A Study On Sparse Coding Mechanism Of Task-related Neuronal Population/Ensemble Activity

ObjectiveNeurons in neuronal population participate in functional coding with different levels in different tasks:(1) functional neuronal ensembles play an important role in behavior execution (2) other neurons, who are not involved in the neuronal ensembles, also have task-related activity and take part in functional coding. As recognized, neuronal activity has sparse property. How many neurons are necessary for implanted brain machine interface, if we choose neurons randomly from a large population? In this dissertation, we studied sparse coding mechanism of task-related neuronal population/ensemble activity in two different tasks. Part1study the neuronal ensemble coding in rat prefrontal cortex during working memory task; part2study the neuronal population coding in macaques primary motor cortex in brain machine interface and explore the minimum size of neurons from neuronal population can encode task-related information effectively.Part1:Neuronal ensemble coding in rat prefrontal cortex during working memory task1. Methods:1) Laboratory animals:Male SD rats, weighted300-350g.2) Rats were trained in the Y-maze until the rats can reach80%correct rate in working memory task.3) Rats were implanted16-channel microelectrode arrays in the prefrontal cortex.4) After recovery, neuronal population activity in the prefrontal cortex was recorded when the rats were working in Y-maze working memory task, using the Cerebus data acquisition system.5) Neuronal population action potential space-temporal sequences were obtained from the original signals.6) Neuronal ensemble rate/entropy coding in rats prefrontal cortex during working memory task. 7) From the view of behavioural characteristics of rats, neuronal ensemble rate/entropy coding patterns to study the neuronal ensemble coding mechanism in rat prefrontal cortex during working memory task.2. Results:During the Y-maze working memory task, the neuronal ensemble activity was observed. Neuronal ensemble rate coding and entropy coding were used to analyze the properties of neuronal population action potential space-temporal sequences in rat prefrontal cortex during working memory task. We get the following results:1) Neuronal ensemble rate coding in rat prefrontal cortex during working memory task.In cocrrect trials, the averaged pre-event firing rates (10.8607±1.8176Hz) are higher than the averaged post-event firing rates (8.1714±1.1521Hz)(P<0.05). However, in incorrect trials, the two groups showed no statistically significant (P>0.05).2) Neuronal ensemble entropy coding in rat prefrontal cortex during working memory task.In correct trials, the averaged pre-event entropy values (0.1088±0.0305bit) are higher than the averaged post-event entropy values (0.0862±0.0083bit)(P<0.05). However, in incorrect trials, the two groups showed no statistically significant (P>0.05).Part2:Neuronal population coding in macaques primary motor cortex in brain machine interface1. Methods:1) Laboratory animals:Male macaques (Macaca mulatta), weighted10.1kg and7.9kg, respectively.2) Macaques were trained in brain machine interface. At the same time, neuronal population activity in macaques primary motor cortex was recorded when the monkeys were performing the center-out task, using the Plexon data acquisition system. The behavior results and EMGs were also recorded.3) Neuronal population action potential space-temporal sequences were obtained from the original signals.4) Build up a liner model to decode neuronal signal and realize the direct cursor control by neuronal activity.5) Different sizes of neurons were chosen randomly from population to accomplish the cursor control by neuronal activity and explored how many neurons are necessary for implanted brain machine interface6) Study the directional tuning property of neurons and EMGs activity in brain machine interface.2. Results:1) Behavior resultsIn HC sessions, the two monkeys can achive the task with almost100%correct rate. In BC sessions, the two monkeys can control the cursor on screen by their neuronal activity, even without moving their arms. For Monkey1, in BCWH/BCWOH sessions,40-neuron model has higher correct rate than10-neuron model. For Monkey2, in BCWH sessions, the more neurons are involved, the better the model can work; in BCWOH sessions, only19-neuron model can achive the task. After a long time of training (65days), the correct rate reached only80%.2) Directional neuronal tuningDirectional neuronal tuning property of neurons was analyzed and model neurons kept the stable prefered directions.Prefered direction of each neuron was predicted according to firing rate. The angular bias between predicted and actual angels was calculated. The results show that the more neurons were involved in the model, the larger angular bias was. However, as the training continues, the angular bias decreased gradully.3) Directional tuning of EMGsFor Monkey1, in HC sessions, left triceps and wrist extensor have obvious activity; while in BCWH/BCWOH sessions, the magnitude of EMGs decreased; For Monkey2, in HC sessions, left biceps, triceps and wrist extensor have obvious activity; While in BCWH/BCWOH sessions, the magnitude of EMGs decreased. Moreover, the EMGs also have directional tuning property.Conclusions:We studied sparse coding mechanism of task-related neuronal population/ensemble activity in two different tasks. From the view of behavioural characteristics, neuronal ensemble coding patterns and directional tuning property to study the neuronal ensemble sparse coding mechanism, we get the conclusion as follows:1. During the Y-maze working memory task, the neuronal ensemble activity was observed in correct trials. However, in incorrect trials, the neuronal activity has not changed appreciably. Therefore, neuronal ensembles in rat prefrontal cortex encode the working memory task.2. Neurons in neuronal population participate in functional coding with different levels in different tasks:(1) functional neuronal ensembles play an important role in behavior execution (2) other neurons, who are not involved in the neuronal ensembles, also have task-related activity and take part in functional coding.3. Neuronal activity has sparse property. In this center-out task, we chose40%(19/47) and24%(40/166) neurons randomly from population, which can realize brain machine interface effectively.