Skip to content
Large scale 3D Neuron Tracing/Neuron reconstruction in Python for 3D microscopic images powered by the Rivulet2 algorithm. Pain-free Install & use in 5 mins. Also works for general centreline-extraction tasks with mhd images.
Python C MATLAB Shell C++
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
filtering Fixed many small bigs during benchmarking Nov 10, 2016
rivuletpy Fix the Soma class by adding pad and save Nov 25, 2019
tests Fixed the swc2vtk for tiff inputs Jul 30, 2019
.gitignore Remove libtiff from source & install from github Mar 11, 2017
LICENSE.txt Update LICENSE.txt Nov 16, 2016
setup.cfg Package for Dec 9, 2016 Update the recommended version for scikit-image to >=0.14.2 Aug 1, 2019 travis setup script Oct 11, 2016 Make Rivuletpy ready for packaging Dec 9, 2016


Example Neuron Tracings

alt text

Example Lung Airway Tracing

alt text

Rivuletpy == Rivulet2

Rivuletpy is a Python3 toolkit for automatically reconstructing single neuron models from 3D microscopic image stacks & other tree structures from 3D medical images.

It is actively maintained and being used in industry scale image analysis applications.

The project was initiated in the BigNeuron project

The rtrace command is powered by the Rivulet2 algorithm published in IEEE Trans. TMI:

[1] S. Liu, D. Zhang, Y. Song, H. Peng and W. Cai, "Automated 3D Neuron Tracing with Precise Branch Erasing and Confidence Controlled Back-Tracking," in IEEE Transactions on Medical Imaging. URL:

PDF []

The predecessor Rivulet1 was published on Neuroinformatics:

[2] Siqi Liu, Donghao Zhang, Sidong Liu, Dagan Feng, Hanchuan Peng, Weidong Cai, "Rivulet: 3D Neuron Morphology Tracing with Iterative Back-Tracking", Neuroinformatics, Vol.14, Issue 4, pp387-401, 2016.

A C++ implementation of the Rivulet2 algorithm is also available in the lastest Vaa3D sources under the Rivulet Plugin (Not yet available in the released build). However you can build Vaa3D easily on Mac/Linux following the Vaa3D wiki carefully.

Issues / questions / pull requests

Issues should be reported to the Rivuletpy github repository issue tracker. The ability and speed with which issues can be resolved depends on how complete and succinct the report is. For this reason, it is recommended that reports be accompanied with a minimal but self-contained code sample that reproduces the issue, the observed and expected output, and if possible, the commit ID of the version used. If reporting a regression, the commit ID of the change that introduced the problem is also extremely valuable information.

Questions are also welcomed in the Rivuletpy github repository issue tracker. If you put on a question label. We consider every question as an issue since it means we should have made things clearer/easier for the users.

Pull requests are definitely welcomed! Before you make a pull requests, please kindly create an issue first to discuss the optimal solution.


0. Setup the Anaconda environment

$ conda create -n riv python=python3.6 anaconda # We tested on 3.5 and 3.6. Other python versions >= 3.4 should also work
$ source activate riv

1. Setup the dependencies

To install rivuletpy with pip, you need to install the following packages manually beforehand since some dependencies of rivuletpy uses them in their setup scripts

  • numpy>=1.8.0
  • scipy>=0.17.0
  • Cython>=0.25.1
  • tqdm-dev
  • libtiff-dev
  • SimpleITK
(riv)$ conda install numpy scipy matplotlib cython tqdm git+
(riv)$ conda install -c simpleitk simpleitk  # Install SimpleITK for the support of load mhd

2. Install Up-to-date Rivuletpy from source

Optionally you can install Rivuletpy from the source files

(riv)$ git clone
(riv)$ cd rivuletpy
(riv)$ python develop # Needed since we use the fast-forward 'tqdm' and 'pylibtiff'
(riv)$ pip3 install -e .

This installs Rivuletpy into your virtualenv in "editable" mode. That means changes made to the source code are seen by the installation. To install in read-only mode, omit the -e.

Test Installation

In ./rivuletpy/ sh

This will download a simple neuron image and perform a neuron tracing with rivulet2 algorithm. If you encountered any issues while installing Rivuletpy, you are welcome to raise an issue for the developers in the issue tracker


  • Reconstruct single neuron file.

