Print Close

VB_toolbox: A tool for investigating neural feature gradients in Python and MATLAB

Presented During:

Software Demonstrations

Poster No:

1282

Submission Type:

Abstract Submission

Authors:

Claude Bajada¹^,2^,3, Lucas da Costa Campos²^,4, Svenja Caspers²^,5^,6, Richard Muscat¹, Geoff Parker⁷^,8^,9, Matthew Lambon Ralph¹⁰, Lauren Cloutman³, Nelson Trujillo-Barreto³

Institutions:

¹Department of Physiology and Biochemistry, Faculty of Medicine and Surgery, The University of Malta, Msida, Malta, ²Institute of Neuroscience and Medicine (INM-1), Research Centre Jülich, Jülich, Germany, ³Division of Neuroscience & Experimental Psychology, The University of Manchester, Manchester, United Kingdom, ⁴Institute of Complex Systems (ICS-2), Research Centre Jülich, Jülich, Germany, ⁵Institute for Anatomy I, Medical Faculty, Heinrich-Heine-University Düsseldorf, Düsseldorf, Germany, ⁶JARA-BRAIN, Jülich-Aachen Research Alliance, Jülich, Germany, ⁷Centre for Medical Image Computing, Department of Computer Science, University College London, London, NA, ⁸Queen Square MS Centre, Department of Neuroinflammation, UCL Institute of Neurology, University College London, London, United Kingdom, ⁹Bioxydyn Limited, Manchester, United Kingdom, ¹⁰MRC Cognition and Brain Sciences Unit, University of Cambridge, Cambridge, United Kingdom

First Author:

Claude Bajada
Department of Physiology and Biochemistry, Faculty of Medicine and Surgery, The University of Malta|Institute of Neuroscience and Medicine (INM-1), Research Centre Jülich|Division of Neuroscience & Experimental Psychology, The University of Manchester
Msida, Malta|Jülich, Germany|Manchester, United Kingdom

Co-Author(s):

Lucas da Costa Campos
Institute of Neuroscience and Medicine (INM-1), Research Centre Jülich|Institute of Complex Systems (ICS-2), Research Centre Jülich
Jülich, Germany|Jülich, Germany

Svenja Caspers
Institute of Neuroscience and Medicine (INM-1), Research Centre Jülich|Institute for Anatomy I, Medical Faculty, Heinrich-Heine-University Düsseldorf|JARA-BRAIN, Jülich-Aachen Research Alliance
Jülich, Germany|Düsseldorf, Germany|Jülich, Germany

Richard Muscat
Department of Physiology and Biochemistry, Faculty of Medicine and Surgery, The University of Malta
Msida, Malta

Geoff Parker
Centre for Medical Image Computing, Department of Computer Science, University College London|Queen Square MS Centre, Department of Neuroinflammation, UCL Institute of Neurology, University College London|Bioxydyn Limited
London, NA|London, United Kingdom|Manchester, United Kingdom

Matthew Lambon Ralph
MRC Cognition and Brain Sciences Unit, University of Cambridge
Cambridge, United Kingdom

Lauren Cloutman
Division of Neuroscience & Experimental Psychology, The University of Manchester
Manchester, United Kingdom

Nelson Trujillo-Barreto
Division of Neuroscience & Experimental Psychology, The University of Manchester
Manchester, United Kingdom

Introduction:

There has been an increasing interest in "gradient analysis". Although the technique has been used in the neuroimaging literature since Johansen-Berg et al. (2004), a recent surge in interest occurred when Margulies et al. (2016) embedded the default mode network within a gradient of macroscopic cortical organisation. Gradient analyses in the literature rely on spectral graph theory and the eigendecomposition of the graph laplacian. The second smallest eigenpair of this matrix represents the principal gradient of similarity. In this abstract we introduce a new toolbox built in Python and MATLAB for carrying out gradient analysis using a simple command line interface. The toolbox performs gradient analyses on cortical surfaces and is able to perform them at a whole brain level, using ROI approaches, or a searchlight across the cortex (Kriegeskorte et al. 2006).

