The Brain MRI thread (what you can do/measure from a brain MRI and how to analyze)

#1

eg BPF fraction is one impt metric, but there ae others

It’s now trivially easy to calculate (as well the size of all ur brain regions) if you have a MRI and just use the right coding agents to autocalculate…

B/c you can do MRIs frequently, you can also track this over time as a way of doing interventions

====

[there’s an impt between no-threshold and 50% threshold which may move the percentages a few pts]

“Different segmentation methods produce BPF differences of 3.8-17%”

“There is NO consensus gold standard for BPF calculation”

“BPF values CANNOT be directly compared between studies using different methods”

most published norms use SIENAX or SPM. These methods can differ by up to 17 percentage points due to different:

  • Tissue classification thresholds
  • Brain edge definitions
  • CSF inclusion criteria
#2

From a MRI scan you can also calculate…

What are Cortical Layers?

Cortical layers (L1, L2/3, L4, L5, L6) are microscopic structures in the gray matter:

  • Layer I (Molecular): ~100μm thick
  • Layer II/III (Granular/Pyramidal): ~300-500μm
  • Layer IV (Internal Granular): ~200-300μm
  • Layer V (Pyramidal): ~300-500μm
  • Layer VI (Multiform): ~400-600μm

Your MRI resolution: 1mm (1000μm) voxels

Problem: You cannot resolve 100-500μm structures with 1000μm voxels. It’s like trying to read fine print with blurry vision.

What I CAN Calculate

1. Total Cortical Thickness:

2. Depth-Dependent Thickness (Superficial vs Deep): I can measure if your superficial cortex (near pial surface) is thinner than deep cortex (near white matter).

3. Regional Thickness (by lobe/region):

  • Frontal cortex thickness
  • Parietal thickness
  • Temporal thickness
  • Occipital thickness

4. Specific Region Thickness:

  • dlPFC thickness
  • Visual cortex thickness
  • Motor cortex thickness

===

You can also ask the AI to calculate PVS spaces and WMH for you just from the MRI

I have extremely high WM to GM ratio…
0.64
GM:WM Ratio

#3

if you want fMRI, cognitivefxusa.com

1 Like
#4

And there’s an actual prediction available from your own numbers. Mota & Herculano-Houzel (Science, 2015) found a universal cortical folding law across mammals: A_total × √T scales with A_exposed^1.305. Thinner cortex folds more for a given area. Your thickness is slightly low and your surface area estimated slightly high — both push the same direction. So the law predicts your GI runs somewhat above average. Moderate-to-low confidence; it’s a chain of three estimates.

Pial surface = the GM/CSF boundary, the outer edge of cortex. Cortical thickness is the distance between them.

So to your question: “surface area” in FreeSurfer means the white surface by default — SurfArea in aparc.stats, WhiteSurfArea in the header. The Glahn paper says it explicitly: “Area was measured at the white surface.” Which means the trait that correlates with cognition, and that’s genetically independent of thickness, is geometrically the surface of your white matter. Not a gray matter measurement at all, despite being called cortical surface area.

Pial area is the other one, and it’s larger — roughly 10–20% larger in humans, because gyral crowns are convex (pial sits outside on a bigger radius) and that outweighs the sulcal fundi where it goes the other way.

why surface area has ICC ≈ 0.70 across scanners while thickness has ≈ 0.50.

Move the white surface outward by δ everywhere:

  • Thickness changes by exactly δ. Full transmission of the error.
  • Area changes by δ × (local mean curvature). Gyral crowns gain area, sulcal fundi lose it — and on a heavily folded closed surface, mean curvature integrates to near zero. The error largely cancels.

Same boundary, same displacement, and the two measures inherit it completely differently. Thickness is a difference between two surfaces, so errors add; area is an integral over one surface, so errors partially annihilate.

#5

Consequence: almost everything people want thickness to correlate with — intelligence, education, expertise — loads on surface area or total volume instead. The brain-size/IQ correlation (r ≈ 0.2ish, Pietschnig meta-analysis) is mostly area. Thickness-specific IQ correlations are small enough that Marek et al.'s 2022 Nature paper on brain-wide association studies is the relevant frame: typical effect sizes r ≈ 0.01–0.1, needing thousands of subjects to stabilize. So “robustly correlates” and “usefully correlates” come apart here. UK Biobank will find you a significant thickness–anything association. It will explain 0.4% of variance.

Two things I’d call genuinely robust and genuinely interesting

Mechanics. Mota & Herculano-Houzel (2015) found that cortical folding across mammals follows a scaling law where total area × √thickness relates to exposed area with a fixed exponent — folding as a thin-plate buckling problem. Thicker plates buckle at longer wavelength, so thick cortex is smooth cortex. This holds across four orders of magnitude of brain mass and, from what I recall, within humans too. That’s the most law-like correlate of thickness I know of, and it’s physics, not psychology.

The cytoarchitectural gradient. Thickness anticorrelates strongly with intracortical myelin (T1w/T2w ratio) along the sensory-to-association axis. Thin, heavily myelinated primary sensory cortex at one pole; thick, lightly myelinated paralimbic at the other. Very reproducible, and it lines up with Margulies’ principal connectivity gradient — which is a nice hint that thickness is telling you about hierarchical position, not about the person.