The script rtrace command will be installed

$ rtrace --help
usage: rtrace [-h] -f FILE [-o OUT] [-t THRESHOLD] [-z ZOOM_FACTOR]
              [--save-soma] [--no-save-soma] [--speed]
              [--quality] [--no-quality] [--clean] [--no-clean] [--silent]
              [--no-silent] [-v] [--no-view]
              [--tracing_resolution TRACING_RESOLUTION] [--vtk]

Arguments to perform the Rivulet2 tracing algorithm.

optional arguments:
  -h, --help            show this help message and exit
  -f FILE, --file FILE  The input file. A image file (*.tif, *.nii, *.mat).
  -o OUT, --out OUT     The name of the output file
  -t THRESHOLD, --threshold THRESHOLD
                        threshold to distinguish the foreground and
                        background. Default 0. If threshold<0, otsu will be
  -z ZOOM_FACTOR, --zoom_factor ZOOM_FACTOR
                        The factor to zoom the image to speed up the whole
                        thing. Default 1.
  --save-soma           Save the automatically reconstructed soma volume along
                        with the SWC.
  --no-save-soma        Don't save the automatically reconstructed soma volume
                        along with the SWC (default)  
  --speed               Use the input directly as speed image
  --quality             Reconstruct the neuron with higher quality and
                        slightly more computing time
  --no-quality          Reconstruct the neuron with lower quality and slightly
                        more computing time
  --clean               Remove the unconnected segments (default). It is
                        relatively safe to do with the Rivulet2 algorithm
  --no-clean            Keep the unconnected segments
  --silent              Omit the terminal outputs
  --no-silent           Show the terminal outputs & the nice logo (default)
  -v, --view            View the reconstructed neuron when rtrace finishes
  --tracing_resolution TRACING_RESOLUTION
                        Only valid for mhd input files. Will resample the mhd
                        array into isotropic resolution before tracing.
                        Default 1mm
  --vtk                 Store the world coordinate vtk format along with the

Example Usecases with single neurons in a TIFF image

$ rtrace -f example.tif -t 10 # Simple like this. Reconstruct a neuron in example.tif with a background threshold of 10
$ rtrace -f example.tif -t 10 --quality # Better results with longer running time
$ rtrace -f example.tif -t 10 --quality -v # Open a 3D swc viewer after reconstruction 

Example Usecases with general tree structures in a mhd image

$ rtrace -f example.mhd -t 10 --tracing_resolution 1.5 --vtk # Perform the tracing under an isotropic resolution of 1.5mmx1.5mmx1.5mm and output a vtk output file under the world coordinates along side the swc.
$ rtrace -f example.mhd -t 10 --tracing_resolution 1.5 --vtk --speed # Use the input image directly as the source of making speed image. Recommended if the input mhd is a probablity map of centerlines.

Please note that Rivulet2 is powerful of handling the noises, a relatively low intensity threshold is preferred to include all the candidate voxels.

  • Compare a swc reconstruction against the manual ground truth
$ compareswc --help
usage: compareswc [-h] --target TARGET --groundtruth GROUNDTRUTH
                  [--sigma SIGMA]

Arguments for comparing two swc files.

optional arguments:
  -h, --help            show this help message and exit
  --target TARGET       The input target swc file.
  --groundtruth GROUNDTRUTH
                        The input ground truth swc file.
  --sigma SIGMA         The sigma value to use for the Gaussian function in

$ compareswc --target r2_tracing.swc --groundtruth hand_tracing.swc
0.9970 0.8946 0.9865 1 3

The compareswc command outputs five numbers which are in order:

precision, recall, f1-score, No. connection error type A, No. connection error type B


What if I see on Mac OS ImportError: Failed to find TIFF library. Make sure that libtiff is installed and its location is listed in PATH|LD_LIBRARY_PATH|..


brew install libtiff

What if I see ...version `GLIBCXX_3.4.21' not found... when I run rtrace under Anaconda?


(riv)$ conda install libgcc # Upgrades the gcc in your conda environment to the newest

What if I see Intel MKL FATAL ERROR: Cannot load or

Try to get rid of the mkl in your conda, it has been reported to cause many issues

(riv)$ conda install nomkl numpy scipy scikit-learn numexpr
(riv)$ conda remove mkl mkl-service


The build-time and runtime dependencies of Rivuletpy are:

You can’t perform that action at this time.