Limits...
Improving spatial localization in MEG inverse imaging by leveraging intersubject anatomical differences.

Larson E, Maddox RK, Lee AK - Front Neurosci (2014)

Bottom Line: Specifically, we argue that differences in subject brain geometry yield differences in point-spread functions, resulting in improved spatial localization across subjects.Using a linear minimum-norm inverse to localize this brain activity, we demonstrate that a substantial increase in the spatial accuracy of MEG source localization can result from combining data from subjects with differing brain geometry.Finally, we use a simple auditory N100(m) localization task to show how this effect can influence localization using a recorded neural dataset.

View Article: PubMed Central - PubMed

Affiliation: Institute for Learning and Brain Sciences, University of Washington Seattle, WA, USA.

ABSTRACT
Modern neuroimaging techniques enable non-invasive observation of ongoing neural processing, with magnetoencephalography (MEG) in particular providing direct measurement of neural activity with millisecond time resolution. However, accurately mapping measured MEG sensor readings onto the underlying source neural structures remains an active area of research. This so-called "inverse problem" is ill posed, and poses a challenge for source estimation that is often cited as a drawback limiting MEG data interpretation. However, anatomically constrained MEG localization estimates may be more accurate than commonly believed. Here we hypothesize that, by combining anatomically constrained inverse estimates across subjects, the spatial uncertainty of MEG source localization can be mitigated. Specifically, we argue that differences in subject brain geometry yield differences in point-spread functions, resulting in improved spatial localization across subjects. To test this, we use standard methods to combine subject anatomical MRI scans with coregistration information to obtain an accurate forward (physical) solution, modeling the MEG sensor data resulting from brain activity originating from different cortical locations. Using a linear minimum-norm inverse to localize this brain activity, we demonstrate that a substantial increase in the spatial accuracy of MEG source localization can result from combining data from subjects with differing brain geometry. This improvement may be enabled by an increase in the amount of available spatial information in MEG data as measurements from different subjects are combined. This approach becomes more important in the face of practical issues of coregistration errors and potential noise sources, where we observe even larger improvements in localization when combining data across subjects. Finally, we use a simple auditory N100(m) localization task to show how this effect can influence localization using a recorded neural dataset.

No MeSH data available.


Combining information across subjects results in decreases in both the error in the estimation of centroid location (accuracy), and the point-spread of the data (precision). The centroid error (A–C) was measured by the difference between the true activation location and the centroid calculated from the top V points (averaged across spatial locations for V = 1, 2, 5, 10, 25, 50, or 100 in A; V = 25 for B,C) across different numbers of subjects (1 through 20). The point spread (D–E) was similarly measured as the average distance between the location of true activation and each of the top V spatial locations. Mean ± 2 s.e.m. (across 20,424 cortical locations) is shown by the lines and shaded backgrounds. Note that s.e.m. values are mostly small enough to be masked by the mean lines. The improvement due to combining data across subjects as a function of cortical location is shown in (B,C,E,F). There were large decreases across cortex when comparing 20 subjects to 1 in both the centroid estimation error (C) and the point-spread (F), indicating that the differing subject anatomical structure has reduced the spatial uncertainty and reduced localization error. Note also that the best absolute error values (e.g., some near 1 mm) are only valid for ideal conditions.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4202703&req=5

Figure 2: Combining information across subjects results in decreases in both the error in the estimation of centroid location (accuracy), and the point-spread of the data (precision). The centroid error (A–C) was measured by the difference between the true activation location and the centroid calculated from the top V points (averaged across spatial locations for V = 1, 2, 5, 10, 25, 50, or 100 in A; V = 25 for B,C) across different numbers of subjects (1 through 20). The point spread (D–E) was similarly measured as the average distance between the location of true activation and each of the top V spatial locations. Mean ± 2 s.e.m. (across 20,424 cortical locations) is shown by the lines and shaded backgrounds. Note that s.e.m. values are mostly small enough to be masked by the mean lines. The improvement due to combining data across subjects as a function of cortical location is shown in (B,C,E,F). There were large decreases across cortex when comparing 20 subjects to 1 in both the centroid estimation error (C) and the point-spread (F), indicating that the differing subject anatomical structure has reduced the spatial uncertainty and reduced localization error. Note also that the best absolute error values (e.g., some near 1 mm) are only valid for ideal conditions.

