Skip to content

Latest commit

 

History

History
169 lines (109 loc) · 6.13 KB

README.md

File metadata and controls

169 lines (109 loc) · 6.13 KB

Fetal CP Surface Extraction

Version MIT License ci

Fetal brain cortical plate surface extraction using CIVET marching-cubes (sphere_mesh).

Figure

Abstract

pl-fetal-surface-extract consumes binary volumetric .mnc brain masks to produce surfaces as .obj files with standard connectivity (81,920 triangles). This program is suitable for targeting the inner cortical plate surface (gray-white matter boundary) of high-quality human fetal brain MRI segmentation for subjects between 23-35 gestational weeks of age.

Background

Polygonal surface mesh representations of brain hemispheres are useful for measuring cortical thickness, image registration, and quantitative regional analysis.

Surface Extraction Algorithm

  1. Preprocess mask using mincmorph to fill in disconnected voxels (improve mask quality)
  2. Marching-cubes -> spherical topology surface mesh with unknown number of triangles
  3. Calculate distance error
  4. If maximum distance error is too large, then redo marching-cubes with subsampling
  5. Sphere-to-sphere interpolation -> resample mesh to standard connectivity of 81,920 triangles, preserving morphology
  6. Calculate smoothness error
  7. Run adapt_object_mesh to achieve desired smoothness. Number of iterations is predicted by a regression

Marching Cubes

We have evaluated two implementations of marching-cubes:

We found that CIVET's implementation of the marching-cubes algorithm, sphere_mesh, is more accurate than scikit-image's implementation, which requires FWHM blurring of the volume. Moreover, sphere_mesh guarantees a spherical topology.

How much Smoothing? Estimation

We developed a model which predicts the number of adapt_object_mesh smoothing iterations necessary, given a desired smoothness error threshold. The model was created from the experiment described here:

Pipeline

It is recommended to use the Fetal Brain MRI Surface Extraction ChRIS pipeline.

Installation

pl-fetal-surface-extract is a ChRIS plugin, meaning it can run from either within ChRIS or the command-line.

Usage

extract_cp reads mask files from an input directory and creates the resulting surface files in an output directory.

Input

Input files should be MINC .mnc files representing a mask of the white matter (WM) for a single brain hemisphere (either left or right). WM should be indicated by a value of 1, background value should be 0.

If the input directory contains multiple masks, they will all be processed individually and in parallel.

Options

--inflate_to_sphere_implicit

Arguments to pass to inflate_to_sphere_implicit. The default value (200, 200) should work for fetal brains. While it shouldn't be necessary, increasing the values shouldn't do any harm, and would help compensate for larger brain sizes.

--keep-mask

Copy input mask file to output directory.

--keep-mask is a workaround for using pl-fetal-surface-extract as part of a ChRIS pipeline. It eliminates the need for an extra ts plugin step.

--distance-threshold

Set acceptable maximum distance error.

Without subsampling, "sulci bridge" errors can happen.

Sulci bridge error

This error can be detected by large distance error. If distance error exceeds the distance threshold, then we rerun marching cubes with subsampling enabled.

--target-smoothness

Set target mean smoothness error.

Low quality surface

The result of marching cubes, especially with subsampling enabled, can be poor quality. Taubin smoothing is applied to improve surface quality. The number of smoothing iterations is decided by the value for target smoothness.

Local Usage

To get started with local command-line usage, use Apptainer to run pl-fetal-surface-extract as a container:

apptainer exec docker://fnndsc/pl-fetal-surface-extract extract_cp input/ output/

To print its available options, run:

apptainer exec docker://fnndsc/pl-fetal-surface-extract extract_cp --help

Suggested Software

Before pl-fetal-surface-extract

Development

Instructions for developers.

Building

Build a local container image:

docker build -t localhost/fnndsc/pl-fetal-surface-extract .

Get JSON Representation

Run chris_plugin_info to produce a JSON description of this plugin, which can be uploaded to a ChRIS Store.

docker run --rm localhost/fnndsc/pl-fetal-surface-extract chris_plugin_info > chris_plugin_info.json

Local Test Run

Mount the source code extract_cp.py into a container to test changes without rebuild.

docker run --rm -it --userns=host -u $(id -u):$(id -g) \
    -v $PWD/extract_cp:/opt/conda/lib/python3.10/site-packages/extract_cp:ro \
    -v $PWD/in:/incoming:ro -v $PWD/out:/outgoing:rw -w /outgoing \
    localhost/fnndsc/pl-fetal-surface-extract extract_cp /incoming /outgoing