Effect of Spatial Smoothing on Task fMRI ICA and Functional Connectivity

Zikuan Chen, Vince D. CalhounFrontiers in Neuroscience 2018 · DOI

TL;DR

This paper investigates how spatial smoothing, a common preprocessing step in fMRI, affects the analysis of task-based brain activity using Independent Component Analysis (ICA). They found that while smoothing generally improves signal-to-noise, it has differential effects on single-subject vs. multi-subject analyses, impacting task identification, activation volume, and functional connectivity. The authors suggest optimal smoothing ranges for each scenario to balance these effects.

What Problem Does This Paper Address?

Functional Magnetic Resonance Imaging (fMRI) is a powerful tool for understanding brain activity. A standard step in processing fMRI data is spatial smoothing, typically using a Gaussian kernel, to reduce noise and improve the signal-to-noise ratio (SNR). However, the extent to which this smoothing impacts subsequent analyses, particularly those involving complex decomposition techniques like Independent Component Analysis (ICA) for task-based studies, is not fully understood. ICA is used to identify distinct brain networks or components from fMRI data. For task fMRI, it's crucial to not only identify these components but also to understand how they relate to the task itself (task identification) and how they interact with each other (functional connectivity). This paper aims to systematically explore the effects of varying degrees of spatial smoothing on these key aspects of task fMRI analysis using ICA, comparing results from single-subject and multi-subject datasets.

Methods

The core methodology involves applying ICA to task fMRI data that has undergone varying levels of spatial smoothing.

  1. ICA Decomposition: Task fMRI data was decomposed into independent components (spatial maps and timecourses) using ICA. For multi-subject analysis, a group ICA approach was used.
  2. Task Identification: To identify components related to the task, the timecourses of the ICA components were correlated with the experimental task paradigm's timecourse. A high correlation (e.g., > 0.9) indicated successful task identification.
  3. Task Activation Volume Measurement: For the identified task component, the volume of significant activation at the task response foci was measured.
  4. Functional Connectivity (FC) Calculation: ICA components were ordered based on their correlation with the task timecourse. The functional connectivity matrix was then calculated by computing the temporal correlations between the timecourses of these ordered components.
  5. Spatial Correlation (scorr) Calculation: Similar to FC, a spatial correlation matrix was computed between the spatial maps of the ICA components.
  6. Spatial Smoothing Variation: The fMRI data was preprocessed with Gaussian smoothing kernels of varying Full Width at Half Maximum (FWHM) values, ranging from 1 mm to 35 mm. This was done for both single-subject and multi-subject datasets.

The study specifically compared the effects of smoothing on single-subject ICA versus multi-subject (group) ICA.

Datasets

The study utilized two types of datasets:

  1. Single-Subject Dataset: One isotropic dataset acquired from a single healthy subject (age 44, male) performing a finger-tapping task. The voxel size was 3x3x3 mm³.
  2. Multi-Subject Dataset: A group of 20 healthy subjects (age 38 ± 10, 13 males) who also performed a finger-tapping task. These datasets were anisotropic with voxel sizes of 3.75x3.75x4.55 mm³.

Both datasets were acquired using a 3T Siemens TrioTim scanner at the Mind Research Network (MRN). The study used existing datasets, not newly introduced benchmarks.

Key Insights and Ideas

  • Differential Impact: Spatial smoothing doesn't affect all aspects of ICA analysis equally, and its impact differs significantly between single-subject and multi-subject analyses. This highlights the importance of considering the analysis strategy when choosing smoothing parameters.
  • Task Identification vs. Activation Volume: Smoothing generally decreases the accuracy of identifying the primary task component in single-subject ICA but increases the measured task activation volume, especially in multi-subject ICA. This suggests a trade-off between precise component identification and capturing the extent of brain activation.
  • Functional Connectivity Enhancement: Spatial smoothing appears to strengthen functional connectivity (FC) more in single-subject ICA than in multi-subject ICA. This could be due to smoothing enhancing the temporal correlations between components in individual brains.
  • Correlation Scale Invariance: The paper touches upon correlation scale invariance theory, suggesting that smoothing can influence correlation measures by altering noise levels and signal strengths, which is particularly relevant for FC and spatial overlap measures.
  • Optimal Smoothing Ranges: The study provides specific recommendations for FWHM smoothing: 2-3 voxels (approx. 6-9 mm) for single-subject ICA to achieve optimal FC balance, and a wider range of 2-5 voxels (approx. 6-15 mm) for multi-subject ICA.

