Skip to content

FrangakisLab/mcm-cryoet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Mean curvature motion and levelset motion for cryo-ET

Three-dimensional visualization of biological samples is essential for understanding their architecture and function. However, it is often challenging due to the macromolecular crowdedness of the samples and low signal-to-noise ratio of the cryo-electron tomograms. Denoising and segmentation techniques address this challenge by increasing the signal-to-noise ratio and by simplifying the data in images. Here, mean curvature motion is presented as a method that can be applied to segmentation results, created either manually or automatically, to significantly improve both the visual quality and downstream computational handling. Mean curvature motion is a process based on nonlinear anisotropic diffusion that smooths along edges and causes high-curvature features, such as noise, to disappear. In combination with level-set methods for image erosion and dilation, the application of mean curvature motion to electron tomograms and segmentations removes sharp edges or spikes in the visualized surfaces, produces an improved surface quality, and improves overall visualization and interpretation of the three-dimensional images.

Unlike other nonlinear anisotropic diffusion methods the tools presented here do not include contrast dependent parameters and are thus applicable to binary and grayscale images with arbitrary ranges of values or contrast (i.e. can be applied to light-on-dark and dark-on-light tomograms the same way).

If you use these tools, please cite:

Contents:


Examples

Smoothing a manual segmentation with staircase-artifacts and holes

Movie mcm_close.py

Existing manual segmentations in EM/MRC-Format can be smoothed and filled in using mcm_close.py. The tool is conceptually similar to binary morphological closing and gauss filtering, but should yield better results with close to no parameter tuning. The program diffuses the input signal depending on the local image gradient (adjust strength with -a/--alpha) and local mean curvature (adjust strength with -b/--beta).

The volume shown above is a binary segmemtation of size 900 x 700 x 250. The movie was generated using the following parameters (replace ITERNUM with the iteration number of your choice). The --binary option ensures rescaling to data range of [0 1] after diffusion.

# Run mcm_close for ITERNUM iterations.
mcm_close.py -i vol_in.mrc -o vol_out.mrc -p ITERNUM -a 0.5 -b 0.5 --binary

Optionally, it is possible to threshold the output image, to obtain a new binary output:

# Run mcm_close for ITERNUM iterations and threshold
mcm_close.py -i vol_in.mrc -o vol_out.mrc -p ITERNUM -a 0.5 -b 0.5 --binary --threshold 0.5

For non-binary input images, turn off rescaling using --no-binary

# Run mcm_close for ITERNUM iterations, no rescaling
mcm_close.py -i vol_in.mrc -o vol_out.mrc -p ITERNUM -a 0.5 -b 0.5 --no-binary

Reducing noise in a membrane segmentation

Movie mcm_open1.py Movie mcm_open2.py

Noisy membrane segmentations or probability maps can be improved by mcm_open.py. Following a similar principle as in the example above, levelset-based erosion and mean curvature motion are applied first, followed by levelset based dilation coupled with mean curvature motion.

The volume shown above is a simulated noisy segmemtation of size 900 x 700 x 250 with values scaled between 0 and 1. The movie was generated using the following parameters (replace ITERNUM with the iteration number of your choice). The --binary option ensures rescaling to data range of [0 1] after diffusion.

# Run mcm_open for ITERNUM iterations.
mcm_open.py -i vol_in.mrc -o vol_out.mrc -p ITERNUM -a 0.5 -b 0.5 --binary

As above, the tool can also produce thresholded output or proceed without any rescaling.

# Run mcm_open for ITERNUM iterations and threshold
mcm_open.py -i vol_in.mrc -o vol_out.mrc -p ITERNUM -a 0.5 -b 0.5 --binary --threshold 0.5
# Run mcm_open for ITERNUM iterations, no rescaling
mcm_open.py -i vol_in.mrc -o vol_out.mrc -p ITERNUM -a 0.5 -b 0.5 --no-binary

Finding the shortest path through a binary mask

Movie mcm_close.py

The tool geodesic_trace.py allows finding the connection between two points through a binary mask (e.g. a filament segmentation). Filling a mask from a start point X using levelset motion enables recording the path of the diffusion front until an endpoint Y is reached. Tracing back the path of the diffusion from Y to X yields the shortest connection between the points through the mask. This is conceptually similar to pathfinding using Dijkstra's algorithm.

For the sake of consistency, the tool is demonstrated using a membrane mask. The volume shown above is a binary segmemtation of size 900 x 700 x 250. The movie was generated using the following parameters:

# Run geodesic_trace
geodesic_trace.py -i in_mask.mrc -ov out_trace.mrc -ot trace_coords.txt -x 352,222,47 -y 110,44,30

Tools

mcm_3D

Smooth a volume using mean curvature motion only.

