to the second epoch. The majority of neurons did not show any motor-related response.
The analysis revealed that the VLPF contains neurons responding to observation of both goal-directed and non-goal-directed actions. The majority of these neurons coded a specific stimulus, such as a monkey grasping an object, while fewer neurons coded objects moving in space.
Observation of object motion
Observation of object motion during monkey neuron activation is rare. Although some neurons have been shown to respond to the movement of a mirror, their response is limited. In the current study, we show that a subset of V6A neurons are activated during passive observation of object motion. These neurons likely code specific aspects of the observed movement.
To assess the nature of this response, we used a polar plot of number of neurons responding to a video stimulus. Each dot represents a neuron, and a vertical scale bar indicates the number of spikes per second. For each dot, a value around zero corresponds to a low selectivity index, and a value of one signifies a neuron that fires only in the most desirable condition. The numbers of neurons are then divided by the number of effective stimuli. This is a simple, two-way ANOVA (see Figure).
The polar plot shows that neurons that fire during the basic condition, discharged during both phases of the video. The obscuration of the video, however, decreased the amount of neuronal activity. The decrease was most pronounced when the object was seen outside of the grasping context.
During grasping, most neurons responded. The HS category was the most active. These neurons are coded to represent the high order representation of the observed action. These cells are particularly sensitive to self-actions and affordance. They also respond to the presence of another agent. In addition, HS cells discharge more strongly when the other agent’s hand grasps the object.
Several of these neurons are selective for Epoch 2. They can code different aspects of the observed movement. These include the spatial location of the moving effector, the type of object, and the interaction between the effector and the object. This may be important in predicting the outcome of the reach-to-grasp task.
The V6A area is responsible for sending signals to the dorsal stream and to other areas of the brain. Some of these signals are motivational. These signals may be processed in the late feedback mode. It has been suggested that the HS neurons code the context of the action and provide the necessary cues to predict its outcome.
K-means clustering algorithm
Optimal clustering of neural activity traces is a computationally NP-hard problem. In this study, we used a K-means clustering algorithm to assign neurons to a cluster based on their activity patterns. For the first step, we allocated two centroids, each representing a data point in the data set. We then performed a for loop for each iteration of the algorithm.
Then, we analyzed individual neurons’ temporal activity patterns. We found that a large fraction of them had a strong correlation to the most common event in the video (running) or the least common event in the video (grooming). The most significant correlation was between running and grooming, with the strongest correlation occurring before movement onset.
Our study was also able to determine the best neuronal selectivity for a limited number of stimuli. We found that the difference between the discharge to the best and worst stimulus was the highest. This is a good indicator of neuronal selectivity for a given stimulus.
We then analyzed the correlation between the corresponding value of the two indexes, the T corr and the selectivity index, and the corresponding values in the resulting plot. The results indicate that the T corr may vary from image field to image field. However, it seems that most neurons have high degrees of preference in both indexes.
The k-means clustering algorithm is useful for a variety of applications. For example, it can be used to group cricket players based on their positions. It also allows for other distance measures. For this study, we used the silhouette function, which provided a measure of cluster separation.
The algorithm was applied to a 3D space, and the resulting clusters were arranged by physical location. Using the squared Euclidean distance, we were able to calculate the silhouette value. It ranged from +1 to -1. This value was the metric of success, with values greater than 0.6 indicating successful clustering.
The k-means clustering process can be performed on many different kinds of data. For this experiment, we used a combination of human and monkey videos. The monkeys were asked to view the videos on a monitor, and they were required to maintain their hand in a resting position.
Functional reconstruction of monkey neuron activation
Using electrical microstimulation, we studied the role of monkey area F5 in grasping. We found that a single neuron could elicit hand movements through the use of somatosensory stimuli. These findings suggest that area F5 has high order hand control, akin to human high order pragmatic control.
Area F5 is located in the rostral part of the ventral premotor cortex, a region involved in the coding of motor goals. We also found that, during grasping with the right hand, area F5 neurons discharge. These findings suggest that monkey area F5 is an active area. However, there are limitations to the study. These include the fact that the somatosensory stimuli are not highly excitable, as was the case with other cortical areas.
A limited number of neurons also showed peculiar behavior. These neurons discharged during the first epoch of video. Their activation was stronger during the initial phase of the movement, which is similar to how the human mirror neurons respond to the visual observation of a single motor act. These neurons were found to code the context of the observed action. This may provide enough cues to predict the outcome of the movement.
These findings indicate that bilateral connections exert a complex role in the perception and control of motor acts. In addition, these results confirm the relevance of bilateral high order hand control.
Our studies suggest that area PFG and area F5 share a common role in the coding of action intention. These areas are also associated with the control of two-hands interaction. This suggests that the coordination of the activity of these areas is necessary for the organization of intentional actions. Moreover, inactivation of area F5 and area PFG would result in the misuse of the ipsilateral hand and neglect of the contralateral space. This inactivation would therefore impact the performance of the motor act.
During the basic condition, most of the neurons show no response. However, this is not the case during the obscuration of the video. In this condition, more than one-third of the neurons responded equally to both motor acts. This suggests that some of these neurons were selective for the first epoch of video.