fnndsc/ep-surface_fit_parameterized repository overview

⁠surface_fit experiment

ep-surface_fit_parameterized is a ChRIS⁠ plugin for experimenting with the parameters of surface_fit (ASP algorithm from CIVET).

⁠Usage

surface_fit_script.pl is a Perl wrapper for surface_fit. ep_surface_fit(.py) is a Python script for running surface_fit_script.pl as a ChRIS ds-plugin on multiple subjects.

ep_surface_fit processes every laplacian grid (*.mnc) + starting surface (*.obj) pair found in its input directory. For every *.mnc file found, ep_surface_fit will search for a *.obj surface file in the same directory to use as a starting surface.

When multiple inputs are found, they are processed in parallel.

⁠Parameters

Multiple stages of surface_fit can be run by specifying multiple values as a comma-separated list. If some parameter values are given as CSV whereas others are given as singular, the singular value is reused for later iterations. Example:

ep_surface_fit --iter-outer 100,100,400 --stretch-weight 80,60,40 --laplacian-weight 1e-4 ...

The schedule is interpreted as:

1. 100 iterations with sw=80 lw=1e-4
2. 100 iterations with sw=60 lw=1e-4
3. 400 iterations with sw=40 lw=1e-4

⁠--size

Number of triangles in the surface mesh, i.e. resolution

• 20480 improves performance and is more suitable for fetal brains 20-28 GA
• 81920 is standard
• 327680 is used for high-resolution adult human brain

⁠--step-size

Distance to move per iteration.

• large value converges faster
• small value decreases risk of self-intersection

...