Mentions: With MEG activations for each subject projected onto the average brain, we then sought to quantify the quality of the source localization by measuring both the activity centroid estimation and the point spread when using one subject's or the average of several subjects' activations to localize activity. To approximate the outcome of any given statistical approach—which would need to be selected based on the number of subjects used and the noise structure, among other things, in a non-simulated scenario—we selected the V points with the largest sum of activation magnitudes across subjects. We used different values of V as 1, 2, 5, 10, 25, 50, or 100 vertices as a surrogate for how many points a given statistical test could ideally recover from background noise, which is analogous to choosing a particular threshold on the activity level to declare points “significant.” With these V points selected, we calculated the Euclidean distance from the spatial centroid of these points to the true, original activation location (thereby quantifying the accuracy; see Figures 2A–C), as well as the average distance from each of the V points to the original activation location (thereby quantifying the point spread; see Figures 2D–F). Note that, since we simulate activity at a single source vertex, activation at any other vertex would constitute a false positive according to signal detection theory; our point spread measure thus quantifies the spatial distribution across cortex of the false positives that arise from the inverse solution. For a given desired subject count N (spanning from one to the number of subjects recorded from), the centroid estimation error and point-spread were determined by using the average over 50 random sub-samplings of the subjects of size N (with 50 iterations chosen to reduce computation time while providing sufficiently smooth estimates as a function of space and number of subjects). In our plots of centroid error and point-spread (Figures 2A,C) we have masked the inner portion of the medial wall, since this may contain structures (corpus callosum, midbrain) with unreliable MEG sensitivity and localization.


Improving spatial localization in MEG inverse imaging by leveraging intersubject anatomical differences.

Larson E, Maddox RK, Lee AK - Front Neurosci (2014)

Combining information across subjects results in decreases in both the error in the estimation of centroid location (accuracy), and the point-spread of the data (precision). The centroid error (A–C) was measured by the difference between the true activation location and the centroid calculated from the top V points (averaged across spatial locations for V = 1, 2, 5, 10, 25, 50, or 100 in A; V = 25 for B,C) across different numbers of subjects (1 through 20). The point spread (D–E) was similarly measured as the average distance between the location of true activation and each of the top V spatial locations. Mean ± 2 s.e.m. (across 20,424 cortical locations) is shown by the lines and shaded backgrounds. Note that s.e.m. values are mostly small enough to be masked by the mean lines. The improvement due to combining data across subjects as a function of cortical location is shown in (B,C,E,F). There were large decreases across cortex when comparing 20 subjects to 1 in both the centroid estimation error (C) and the point-spread (F), indicating that the differing subject anatomical structure has reduced the spatial uncertainty and reduced localization error. Note also that the best absolute error values (e.g., some near 1 mm) are only valid for ideal conditions.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4202703&req=5

