Research Article

Assumption-Free Assessment of Corpus Callosum Shape: Benchmarking and Application

Table 1

Case study mixed effects model results.

Sole predictor

VolumeLPC1EPC1

2.411.703.05
(1.88, 2.95)(1.28, 2.12)(2.51, 3.58)
Intercept53.6153.6153.62
(52.91, 54.30)(52.89, 54.32)(52.92, 54.32)
Random effects
 Intercept99.9 (9.99)106.4 (10.31)102.57 (10.13)
 Residual17.86 (4.23)17.3 (4.16)16.91 (4.11)
Observations1,9841,9841,984
AIC13,574.5313,587.3513,530.38

With covariates

VolumeLPC1EPC1

0.820.220.55
(0.07, 1.57)(-0.43, 0.86)(-0.30, 1.40)
Covariates
 Intracranial volume†1.000.931.03
(0.03, 1.98)(-0.06, 1.92)(0.03, 2.03)
 Handedness (right)0.951.041.04
(-1.64, 3.55)(-1.57, 3.64)(-1.56, 3.64)
 Handedness degree-2.19-2.30-2.31
(-4.59, 0.21)(-4.75, 0.15)(-4.72, 0.09)
 Gender (Female)2.312.402.52
(0.36, 4.27)(0.43, 4.37)(0.54, 4.50)
 Age‡0.650.570.46
(-1.66, 2.97)(-1.87, 3.00)(-1.85, 2.77)
 Years of education0.440.440.45
(0.09, 0.78)(0.09, 0.79)(0.10, 0.79)
Intercept51.1451.6651.66
(40.99, 61.29)(40.96, 62.37)(40.96, 62.37)
Random effects
 Intercept7.407.437.43
 Residual3.783.793.78
Observations675675675
AIC4,510.754,515.244,513.51

Note. All predictors have been converted into z scores to allow clear comparison across measures. Volume = total two-dimensional volume of the corpus callosum mid-sagittal slice. LPC1 = position on the first component in tangent shape space derived from Procrustes aligned landmarks. EPC1 = position on the first component in n-dimensional shape space derived from a combination of eFourier and principal components analysis techniques. For fixed effects, values in brackets are 99% confidence intervals. For random effects, values reported are variance and values in brackets are standard deviations. p<0.05, p<0.01, p<0.001. †Intracranial volume was converted to a z score to address model identifiability problems. ‡Age is scaled to years beyond 40 (youngest age in the cohort). Observations are lower for volume analyses due to some failures in Freesurfer processing. Similarly, the sample size is lower when covariates are included due to missingness in those variables.