Limits...
Validation of Shared and Specific Independent Component Analysis (SSICA) for Between-Group Comparisons in fMRI

View Article: PubMed Central - PubMed

ABSTRACT

Independent component analysis (ICA) has been widely used to study functional magnetic resonance imaging (fMRI) connectivity. However, the application of ICA in multi-group designs is not straightforward. We have recently developed a new method named “shared and specific independent component analysis” (SSICA) to perform between-group comparisons in the ICA framework. SSICA is sensitive to extract those components which represent a significant difference in functional connectivity between groups or conditions, i.e., components that could be considered “specific” for a group or condition. Here, we investigated the performance of SSICA on realistic simulations, and task fMRI data and compared the results with one of the state-of-the-art group ICA approaches to infer between-group differences. We examined SSICA robustness with respect to the number of allowable extracted specific components and between-group orthogonality assumptions. Furthermore, we proposed a modified formulation of the back-reconstruction method to generate group-level t-statistics maps based on SSICA results. We also evaluated the consistency and specificity of the extracted specific components by SSICA. The results on realistic simulated and real fMRI data showed that SSICA outperforms the regular group ICA approach in terms of reconstruction and classification performance. We demonstrated that SSICA is a powerful data-driven approach to detect patterns of differences in functional connectivity across groups/conditions, particularly in model-free designs such as resting-state fMRI. Our findings in task fMRI show that SSICA confirms results of the general linear model (GLM) analysis and when combined with clustering analysis, it complements GLM findings by providing additional information regarding the reliability and specificity of networks.

No MeSH data available.


Related in: MedlinePlus

Performance of SSICA in partially specific conditions. Left panel shows the reconstruction performance of partially-specific patches as measured by r2 averaged over 600 extracted components (50 permutations × 4 anatomical noise levels × 3 partially-specific patches on average) at each power ratios. Power ratio is defined as the ratio of the simulated patch's SNR value in the opposite group to the matching group. Error bars indicate standard deviation. Blue and red curves show the average r2 when the matching component could be selected from all extracted components, and from just the specific components, respectively. Right panel shows the classification performance of the partially specific patches as the percentage of times in which the SSICA classified the partially- specific patches as a specific component, calculated at different power ratio levels.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC5037228&req=5

Figure 7: Performance of SSICA in partially specific conditions. Left panel shows the reconstruction performance of partially-specific patches as measured by r2 averaged over 600 extracted components (50 permutations × 4 anatomical noise levels × 3 partially-specific patches on average) at each power ratios. Power ratio is defined as the ratio of the simulated patch's SNR value in the opposite group to the matching group. Error bars indicate standard deviation. Blue and red curves show the average r2 when the matching component could be selected from all extracted components, and from just the specific components, respectively. Right panel shows the classification performance of the partially specific patches as the percentage of times in which the SSICA classified the partially- specific patches as a specific component, calculated at different power ratio levels.

Mentions: As described in Section Robustness of SSICA with Regards to the Orthogonally Assumptions, partially-specific components were generated to simulate cases where a brain network is not totally missing within the opposite group (i.e., having rather a small but non-zero power), while it is significantly activated in the matching group. A total of 1000 (50 × 4 × 5) hybrid fMRI datasets were generated by various combinations of patches as shared and partially-specific components (50 random permutations of up to 2 partially-specific patches per group), at different anatomical noise levels (n = 0, 1, 2, 3 voxels), and using different power ratios which specify the relative SNR of the partially-specific components between groups (0.1–0.9 in 0.2 steps). We calculated the reconstruction performance of the partially-specific components at different power ratios by calculating r2 between the corresponding patch and the most similar extracted component either among all the components, or just among the specific ones. Figure 7 left panel shows these results based on the entire extracted components in blue and based on the specific components in red. As expected, SSICA is able to extract the partially-specific components with high accuracy either as shared or specific component (r2 ≈ 0.9 left panel, blue curve). However, only at power ratios ≤0.5, the corresponding extracted component was among the specific components. At power ratios more than 0.5, SSICA labeled the corresponding partially-specific component most often as a shared component. To better quantify the classification performance, the percentage of times in which the SSICA classified the partially-specific patches as specific was calculated (i.e., the number of cases in which the best correlated component with the patch was among the extracted specific components). Figure 7 right panel shows classification performance of the partially-specific components averaged over all hybrid simulated datasets at different relative SNR values. The classification performance curve closely mimics that of the reconstruction performance among the specific components. Overall, this simulation demonstrates that SSICA has the flexibility to extract and label partially specific components as specific when the component's SNR is at least twice larger in one group than in the other. In situations where a component's power is more or less comparable across groups (power ratio between 0.5 and 1), it may be more natural to label this component as shared and then to evaluate differences in the power of shared components across groups.


