Consistency analysis

The consistency analyses of each measure are carried out at the voxel-level within the bundle masks in common space using this consistency script. The individual mask corresponding to each subject and session in the common space are provided as input. The overlap between masks across sessions and subjects is then performed as described in the consistency measures (See section Consistency metrics). Incomplete overlap of mask between subjects and sessions is compensated by densifying each measure in the affected regions voxel-wise, using the average value estimated from the available subjects or sessions. The averaged masks used for the computation of statistical measurements are then obtained subject-wise or session-wise by mathematical union.

Bundle-averaged

For each bundle, a density map is used to generate a binary mask of the whole bundle. To minimize the effect of partial volume, each whole bundle mask was eroded by one voxel to generate a conservative bundle mask that we called the “safe mask”.

Bundle-profile

To generate the bundle-profile (also called track-profiles), Tractometry Flow is applied to each subject-specific bundle to obtain 10 binary mask corresponding to 10 equidistant sections. Each binary mask is then intersected with the safe mask.

../_images/mask_consistency_analysis.png

Pipeline to generate profiles masks for each bundle.

Left and right masks are merged for each average and section bundle mask. Finally, DTI, HARDI, NODDI and MTI measurements are extracted for each average and profile masks for each bundle over session to assess their distribution.

All steps to build the mask used SCIL scripts.

Consistency metrics

Reliability

Image Intra-Class Correlation coefficient (I2C2, Shou et al., 2014), a generalization of the Intra-Class Correlation coefficient (ICC, Koo and Li, 2016, Bruton et al., 2000) to n-dimensional images (one-way random effect, absolute agreement) was used to evaluate the reliability of MRI measurements.

Variability

The variability induced by within-subject and between-subject effects on the measures is quantified using two coefficients of variation per measure: Within- and Between-Variability.

  • Within-Variability (CVw)

CVw is used to evaluate the dispersion of observations when repeatedly measuring a single individual (i.e., reproducibility). It represents the amount of random error or noise contributing to the measure. The CVw is first estimated per subject over their respective imaging sessions and then averaged across session.

  • Between-Variability (CVb)

CVb is used to evaluate the sample heterogeneity. The CVb is obtained by first averaging each subject session-wise, to then estimate the CV over those averages.

Results

Results are displayed with Plotly. The plots are interactives, click on the legend items to select and/or unselect the items.

See section Consistency for results.