Python wrapper

mcm_3D.py -i input.em -o output.em -p 20

Python code

import pymcm.mcm as mcm

outvol = mcm.mcm(invol, iterations=20, verbose=True)

Pure C application

mcm_3D input.em output.em 20

CUDA-accelerated application

mcm_3D_cuda input.em output.em 20


mcm_levelset

Smooth a volume using a combination of mean curvature motion and levelset motion.

Python wrapper

mcm_levelset.py -i input.em -o output.em -p 20 -a 0.5 -b 0.5

Python code

import pymcm.mcm as mcm

outvol = mcm.mcm_levelset(invol, iterations=20, alpha=0.5, beta=0.5, verbose=True)

Pure C application

mcm_levelset input.em output.em 20 0.5 0.5

CUDA-accelerated application

mcm_levelset_cuda input.em output.em 20 0.5 0.5


mcm_open

Morphological opening with levelset and mean curvature motion.

Python wrapper

mcm_open.py -i input.em -o output.em -p 20 -a 0.5 -b 0.5

Python code

import pymcm.mcm as mcm

# Alpha needs to be between 0 and 1
alpha = 0.5
assert(0 <= alpha <= 1)

# Erosion
outvol = mcm.mcm_levelset(invol, iterations=20, alpha=-1*alpha, beta=0.5, verbose=True)
invol = outvol.copy()

# Dilation
outvol = mcm.mcm_levelset(invol, iterations=20, alpha=alpha, beta=0.5, verbose=True)


mcm_close

Morphological closing with levelset and mean curvature motion.

Python wrapper

mcm_close.py -i input.em -o output.em -p 20 -a 0.5 -b 0.5

Python code

import pymcm.mcm as mcm

# Alpha needs to be between 0 and 1
alpha = 0.5
assert(0 <= alpha <= 1)

# Dilation
outvol = mcm.mcm_levelset(invol, iterations=20, alpha=-alpha, beta=0.5, verbose=True)
invol = outvol.copy()

# Erosion
outvol = mcm.mcm_levelset(invol, iterations=20, alpha=-1*alpha, beta=0.5, verbose=True)


geodesic_trace

Finds the shortest geodesic path through a binary mask given a start and end point.

Python wrapper

geodesic_trace.py -i mask.em -ov output_vol.em -ot output_coords -x 10,10,10 -y 30,30,30 -m 10000

Python code

import pymcm.mcm as mcm

outvol, outtrace = mcm.trace(invol, x, y, maxstep=10000, verbose=True)

Pure C application

geodesic_trace mask.em output_vol.em 10 10 10 30 30 30

CUDA-accelerated application

geodesic_trace_cuda mask.em output_vol.em 10 10 10 30 30 30

Installation

TL;DR

# Install C/C++ compilers and optionally CUDA 
# e.g. on Ubuntu
sudo apt install build-essentials
sudo apt install nvidia-cuda-dev nvidia-cuda-toolkit

# Installs all other pre-requisites (a little overkill)
conda create -n mcm -c conda-forge python=3.9 scikit-build numpy mrcfile cython cmake=3.23
conda activate mcm

# Build
git clone https://github.com/FrangakisLab/mcm-cryoet.git
cd https://github.com/FrangakisLab/mcm-cryoet.git
pip install .

# MCM-Levelset combi
mcm_levelset.py --help

# MCM alone
mcm_3D.py --help

# Geodesic trace
geodesic_trace.py --help

The package contains three flavours of the filters:

  1. Pure C programs
  2. C++/CUDA-accelerated versions of the C code.
  3. A python extension and scripts wrapping the C and C++/CUDA programs (recommended)

All can be built and installed independently. The pure C/C++/CUDA applications read data only in EM format, while the python wrappers also accept MRC-files as input and allow setting some additional parameters. If the CUDA-accelerated library is found, it will be preferred by the python wrappers.

Prerequisites

Python package (recommended)

  • CMake >= 3.23
  • C/C++ compiler
  • Python >= 3.9
  • Python packages:
    • skbuild >= 0.15
    • numpy
    • mrcfile
    • Cython
  • optional: CUDA toolkit

Pure C applications

  • CMake >= 3.23
  • C/C++ compiler

CUDA-accelerated applications

  • CMake >= 3.23
  • C/C++ compiler
  • CUDA toolkit

Build

Python package (recommended)

conda create -n mcm -c conda-forge python=3.9 scikit-build numpy mrcfile cython cmake=3.23
conda activate mcm
git clone https://github.com/FrangakisLab/mcm-cryoet.git
cd mcm-cryoet
pip install .

Pure C applications

git clone https://github.com/FrangakisLab/mcm-cryoet.git
cd mcm-cryoet
mkdir build; cd build
cmake ..
make
# Executables now in build/bin/

CUDA-accelerated applications

git clone https://github.com/FrangakisLab/mcm-cryoet.git
cd mcm-cryoet
mkdir build; cd build
cmake ..
make
# Executables now in build/bin/

About

Mean curvature motion for cryo-electron tomograms

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published