Motor intention in the posterior parietal cortex: experimental data analysis and functional modeling study
The complexity of processes occurring in the brain is an intriguing issue not just for scientists and medical doctors, but the humanity in general. The cortex ability to perceive and analyze an enormous amount of information in an instance of time, the parallelism and computational efficiency are among the questions that attract attention. Even a simple, everyday gesture, for example, reaching for a cup of coffee, evokes a flow of signals in the brain. It goes from the primary visual region, that locates the cup on the table, to the primary motor region that sends the precise coordinates to the hand, and the instruction what to do next. The sequence of signal transmission and transformation continues through several regions, sensory, associative and motor ones. In this study, we will focus on the posterior parietal cortex, the region involved in the transformation of visual inputs into the preliminary motor plans. The years of experimental work revealed mechanisms for integration of multimodal signals, coordinate transformations, information representation in multiple coordinate frames, and many other. Still, a single encompassing theory about movement generation in the parietal cortex does not exist, and is a matter of debate. This study contributes to the analysis of motor intention in the 7a parietal region. The motor intention, a high-level cognitive signal, is defined as the preliminary plan for making a movement. From the engineering point of view, encoding of motor parameters in the neural activity is extensively studied within the framework of brain-computer interfaces. The motivation behind these studies is the development of neural prosthesis for the paralyzed persons. The direct cortical prosthesis can significantly improve the lives of paralyzed people, who have lost every other contact with the outside world. Also, this framework opens the possibilities for monitoring the neural processes during the execution of natural movements, and studying the mechanisms behind it. In this work, a method for identification of motor intention from the standard recordings of neural activity, the spike trains, is developed. The data of interest was collected in a series of behavioral experiments involving reaching or saccadic eye movements. The presence and absence of motor intention was monitored in various phases of motion execution, and for different types of movements. All the recordings obtained simultaneously are combined in the same decoding session. Therefore, the analysis is done using the activity of small population of cells (typically 8 to 12 cells). We aim to study the motor intention in a general context which requires using activity of multiple cells. The population size is determined by the experimental procedure. Throughout this study we assume that the motor intention can be red from the spike rates, the assumption supported by the neurophysiological studies. Therefore, all the simultaneously collected spike trains are converted into vectors of spike rates. The results of this study show that motor intention can be decoded from the spike rates. A machine-learning based algorithm is developed to analyze the presence or absence of motor intention in the obtained spike rate vectors. This algorithm, based on standard support vector machines, can distinguish between the segments of recordings that encode motor intention, from those that do not encode it. The goal of the study was to examine the precision of motor intention identification, when the activity of a randomly selected set of cells is analyzed using on such algorithm. Additionally, several relevant parameters were tested. The algorithm precision during different phases of movement execution is tested. Also, the influence of the population size and of the procedure for spike rates computation is examined. The obtained results demonstrated that the motor intention can be extracted from the neural signals with the precision of around 70% for a randomly selected set of cells. For the best groups of cells, this precision was 82%. The motor intention identification was particularly precise during the intervals of preparation and realization of saccadic eye movements. This is in accordance with the known functions of the 7a region, where the majority of cells respond to the eye movements. The algorithm precision is determined by the considered population size. For the bigger population the precision increases. Still, this conclusion holds only on average, since adding one or a couple of randomly selected cells does not have to change the result. Randomly selected cells do not necessary carry the information of interest. The influence of each of the cells, present in one set, is tested in this context. The obtained results indicate redundant coding of motor intention in the parietal cortex. Many cells carry the same information, and some of them can be removed from the set without changing the algorithm precision. Still, removing all of them degrades the result. Finally, the influence of the window size, used to compute spike rates in some of the tests is studied. In general, the precision improves when using bigger windows, the result that is consistent with the literature. Introducing the window for computing spike rates enables automatic identification of motor intention, the method suitable for the brain-computer interface applications. Finally, the analysis of the experimental data is complemented with the study of an appropriately designed model. Modeling the biological processes, in order to reveal additional functionality and test some parameters not accessible through the data, is a widely accepted approach. Still, the development of a model, sufficiently simple for implementation on the standard hardware, sufficiently tractable in the simulations, yet informative enough to capture the main processes of interest, is not straightforward. Our motivation for accepting this approach was to test several parameters that imposed themselves as important in the data analysis step. Due to the nature of the problem itself, the test on an approximative model was the only feasible tactic. The influence of the population size and the window size was assessed in this study. This, additionally, demonstrated the algorithm precision scaling as a function of the number of cells.
Keywords: area 7a ; bio-inspired models ; brain-computer interface ; hand-eyes coordination ; liquid state machines ; machine learning ; motor intention ; motor parameters ; neural signals ; population coding ; posterior parietal cortex ; recurrent networks ; spike trains ; spike rates ; support vector machines ; apprentissage automatique ; intention motrice ; codage de population ; coordination main-oeil ; cortex pariétal postérieur ; fréquence d'impulsions ; interfaces cerveau-machine ; machines à état liquide ; modèles bio-inspirés ; paramètres moteurs ; région 7a ; réseaux récurrents ; signal neurologique ; train d'impulsionsThèse École polytechnique fédérale de Lausanne EPFL, n° 4253 (2008)
Programme doctoral Informatique, Communications et Information
Faculté informatique et communications
Institut de systèmes de communication
Laboratoire de systèmes non linéaires
Record created on 2008-10-30, modified on 2016-12-12