Cerebral modeling and dynamic Bayesian networks

摘要: The understanding and the prediction of clinical outcomes focal or degenerative cerebral lesions, as well assessment rehabilitation procedures, necessitate knowing substratum cognitive sensorimotor functions. This is achieved by activation studies, where subjects are asked to perform a specific task while data their brain functioning obtained through functional neuroimaging techniques. Such animal experiments, have shown that functions offspring activity large-scale networks anatomically connected regions. However, no one-to-one correspondence between activated can be found. Our research aims at how derives from information processing mechanisms, which only explain apparently conflicting data. work falls crossroads interpretation techniques computational neuroscience. Since knowledge in neuroscience permanently evolving, our more precisely defining new modeling formalism building flexible simulator, allowing quick implementation models, for better images. It also providing plausible level networks, mechanisms humans. In this paper, we propose formalism, based on dynamic Bayesian (DBNs), respects following constraints: an oriented, networked architecture, whose nodes (the structures) all different, causality-the structure caused upstream nodes' activation-the explicit representation different time scales (from 1ms many seconds PET scan image acquisition), integrated neuronal populations, imprecision data, nonlinearity uncertainty brain's plasticity (learning, reorganization, modulation). One main problems, nonlinearity, has been tackled thanks extensions Kalman filter. capabilities formalism's current version illustrated phoneme categorization process, explaining activations normal dyslexic subjects.