image/svg+xml
Both High-resolution and atlas SC maps are impacted by smoothing.
- Stronger & wider smoothing increases reliability.
- Moderate smoothing improves identifiability (2mm FWHM)
- Extensive smoothing reduces individual identifiability (>6mm FWHM)
- Deterministic SC performed better at identification
- Probabilistic SC was comparatively more reliable
Concluding remarks:
- Smoothing is crucial for high-resolution SC (2-6 mm FWHM recommended)
- Smoothing can benefit atlas-resolution SC (4-8 mm FWHM recommended)
- Deterministic SC requires stronger smoothing compared to probabilistic
Diffusion images sourced from the
Human Connectome Project:
- 42 individuals with repeat scans
- Deterministic & probabilistic tractography
Different smoothing parameters
compared based on:
- Raliability (robustness)
- Identifiability (individualised)
Other important considerations:
- Computation & storage burden
HRSC mapped at the resolution of cortical vertices
Connectivity-based spatial smoothing performed
- truncated bivariate Gaussian kernel
All SC maps also downsampled to atlas resolution
Spatial smoothing is a well-recognized preprocessing step that is commonly implemented in a wide range of neuroimaging modalities (fMRI, EEG, fNIRS, PET, etc.). This step is undertaken to increase the signal to noise ratio at the expense of spatial specificity [1].
Structural connectivity (SC) is not normally smoothed because most SC maps are constructed at the resolution of brain atlases comprising broad areal parcels.
Recent studies highlight the benefits of investigating SC higher spatial resolutions:
- High-resolution SC (HRSC) accurately captures intricate neural connections [2]
- HRSC detects local modularities in brain networks [3]
- HRCS perform better in neural fingerprinting and predicting behavior [4]
High-resolution connectomes are susceptible to:
- image registration misalignment, tractography artifacts and noise
This can reduce connectome accuracy and test-retest reliability.
We investigate a network analogue of image smoothing to address these key challenges and investigate the impacts of smoothing on connectome reliability and individual identifiability of SC maps at different resolutions.
a. Department of Biomedical Engineering, The University of Melbourne, Australiab. Melbourne Neuropsychiatry Centre, The University of Melbourne, Australiac. Florey Institute of Neuroscience and Mental Health, Melbourne, Australia
Challenges and impacts of spatial smoothing on high-resolution structural connectomes
Sina Mansour L. a, Caio Seguin b, Vanessa Cropley b, Robert E. Smith c, Andrew Zalesky a,b
Study design
Introduction
Results
42 individuals
Two scans
Diffusion MRI