Key Results

  • Task Extraction Performance: For single-subject ICA, task extraction performance (measured by maximal task correlation) decreased with increasing FWHM smoothing. For multi-subject ICA, it increased slightly.
  • Task Activation Volume: Task activation volume (taskV) increased with increasing FWHM smoothing for both single-subject and multi-subject ICA. This effect was more pronounced in multi-subject ICA, with taskV tending to saturate around 20 mm FWHM (~6 voxels).
  • Functional Connectivity (FC):
    • The strength of functional connectivity (std(FC)) generally increased with larger FWHM smoothing for single-subject ICA.
  • The balance of functional connectivity (minimal mean(FC)) was achieved with 2-3 voxel FWHM smoothing for single-subject ICA.
  • Spatial smoothing had a lesser effect on the FC matrices of multi-subject ICA compared to single-subject ICA.
  • Spatial Overlap (scorr): Spatial smoothing showed no obvious effects on the spatial overlap of ICA components, with scorr values generally remaining small (< 0.2).

Limitations and Future Work

  • Incomplete Understanding of FC Changes: The authors acknowledge that the precise mechanisms by which spatial smoothing alters FC matrices via ICA are not fully understood. They note that while smoothing changes noisy structures in raw images, leading to variations in ICA decomposition and thus different FC matrices, the theoretical study of these interactions is ongoing.
  • Theoretical vs. Empirical: While correlation scale invariance theory provides a framework, its complete explanation for the observed effects, particularly on FC, requires further investigation.
  • Future Research: The authors suggest that theoretical studies on the interactions among spatial smoothing, ICA, FC, and group averaging are an ongoing research topic. They also highlight the growing trend towards individual fMRI studies, for which understanding optimal preprocessing steps like spatial smoothing is crucial.

Summary updated

Abstract

Spatial smoothing is a widely used preprocessing step in functional magnetic resonance imaging (fMRI) data analysis. In this work, we report on the spatial smoothing effect on task-evoked fMRI brain functional mapping and functional connectivity. Initially, we decomposed the task fMRI data into a collection of components or networks by independent component analysis (ICA). The designed task paradigm helps identify task-modulated ICA components (highly correlated with the task stimuli). For the ICA-extracted primary task component, we then measured the task activation volume at the task response foci. We used the task timecourse (designed) as a reference to order the ICA components according to the task correlations of the ICA timecourses. With the re-ordered ICA components, we calculated the inter-component function connectivity (FC) matrix (correlations among the ICA timecourses). By repeating the spatial smoothing of fMRI data with a Gaussian smoothing kernel with a full width at half maximum (FWHM) of {1, 3, 6, 9, 12, 15, 20, 25, 30, 35} mm, we measured the spatial smoothing effects. Our results show spatial smoothing reveals the following effects: (1) It decreases the task extraction performance of single-subject ICA more than that of multi-subject ICA; (2) It increases the task volume of multi-subject ICA more than that of single-subject ICA; (3) It strengthens the functional connectivity of single-subject ICA more than that of multi-subject ICA; and (4) It impacts the positive-negative imbalance of single-subject ICA more than that of multi-subject ICA. Our experimental results suggest a 2~3 voxel FWHM spatial smoothing for single-subject ICA in achieving an optimal balance of functional connectivity, and a wide range (2~5 voxels) of FWHM for multi-subject ICA.
BibTeX
@inproceedings{chen2018effect,
  author = {Zikuan Chen and Vince D. Calhoun},
  title = {{Effect of Spatial Smoothing on Task fMRI ICA and Functional Connectivity}},
  booktitle = {Frontiers in Neuroscience},
  year = {2018},
  doi = {10.3389/fnins.2018.00015},
}
Figure