Limits...
MEG and fMRI Fusion for Non-Linear Estimation of Neural and BOLD Signal Changes.

Plis SM, Calhoun VD, Weisend MP, Eichele T, Lane T - Front Neuroinform (2010)

Bottom Line: Neural activity estimates were derived using a dynamic Bayesian network with continuous real valued parameters by means of a sequential Monte Carlo technique.The highly non-linear model of the BOLD response poses a difficult inference problem for neural activity estimation; computational requirements are also high due to the time and space complexity.We show that joint analysis of the data improves the system's behavior by stabilizing the differential equations system and by requiring fewer computational resources.

View Article: PubMed Central - PubMed

Affiliation: The Mind Research Network Albuquerque, NM, USA.

ABSTRACT
The combined analysis of magnetoencephalography (MEG)/electroencephalography and functional magnetic resonance imaging (fMRI) measurements can lead to improvement in the description of the dynamical and spatial properties of brain activity. In this paper we empirically demonstrate this improvement using simulated and recorded task related MEG and fMRI activity. Neural activity estimates were derived using a dynamic Bayesian network with continuous real valued parameters by means of a sequential Monte Carlo technique. In synthetic data, we show that MEG and fMRI fusion improves estimation of the indirectly observed neural activity and smooths tracking of the blood oxygenation level dependent (BOLD) response. In recordings of task related neural activity the combination of MEG and fMRI produces a result with greater signal-to-noise ratio, that confirms the expectation arising from the nature of the experiment. The highly non-linear model of the BOLD response poses a difficult inference problem for neural activity estimation; computational requirements are also high due to the time and space complexity. We show that joint analysis of the data improves the system's behavior by stabilizing the differential equations system and by requiring fewer computational resources.

No MeSH data available.


Estimation RMS error for BOLD (top) and neural activity (bottom) signals. Each point on the plot is an average of 1000 simulations each with its own ground truth. Neural activity becomes denser from left to right. To provide an idea what the signals look like the central panel shows a random sample of three true input and output signals 50 s in duration for each density. Data is available only at the markers (circles: fusion, squares: fMRI) – lines are for visual guides only. Estimates of both methods are compared to the absolute value of neural activity. fMRI estimates are shifted by the out of sequence parameter of −3 s to remove the harmful effect of incorrectly identified delay (Figure 5A).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2991230&req=5

Figure 8: Estimation RMS error for BOLD (top) and neural activity (bottom) signals. Each point on the plot is an average of 1000 simulations each with its own ground truth. Neural activity becomes denser from left to right. To provide an idea what the signals look like the central panel shows a random sample of three true input and output signals 50 s in duration for each density. Data is available only at the markers (circles: fusion, squares: fMRI) – lines are for visual guides only. Estimates of both methods are compared to the absolute value of neural activity. fMRI estimates are shifted by the out of sequence parameter of −3 s to remove the harmful effect of incorrectly identified delay (Figure 5A).

Mentions: Figures 6 and 7 use a fixed neural activity shape across all experiments. To study the effect of the input signal shape on the estimation, we have constructed a simulation in which signals of different density were generated. As previously, we are following the approach of Riera et al. (2004) by placing RBFs of equal standard deviation (σ = 0.02 s) at fixed equally spaced positions and weighting them. Now however we select weights from a zero-mean unit-variance Gaussian distribution (thus allowing negative weights) for a fixed number of RBFs keeping the rest 0. This allows us to control for the amount of neural activity. Figure 8 shows results in 1000 runs for each density level. For a total number of 70 allowed RBFs we have varied RBFs with non-zero weight from 10 to 70 with a step of 10. Resulting neural and corresponding BOLD signals typical for each density are shown in the middle pane. Each point on the figure is a root mean squared error (RMSE) normalized by the power of the true value and averaged over 1000 runs:


MEG and fMRI Fusion for Non-Linear Estimation of Neural and BOLD Signal Changes.

Plis SM, Calhoun VD, Weisend MP, Eichele T, Lane T - Front Neuroinform (2010)

Estimation RMS error for BOLD (top) and neural activity (bottom) signals. Each point on the plot is an average of 1000 simulations each with its own ground truth. Neural activity becomes denser from left to right. To provide an idea what the signals look like the central panel shows a random sample of three true input and output signals 50 s in duration for each density. Data is available only at the markers (circles: fusion, squares: fMRI) – lines are for visual guides only. Estimates of both methods are compared to the absolute value of neural activity. fMRI estimates are shifted by the out of sequence parameter of −3 s to remove the harmful effect of incorrectly identified delay (Figure 5A).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2991230&req=5

Figure 8: Estimation RMS error for BOLD (top) and neural activity (bottom) signals. Each point on the plot is an average of 1000 simulations each with its own ground truth. Neural activity becomes denser from left to right. To provide an idea what the signals look like the central panel shows a random sample of three true input and output signals 50 s in duration for each density. Data is available only at the markers (circles: fusion, squares: fMRI) – lines are for visual guides only. Estimates of both methods are compared to the absolute value of neural activity. fMRI estimates are shifted by the out of sequence parameter of −3 s to remove the harmful effect of incorrectly identified delay (Figure 5A).
Mentions: Figures 6 and 7 use a fixed neural activity shape across all experiments. To study the effect of the input signal shape on the estimation, we have constructed a simulation in which signals of different density were generated. As previously, we are following the approach of Riera et al. (2004) by placing RBFs of equal standard deviation (σ = 0.02 s) at fixed equally spaced positions and weighting them. Now however we select weights from a zero-mean unit-variance Gaussian distribution (thus allowing negative weights) for a fixed number of RBFs keeping the rest 0. This allows us to control for the amount of neural activity. Figure 8 shows results in 1000 runs for each density level. For a total number of 70 allowed RBFs we have varied RBFs with non-zero weight from 10 to 70 with a step of 10. Resulting neural and corresponding BOLD signals typical for each density are shown in the middle pane. Each point on the figure is a root mean squared error (RMSE) normalized by the power of the true value and averaged over 1000 runs:

Bottom Line: Neural activity estimates were derived using a dynamic Bayesian network with continuous real valued parameters by means of a sequential Monte Carlo technique.The highly non-linear model of the BOLD response poses a difficult inference problem for neural activity estimation; computational requirements are also high due to the time and space complexity.We show that joint analysis of the data improves the system's behavior by stabilizing the differential equations system and by requiring fewer computational resources.

View Article: PubMed Central - PubMed

Affiliation: The Mind Research Network Albuquerque, NM, USA.

ABSTRACT
The combined analysis of magnetoencephalography (MEG)/electroencephalography and functional magnetic resonance imaging (fMRI) measurements can lead to improvement in the description of the dynamical and spatial properties of brain activity. In this paper we empirically demonstrate this improvement using simulated and recorded task related MEG and fMRI activity. Neural activity estimates were derived using a dynamic Bayesian network with continuous real valued parameters by means of a sequential Monte Carlo technique. In synthetic data, we show that MEG and fMRI fusion improves estimation of the indirectly observed neural activity and smooths tracking of the blood oxygenation level dependent (BOLD) response. In recordings of task related neural activity the combination of MEG and fMRI produces a result with greater signal-to-noise ratio, that confirms the expectation arising from the nature of the experiment. The highly non-linear model of the BOLD response poses a difficult inference problem for neural activity estimation; computational requirements are also high due to the time and space complexity. We show that joint analysis of the data improves the system's behavior by stabilizing the differential equations system and by requiring fewer computational resources.

No MeSH data available.