Limits...
Detecting subject-specific activations using fuzzy clustering.

Seghier ML, Friston KJ, Price CJ - Neuroimage (2007)

Bottom Line: The output identifies subjects who drive activation in specific cortical regions (local effects) or in voxels distributed across neural systems (global effects).In these cases, atypical activations may not be detected by standard tests, under parametric assumptions.The advantage of using FCP is that it searches all voxels systematically and can identify atypical activation patterns in a quantitative and unsupervised manner.

View Article: PubMed Central - PubMed

Affiliation: Wellcome Trust Centre for Neuroimaging, Institute of Neurology, 12 Queen Square, London WC1N 3BG, UK. m.seghier@fil.ion.ucl.ac.uk

ABSTRACT
Inter-subject variability in evoked brain responses is attracting attention because it may reflect important variability in structure-function relationships over subjects. This variability could be a signature of degenerate (many-to-one) structure-function mappings in normal subjects or reflect changes that are disclosed by brain damage. In this paper, we describe a non-iterative fuzzy clustering algorithm (FCP: fuzzy clustering with fixed prototypes) for characterizing inter-subject variability in between-subject or second-level analyses of fMRI data. The approach identifies the contribution of each subject to response profiles in voxels surviving a classical F-statistic criterion. The output identifies subjects who drive activation in specific cortical regions (local effects) or in voxels distributed across neural systems (global effects). The sensitivity of the approach was assessed in 38 normal subjects performing an overt naming task. FCP revealed that several subjects had either abnormally high or abnormally low responses. FCP may be particularly useful for characterizing outlier responses in rare patients or heterogeneous populations. In these cases, atypical activations may not be detected by standard tests, under parametric assumptions. The advantage of using FCP is that it searches all voxels systematically and can identify atypical activation patterns in a quantitative and unsupervised manner.

Show MeSH

Related in: MedlinePlus

(A) Outlier identification with “SPMd” toolbox of Luo and Nichols (2003). (B) Outlier identification with “distance” toolbox of Kherif et al. (2003).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2724061&req=5

fig10: (A) Outlier identification with “SPMd” toolbox of Luo and Nichols (2003). (B) Outlier identification with “distance” toolbox of Kherif et al. (2003).

Mentions: To compare our FCP approach to others, we re-analyzed our data using SPMd (Luo and Nichols, 2003) and the “distance” toolbox provided by Kherif et al. (2003). With respect to Luo and Nichol's method, we used the recent version “spmd2”, which was developed initially for first (within-subject)-level data diagnosis to ensure the stability of fMRI signals over time. We computed different rates following the multi-subject study of Zhang et al. (2006) and compared the “outlier rate” from SPMd with our global G values, as illustrated in Fig. 10A. This demonstrated a number of consistencies between the two approaches. For example, subjects 37, 2, 19, 1, 38 and 29 dominate the activation across voxels, as indicated by our method (Fig. 8). However, we noticed that (i) at the global level, SPMd did not distinguish between outlier effects that are below the mean from those that are above the mean; this may be important when comparing patients to controls, and (ii) there is no quantitative interpretation of this rate, unlike the FCP approach (G values under hypothesis are equal to 1/Nsub, see Fig. 6). SPMd generates a normality diagnostic image based on Shapiro–Wilk statistics. This allows voxels violating normality (outliers) to be identified. Then, individual images are generated to assess how far a subject is from the group mean at a given voxel (equivalent to our U value). The assessment of these images by SPMd is based on the assumption that the population effect is normal. In contrast, our FCP does not assume normality.


Detecting subject-specific activations using fuzzy clustering.

Seghier ML, Friston KJ, Price CJ - Neuroimage (2007)

(A) Outlier identification with “SPMd” toolbox of Luo and Nichols (2003). (B) Outlier identification with “distance” toolbox of Kherif et al. (2003).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2724061&req=5

fig10: (A) Outlier identification with “SPMd” toolbox of Luo and Nichols (2003). (B) Outlier identification with “distance” toolbox of Kherif et al. (2003).
Mentions: To compare our FCP approach to others, we re-analyzed our data using SPMd (Luo and Nichols, 2003) and the “distance” toolbox provided by Kherif et al. (2003). With respect to Luo and Nichol's method, we used the recent version “spmd2”, which was developed initially for first (within-subject)-level data diagnosis to ensure the stability of fMRI signals over time. We computed different rates following the multi-subject study of Zhang et al. (2006) and compared the “outlier rate” from SPMd with our global G values, as illustrated in Fig. 10A. This demonstrated a number of consistencies between the two approaches. For example, subjects 37, 2, 19, 1, 38 and 29 dominate the activation across voxels, as indicated by our method (Fig. 8). However, we noticed that (i) at the global level, SPMd did not distinguish between outlier effects that are below the mean from those that are above the mean; this may be important when comparing patients to controls, and (ii) there is no quantitative interpretation of this rate, unlike the FCP approach (G values under hypothesis are equal to 1/Nsub, see Fig. 6). SPMd generates a normality diagnostic image based on Shapiro–Wilk statistics. This allows voxels violating normality (outliers) to be identified. Then, individual images are generated to assess how far a subject is from the group mean at a given voxel (equivalent to our U value). The assessment of these images by SPMd is based on the assumption that the population effect is normal. In contrast, our FCP does not assume normality.

Bottom Line: The output identifies subjects who drive activation in specific cortical regions (local effects) or in voxels distributed across neural systems (global effects).In these cases, atypical activations may not be detected by standard tests, under parametric assumptions.The advantage of using FCP is that it searches all voxels systematically and can identify atypical activation patterns in a quantitative and unsupervised manner.

View Article: PubMed Central - PubMed

Affiliation: Wellcome Trust Centre for Neuroimaging, Institute of Neurology, 12 Queen Square, London WC1N 3BG, UK. m.seghier@fil.ion.ucl.ac.uk

ABSTRACT
Inter-subject variability in evoked brain responses is attracting attention because it may reflect important variability in structure-function relationships over subjects. This variability could be a signature of degenerate (many-to-one) structure-function mappings in normal subjects or reflect changes that are disclosed by brain damage. In this paper, we describe a non-iterative fuzzy clustering algorithm (FCP: fuzzy clustering with fixed prototypes) for characterizing inter-subject variability in between-subject or second-level analyses of fMRI data. The approach identifies the contribution of each subject to response profiles in voxels surviving a classical F-statistic criterion. The output identifies subjects who drive activation in specific cortical regions (local effects) or in voxels distributed across neural systems (global effects). The sensitivity of the approach was assessed in 38 normal subjects performing an overt naming task. FCP revealed that several subjects had either abnormally high or abnormally low responses. FCP may be particularly useful for characterizing outlier responses in rare patients or heterogeneous populations. In these cases, atypical activations may not be detected by standard tests, under parametric assumptions. The advantage of using FCP is that it searches all voxels systematically and can identify atypical activation patterns in a quantitative and unsupervised manner.

Show MeSH
Related in: MedlinePlus