Validation of Shared and Specific Independent Component Analysis (SSICA) for Between-Group Comparisons in fMRI
Performance of SSICA in partially specific conditions. Left panel shows the reconstruction performance of partially-specific patches as measured by r2 averaged over 600 extracted components (50 permutations × 4 anatomical noise levels × 3 partially-specific patches on average) at each power ratios. Power ratio is defined as the ratio of the simulated patch's SNR value in the opposite group to the matching group. Error bars indicate standard deviation. Blue and red curves show the average r2 when the matching component could be selected from all extracted components, and from just the specific components, respectively. Right panel shows the classification performance of the partially specific patches as the percentage of times in which the SSICA classified the partially- specific patches as a specific component, calculated at different power ratio levels.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC5037228&req=5

Figure 7: Performance of SSICA in partially specific conditions. Left panel shows the reconstruction performance of partially-specific patches as measured by r2 averaged over 600 extracted components (50 permutations × 4 anatomical noise levels × 3 partially-specific patches on average) at each power ratios. Power ratio is defined as the ratio of the simulated patch's SNR value in the opposite group to the matching group. Error bars indicate standard deviation. Blue and red curves show the average r2 when the matching component could be selected from all extracted components, and from just the specific components, respectively. Right panel shows the classification performance of the partially specific patches as the percentage of times in which the SSICA classified the partially- specific patches as a specific component, calculated at different power ratio levels.
Mentions: As described in Section Robustness of SSICA with Regards to the Orthogonally Assumptions, partially-specific components were generated to simulate cases where a brain network is not totally missing within the opposite group (i.e., having rather a small but non-zero power), while it is significantly activated in the matching group. A total of 1000 (50 × 4 × 5) hybrid fMRI datasets were generated by various combinations of patches as shared and partially-specific components (50 random permutations of up to 2 partially-specific patches per group), at different anatomical noise levels (n = 0, 1, 2, 3 voxels), and using different power ratios which specify the relative SNR of the partially-specific components between groups (0.1–0.9 in 0.2 steps). We calculated the reconstruction performance of the partially-specific components at different power ratios by calculating r2 between the corresponding patch and the most similar extracted component either among all the components, or just among the specific ones. Figure 7 left panel shows these results based on the entire extracted components in blue and based on the specific components in red. As expected, SSICA is able to extract the partially-specific components with high accuracy either as shared or specific component (r2 ≈ 0.9 left panel, blue curve). However, only at power ratios ≤0.5, the corresponding extracted component was among the specific components. At power ratios more than 0.5, SSICA labeled the corresponding partially-specific component most often as a shared component. To better quantify the classification performance, the percentage of times in which the SSICA classified the partially-specific patches as specific was calculated (i.e., the number of cases in which the best correlated component with the patch was among the extracted specific components). Figure 7 right panel shows classification performance of the partially-specific components averaged over all hybrid simulated datasets at different relative SNR values. The classification performance curve closely mimics that of the reconstruction performance among the specific components. Overall, this simulation demonstrates that SSICA has the flexibility to extract and label partially specific components as specific when the component's SNR is at least twice larger in one group than in the other. In situations where a component's power is more or less comparable across groups (power ratio between 0.5 and 1), it may be more natural to label this component as shared and then to evaluate differences in the power of shared components across groups.

View Article: PubMed Central - PubMed

ABSTRACT

Independent component analysis (ICA) has been widely used to study functional magnetic resonance imaging (fMRI) connectivity. However, the application of ICA in multi-group designs is not straightforward. We have recently developed a new method named “shared and specific independent component analysis” (SSICA) to perform between-group comparisons in the ICA framework. SSICA is sensitive to extract those components which represent a significant difference in functional connectivity between groups or conditions, i.e., components that could be considered “specific” for a group or condition. Here, we investigated the performance of SSICA on realistic simulations, and task fMRI data and compared the results with one of the state-of-the-art group ICA approaches to infer between-group differences. We examined SSICA robustness with respect to the number of allowable extracted specific components and between-group orthogonality assumptions. Furthermore, we proposed a modified formulation of the back-reconstruction method to generate group-level t-statistics maps based on SSICA results. We also evaluated the consistency and specificity of the extracted specific components by SSICA. The results on realistic simulated and real fMRI data showed that SSICA outperforms the regular group ICA approach in terms of reconstruction and classification performance. We demonstrated that SSICA is a powerful data-driven approach to detect patterns of differences in functional connectivity across groups/conditions, particularly in model-free designs such as resting-state fMRI. Our findings in task fMRI show that SSICA confirms results of the general linear model (GLM) analysis and when combined with clustering analysis, it complements GLM findings by providing additional information regarding the reliability and specificity of networks.

No MeSH data available.


Related in: MedlinePlus