Methods:

The program extracts the principal similarity gradient across the cortical surface extracting the second smallest eigenpair. The eigenvector is used to visualise the gradients while the eigenvalue is used to quantify the "gradedness" of the feature. Three types of maps can be extracted from the program. The first is a whole-brain gradient map, the second is a set of gradient maps restricted to regions of interest (ROIs) (Yeo et al. 2011) and the associated eigenvalue for the ROIs. The final map is a searchlight analysis extracting the eigenvalue of nearest neighbouring vertices. This approach detects feature edges across the cortex and is similar in concept to the regional homogeneity (ReHo) approach (Zang et al. 2004). The code was tested on two sets of data. A synthetic set where the structure is known, and a single preprocessed resting state fMRI dataset from the Adult Human Connectome Project. To generate the surrogate data, we performed k-means clustering over the vertices of the inflated brain mesh. We replicated a randomly chosen fMRI time series on the vertex of each cluster, the data being identical in the vertices of each cluster, and distinct between. The toolbox is available in MATLAB and Python (github.com/VBIndex). Both implementations of the toolbox were parallelized in a similar manner: The searchlight and ROI analyses were parallelized using a divide-and-conquer approach, where several threads are spawned, each responsible for a disjoint set of voxels or ROIs, depending on the specific analyses to be carried out. The full brain analyses was parallelized using parallel eigenproblem solvers, available in the LAPACK packages bundled with MATLAB and Anaconda. The Python implementation of the toolbox is recommended for users. It can be installed using `pip install vb_toolbox`. The analyses were performed using both packages, with differences within numerical precision. For consistency, all the results we show were calculated using the Python version. The program ran on 2 x Intel CPUs E5-2690 @ 2.90GHz (16 physical cores, 32 logical cores) 20 x 16GB DDR3 ECC Memory, 1600 MHz running Debian 9.8 and Anaconda 2019.10 build py37_0.

Results:

The code ran the analysis using up to 32 threads in a reasonable time. The full brain gradient took 35 min using 26.4 GB RAM. The clustered data took 1 min 48 s using 2068 MB RAM, the searchlight analysis completed in 24 s using 703 MB of RAM. These analyses can be done on commercially available computers. Fig. 1 shows the results from synthetic data. Fig. 2 shows the result for the real HCP dataset.

Supporting Image: OHBM-FIg_1-Submission.png

Supporting Image: OHBM-FIg_2-Submission.png

Conclusions:

We have presented a new, easy to use and install, toolbox for producing gradient maps based on the eigen decomposition of the graph laplacian of an affinity matrix. Such maps can be used for further statistical analysis as has been shown by (Haak et al. 2016) for quantitative comparisons across individuals or to explore anatomical relationships in the cortex within the same individual. In particular, the searchlight maps allow for analysis of local transitions to compliment large scale gradient maps.

Modeling and Analysis Methods:

Methods Development ¹

Other Methods

Neuroinformatics and Data Sharing:

Informatics Other ²

Keywords:

Data analysis

FUNCTIONAL MRI

Informatics

Other - Software; Gradients

^1|2Indicates the priority used for review

My abstract is being submitted as a Software Demonstration.

Yes

Please indicate below if your study was a "resting state" or "task-activation” study.

Resting state

Healthy subjects only or patients (note that patient studies may also involve healthy subjects):

Healthy subjects

Was any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.

Yes

Was any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.

Not applicable

Please indicate which methods were used in your research:

Functional MRI

Other, Please specify - Gradient analysis

For human MRI, what field strength scanner do you use?

3.0T

Which processing packages did you use for your study?

Other, Please list - VB_toolbox

Provide references using author date format

Haak K.V. (2016): Connectopic mapping with resting-state fMRI. Neuroimage 170:83–94.
Johansen-Berg H. (2004): Changes in connectivity profiles define functionally distinct regions in human medial frontal cortex. Proceedings of the National Academy of Sciences of the United States of America 101:13335–13340.
Kriegeskorte N. (2006): Information-based functional brain mapping. Proceedings of the National Academy of Sciences of the United States of America 103:3863–3868.
Margulies D.S. (2016): Situating the default-mode network along a principal gradient of macroscale cortical organization. Proceedings of the National Academy of Sciences of the United States of America 113:12574–12579.
Yeo, B.T.T. (2011): The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology. 106, 1125–1165.
Zang Y. (2004): Regional homogeneity approach to fMRI data analysis. Neuroimage 22:394–400.