Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2011 Aug 22:5:13.
doi: 10.3389/fninf.2011.00013. eCollection 2011.

Nipype: a flexible, lightweight and extensible neuroimaging data processing framework in python

Krzysztof Gorgolewski  1 , Christopher D Burns, Cindee Madison, Dav Clark, Yaroslav O Halchenko, Michael L Waskom, Satrajit S Ghosh
Affiliations

Nipype: a flexible, lightweight and extensible neuroimaging data processing framework in python

Krzysztof Gorgolewski et al. Front Neuroinform. .

Abstract

Current neuroimaging software offer users an incredible opportunity to analyze their data in different ways, with different underlying assumptions. Several sophisticated software packages (e.g., AFNI, BrainVoyager, FSL, FreeSurfer, Nipy, R, SPM) are used to process and analyze large and often diverse (highly multi-dimensional) data. However, this heterogeneous collection of specialized applications creates several issues that hinder replicable, efficient, and optimal use of neuroimaging analysis approaches: (1) No uniform access to neuroimaging analysis software and usage information; (2) No framework for comparative algorithm development and dissemination; (3) Personnel turnover in laboratories often limits methodological continuity and training new personnel takes time; (4) Neuroimaging software packages do not address computational efficiency; and (5) Methods sections in journal articles are inadequate for reproducing results. To address these issues, we present Nipype (Neuroimaging in Python: Pipelines and Interfaces; http://nipy.org/nipype), an open-source, community-developed, software package, and scriptable library. Nipype solves the issues by providing Interfaces to existing neuroimaging software with uniform usage semantics and by facilitating interaction between these packages using Workflows. Nipype provides an environment that encourages interactive exploration of algorithms, eases the design of Workflows within and between packages, allows rapid comparative development of algorithms and reduces the learning curve necessary to use different packages. Nipype supports both local and remote execution on multi-core machines and clusters, without additional scripting. Nipype is Berkeley Software Distribution licensed, allowing anyone unrestricted usage. An open, community-driven development philosophy allows the software to quickly adapt and address the varied needs of the evolving neuroimaging community, especially in the context of increasing demand for reproducible research.

Keywords: Python; data processing; neuroimaging; pipeline; reproducible research; workflow.

PubMed Disclaimer

Figures

Figure 1
Figure 1
Architecture overview of the Nipype framework. Interfaces are wrapped with Nodes or MapNodes and connected together as a graph within a Workflow. Workflows themselves can act as a Node inside another Workflow, supporting a composite design pattern. The dependency graph is transformed before being executed by the engine component. Execution is performed by one of the plug-ins. Currently Nipype supports serial and parallel (both local multithreading and cluster) execution.
Figure 2
Figure 2
Simplified hierarchy of Interface classes. An object-oriented design is used to reduce code redundancy by defining common functionality in base classes, and makes adding new interfaces easier and quicker. MatlabCommand, FSLCommand, and FSCommand extend the CommandLine class to provide functionality specific to executing MATLAB, FSL, and FreeSurfer programs. The SPMCommand class defines functions that simplify wrapping SPM functionality. The dashed line indicates that the SPMCommand class uses the MatlabCommand class to execute the SPM matlab scripts generated by the SPM interfaces.
Listing 1
Listing 1
An example interface wrapping the gzip command line tool and a usage example. This Interface takes a file name as an input, calls gzip to compress it and returns a name of the compressed output file.
Listing 2
Listing 2
Part of the input specification for the Brain Extraction Tool (BET) Interface. Full specification covers 18 different arguments. Each attribute of this class is a Traits object which defines an input and its data type (i.e., list of integers), constraints (i.e., length of the list), dependencies (when for example setting one option is mutually exclusive with another, see the xor parameter), and additional parameters (such as argstr and position which describe how to convert an input into a command line argument).
Figure 3
Figure 3
Graph depicting the processing steps and dependencies for a first level functional analysis workflow. Every output–input connection is represented with a separate arrow. Nodes from every subworkflow are grouped in boxes with labels corresponding to the name of the subworkflow. Such graphs can be automatically generated from a Workflow definition and provide a quick overview of the pipeline.
Figure 4
Figure 4
Workflow modification using iterables and MapNodes. If we take the processing pipeline (A) and set iterables parameter of DataGrabber to a list of two subjects, Nipype will effectively execute graph (B). Identical processing will be applied to every subject from the list. Iterables can be used in a graph on many levels. For example, setting iterables on Smooth FWHM to a list of 4 and 8 mm will result in graph (C). In contrast to iterables, MapNode branches within a node of the graph and also merges the results of the branches, effectively performing a MapReduce operation (D).
Figure 5
Figure 5
HTML help page for dtfit command from Camino. This was generated based on the Interface code: description and example was taken from the class docstring and inputs/outputs were list was created using traited input/output specification.
Figure 6
Figure 6
Graph showing the workflow used for the smoothing methods and parameters comparison. The gray shaded nodes have iterables parameter set. This allows easy iteration over all combinations of FWHM and smoothing algorithms used in the comparison.
Figure 7
Figure 7
Influence of different smoothing methods and their parameters. Upper half shows direct influence of smoothing on the EPI sequence (slice 16, volume 0, run 2). Lower half shows indirect influence of smoothing on the T maps (same slice) of the main contrast.
Figure 8
Figure 8
create_spm_preproc() functions returns this reusable, data independent Workflow. It implements typical fMRI preprocessing with smoothing (SPM), motion correction (SPM), artifact detection (Nipype), and coregistration (FreeSurfer). Inputs and outputs are grouped using IdentityInterfaces. Thanks to this, changes in the configuration of the nodes will not break backward compatibility. For full source code see Supplementary Material.
Figure 9
Figure 9
Single subject fMRI Workflow used for benchmarking parallel execution.
See this image and copyright information in PMC

References

    1. Avants B., Gee J. C. (2004). Geodesic estimation for large deformation anatomical shape averaging and interpolation. Neuroimage 23(Suppl. 1), S139–S15010.1016/j.neuroimage.2004.07.010 - DOI - PubMed
    1. Bednar J. A. (2009). Topographica: building and analyzing map-level simulations from Python, C/C++, MATLAB, NEST, or NEURON components. Front. Neuroinform. 3:8.10.3389/neuro.11.008.2009 - DOI - PMC - PubMed
    1. Callahan S. P., Freire J., Santos E., Scheidegger C. E., Silva C. T., Vo H. T. (2006). “VisTrails: visualization meets data management,” in Proceedings of the 2006 ACM SIGMOD International Conference on Management of Data, Chicago, IL, 745–747
    1. Churchill N. W., Oder A., Abdi H., Tam F., Lee W., Thomas C., Ween J. E., Graham S. J., Strother S. C. (2011). Optimizing preprocessing and analysis pipelines for single-subject fMRI. I. Standard temporal motion and physiological noise correction methods. Hum. Brain Mapp. [Epub ahead of print].10.1002/hbm.21238 - DOI - PMC - PubMed
    1. Cointepas Y., Mangin J., Garnero L., Poline J. (2001). BrainVISA: software platform for visualization and analysis of multi-modality brain data. Neuroimage 13, 98.10.1016/S1053-8119(01)91441-7 - DOI

LinkOut - more resources