The author studies the effect of uncertain conductivity on the electroencephalography （EEG） forward problem. A three-layer spherical head model with different and random layer conductivities is considered. Polynomia...The author studies the effect of uncertain conductivity on the electroencephalography （EEG） forward problem. A three-layer spherical head model with different and random layer conductivities is considered. Polynomial Chaos （PC） is used to model the randomness. The author performs a sensitivity and correlation analysis of EEG sensors influenced by uncertain conductivity. The author addressed the sensitivity analysis at three stages： dipole location and moment averaged out, only the dipole moment averaged out, and both fixed. On average, the author observes the least influenced electrodes along the great longitudinal fissure. Also, sensors located closer to a dipole source, are of greater influence to a change in conductivity. The highly influenced sensors were on average located temporal. This was also the case in the correlation analysis. Sensors in the temporal parts of the brain are highly correlated. Whereas the sensors in the occipital and lower frontal region, though they are close together, are not so highly correlated as in the temporal regions. This study clearly shows that intrinsic sensor correlation exists, and therefore cannot be discarded, especially in the inverse problem. In the latter it makes it possible not to specify the conductivities. It also offers an easy but rigorous modeling of the stochastic propagation of uncertain conductivity to sensorial potentials （e.g., making it suited for research on optimal placing of these sensors）.展开更多
文摘The author studies the effect of uncertain conductivity on the electroencephalography （EEG） forward problem. A three-layer spherical head model with different and random layer conductivities is considered. Polynomial Chaos （PC） is used to model the randomness. The author performs a sensitivity and correlation analysis of EEG sensors influenced by uncertain conductivity. The author addressed the sensitivity analysis at three stages： dipole location and moment averaged out, only the dipole moment averaged out, and both fixed. On average, the author observes the least influenced electrodes along the great longitudinal fissure. Also, sensors located closer to a dipole source, are of greater influence to a change in conductivity. The highly influenced sensors were on average located temporal. This was also the case in the correlation analysis. Sensors in the temporal parts of the brain are highly correlated. Whereas the sensors in the occipital and lower frontal region, though they are close together, are not so highly correlated as in the temporal regions. This study clearly shows that intrinsic sensor correlation exists, and therefore cannot be discarded, especially in the inverse problem. In the latter it makes it possible not to specify the conductivities. It also offers an easy but rigorous modeling of the stochastic propagation of uncertain conductivity to sensorial potentials （e.g., making it suited for research on optimal placing of these sensors）.