Figure 2: Combining information across subjects results in decreases in both the error in the estimation of centroid location (accuracy), and the point-spread of the data (precision). The centroid error (A–C) was measured by the difference between the true activation location and the centroid calculated from the top V points (averaged across spatial locations for V = 1, 2, 5, 10, 25, 50, or 100 in A; V = 25 for B,C) across different numbers of subjects (1 through 20). The point spread (D–E) was similarly measured as the average distance between the location of true activation and each of the top V spatial locations. Mean ± 2 s.e.m. (across 20,424 cortical locations) is shown by the lines and shaded backgrounds. Note that s.e.m. values are mostly small enough to be masked by the mean lines. The improvement due to combining data across subjects as a function of cortical location is shown in (B,C,E,F). There were large decreases across cortex when comparing 20 subjects to 1 in both the centroid estimation error (C) and the point-spread (F), indicating that the differing subject anatomical structure has reduced the spatial uncertainty and reduced localization error. Note also that the best absolute error values (e.g., some near 1 mm) are only valid for ideal conditions.
Mentions: With MEG activations for each subject projected onto the average brain, we then sought to quantify the quality of the source localization by measuring both the activity centroid estimation and the point spread when using one subject's or the average of several subjects' activations to localize activity. To approximate the outcome of any given statistical approach—which would need to be selected based on the number of subjects used and the noise structure, among other things, in a non-simulated scenario—we selected the V points with the largest sum of activation magnitudes across subjects. We used different values of V as 1, 2, 5, 10, 25, 50, or 100 vertices as a surrogate for how many points a given statistical test could ideally recover from background noise, which is analogous to choosing a particular threshold on the activity level to declare points “significant.” With these V points selected, we calculated the Euclidean distance from the spatial centroid of these points to the true, original activation location (thereby quantifying the accuracy; see Figures 2A–C), as well as the average distance from each of the V points to the original activation location (thereby quantifying the point spread; see Figures 2D–F). Note that, since we simulate activity at a single source vertex, activation at any other vertex would constitute a false positive according to signal detection theory; our point spread measure thus quantifies the spatial distribution across cortex of the false positives that arise from the inverse solution. For a given desired subject count N (spanning from one to the number of subjects recorded from), the centroid estimation error and point-spread were determined by using the average over 50 random sub-samplings of the subjects of size N (with 50 iterations chosen to reduce computation time while providing sufficiently smooth estimates as a function of space and number of subjects). In our plots of centroid error and point-spread (Figures 2A,C) we have masked the inner portion of the medial wall, since this may contain structures (corpus callosum, midbrain) with unreliable MEG sensitivity and localization.

Bottom Line: Specifically, we argue that differences in subject brain geometry yield differences in point-spread functions, resulting in improved spatial localization across subjects.Using a linear minimum-norm inverse to localize this brain activity, we demonstrate that a substantial increase in the spatial accuracy of MEG source localization can result from combining data from subjects with differing brain geometry.Finally, we use a simple auditory N100(m) localization task to show how this effect can influence localization using a recorded neural dataset.

View Article: PubMed Central - PubMed

Affiliation: Institute for Learning and Brain Sciences, University of Washington Seattle, WA, USA.

ABSTRACT
Modern neuroimaging techniques enable non-invasive observation of ongoing neural processing, with magnetoencephalography (MEG) in particular providing direct measurement of neural activity with millisecond time resolution. However, accurately mapping measured MEG sensor readings onto the underlying source neural structures remains an active area of research. This so-called "inverse problem" is ill posed, and poses a challenge for source estimation that is often cited as a drawback limiting MEG data interpretation. However, anatomically constrained MEG localization estimates may be more accurate than commonly believed. Here we hypothesize that, by combining anatomically constrained inverse estimates across subjects, the spatial uncertainty of MEG source localization can be mitigated. Specifically, we argue that differences in subject brain geometry yield differences in point-spread functions, resulting in improved spatial localization across subjects. To test this, we use standard methods to combine subject anatomical MRI scans with coregistration information to obtain an accurate forward (physical) solution, modeling the MEG sensor data resulting from brain activity originating from different cortical locations. Using a linear minimum-norm inverse to localize this brain activity, we demonstrate that a substantial increase in the spatial accuracy of MEG source localization can result from combining data from subjects with differing brain geometry. This improvement may be enabled by an increase in the amount of available spatial information in MEG data as measurements from different subjects are combined. This approach becomes more important in the face of practical issues of coregistration errors and potential noise sources, where we observe even larger improvements in localization when combining data across subjects. Finally, we use a simple auditory N100(m) localization task to show how this effect can influence localization using a recorded neural dataset.

No MeSH data available.