After the pre-processing in part 1 Pre Processing (n = 280). We start to build the SuStaIn model.

1. ROI selection

In the table, only ROIs showing significant differences between cases (GDM exposure with obese/overweight child, n = 280) and controls (non-GDM exposure with healthy child, n = 6,149) are displayed, both before and after FDR correction.

For the plot, only ROIs with significant differences between cases (n = 280) and controls (n = 6,149) after FDR correction are shown.

Category Region Before FDR After FDR
Frontal rostral middle frontal ✔️ ✔️
pars triangularis ✔️ ✔️
pars opercularis ✔️ ✔️
superior frontal ✔️ ✔️
pars orbitalis ✔️ ✔️
lateral orbitofrontal ✔️
caudal middle frontal ✔️
Occipital cuneus ✔️ ✔️
Parietal precuneus ✔️ ✔️
superior parietal ✔️ ✔️
Temporal fusiform ✔️
superior temporal ✔️
middle temporal ✔️
Limbic posterior cingulate ✔️
# of significance 14 8

image.png


2. Parameter Setting

Z_val = 1, Z_max = 1.5 ( determined roughly by the 95 percentile of the distribution)

The linear z-score model describes a subtype progression pattern as the linear evolution of biomarkers between different z-scores. Each stage corresponds to a biomarker reaching a new z-score from the set of z-scores for each biomarker, Z_vals. Each biomarker starts with a minimum value of 0 at stage 0 and reaches a maximum of Z_max at the final stage of the progression. The number of stages is determined by the number of biomarkers and z-scores in Z_vals.


3. Results

Subtype 1: Frontal-first, then posterior spread

pars triangularis → rostral middle frontal → pars opercularis → superior frontal → pars orbitalis → precuneus → superior parietal → cuneus

Subtype 2: Posterior-first, then anterior spread

cuneus → superior parietal → precuneus → pars orbitalis → superior frontal → pars opercularis → rostral middle frontal → pars triangularis

Positional Variance Plot, optimal number of subtypes determined by cross-validation.

Positional Variance Plot, optimal number of subtypes determined by cross-validation.

Interpretation from GPT, still need more research…

Subtype 1: Frontal-first, then posterior spread

2. Subtype 2: Posterior-first, then anterior spread