MRI image to mesh (based on fsaverage?)

[mdovgialo]

Use FSSurface with parcelation to create material segmentation of FSAverage scan to white matter, grey matter, CSF, skull, skin, air

FSaverage has brain segmentation, but it seems skin and skull surfaces are only available as .surf files. Faces were scrambled, there is also no cavities.

.surf files can be read by MNE and converted to .obj. Pyvista can read .obj and has filters to select points which are inside of the surface.

create scripts to convert such segmented MRI into a mesh made out of cubes with properly assigned materials, and try to simulate something?

Try to create a cube mesh using the 4 spheres sampled materials, and compare 4 spheres conforming mesh and the cube mesh solutions?

Additional information and tools to segment MRI scans into tissues:

https://link.springer.com/article/10.1007/s12021-020-09504-5#rightslink

https://github.com/gtaberna/mrtim/wiki

TODO:
Add far boundary mesh option

[mdovgialo]

First try of decomposing the FSAVERAGE into materials:

[mdovgialo]

Made a first prototype of converting MRI to a 3D mesh with appropriate format conversions of materials and boundary elements grounding plate definition. Untested yet.

Some hacks still to be fixed, also MFEM complains about

Elements with wrong orientation: 16777216 / 16777216 (NOT FIXED)

Not sure what it's about, need to see how it will simulate...

[mdovgialo]

Fixed elements with wrong orientation by first resampling MRI segmentation to RAS instead of weird skewed/inverted thing

First results of forward modeling! First try on lower resolution with 2mm voxels (around 2 million cubes), 1 mm is 16 million...

Head in air with ground plate:

Volume colors - materials
Boundary - yellow grounding plate, green just a border

GT:

Simulated potential:

Higher contrast:

Seems to be working?

[mdovgialo]

Comparison of kCSD vs kESI

Solution parameters

grid step: 5 mm
R_init: 12 mm

1 mm mesh used to simulate potential to electrodes, fsaverage
Electrodes 3 layers of 10-05

brain mask (nodes to place sources at)

ground truth

forward solved potential

kCSD solution

kESI solution using 2mm mesh corrections

difference between kCSD and kESI

Conclusion

kESI and kCSD solutions are remarkably similar, differences are on the order of 0.001%!!!! Maybe worth trying computing a higher resolution solution?

[mdovgialo]

test of high and low resolution, with brain mask and without

high resolution brain mask

low resolution brain mask

GT CSD!!!

Colors changed for easier readability when showing CSD solution and GT

High resolution With brain mask

solution grid step: 1mm
solution basis source radius: 1.2 mm

lambda chosen by CV

kESI

kCSD

kCSD-kESI

ZERO!!! WHAT???

High resolution without brain mask

solution grid step: 1mm
solution basis source radius: 1.2 mm

lambda chosen by CV

kESI

kCSD

kCSD - kESI

ZERO!!! WHAT???

low resolution without brain mask

solution grid step: 5mm
solution basis source radius: 12 mm

lambda chosen by CV

kESI

kCSD

kCSD - kESI

Very low differences!

low resolution with brain mask

solution grid step: 5mm
solution basis source radius: 12 mm

lambda chosen by CV

kESI

kCSD

kCSD - kESI

Very low differences!

[mdovgialo]

The differences between kESI and kCSD are minimal!

Maybe something is wrong with corrections? How corrections are used? Maybe they need normalisation?

[mdovgialo]

Just in case looking at electrode corrections.

Theoretical Potential (no FEM)

Theoretical potential of a point current source at electrode position, with sigma=0.33, without FEM mesh influence

FEM potential

Potential of a point current source at electrode position, calculated over 2mm mesh with realistic materials

FEM correction for kESI

FEM potential minus theoretical