Event Abstract

Evaluating Structural Connectomics: the effect of the cortical parcellation scheme

  • 1 Mint Labs S.L., Spain
  • 2 University of Barcelona, Facultat de Medicina, Spain
  • 3 University of Barcelona, Faculty of Psychology, Spain
  • 4 Hospital Clinic, Service of Neurology, Spain
  • 5 IDIBAPS, Medical Imaging Platform, Spain

Introduction

The brain is a complex system whose components continuously create complex patterns. Therefore, a natural paradigm for studying the brain is via network analysis. Brain networks span multiple spatial scales, from microscale of individual cells and synapses to the macroscale of cognitive systems and embodied organisms. A comprehensive map of neural connections of the brain is called the "connectome'' [1, 2]. At the macroscopic scale, the connectome can be seen as a network, usually represented as a graph, where each vertex represents well-defined cortical or sub-cortical structures and the edges quantify the structural white matter connectivity as measured with tractography. When this graph is estimated from Diffusion Weighted MRI (dwMRI) data one can speak about structural connectivity. In the process of calculating the connectomes from the measured data, several parameters are involved that can lead to variations of the connectivity matrices. An important step is the identification of the different grey matter structures, such as the deep grey nuclei and cortical gyri, for obtaining the nodes of the network.
In this work, we evaluated the information difference contained in the structural connectomes constructed from the same subject, at different levels of the cortical parcellation, scanned with different dwMRI acquisition techniques (DTI, HARDI and DSI).


Methods and Subjects

Data: MRI acquisitions were performed in 5 healthy volunteers (4 male and 1 female, age:31.2±2.9 years) using a twice refocused spin-echo echo-planar imaging sequence on a 3T Siemens Trio MRI scanner (Erlangen, Germany). Informed consent was obtained prior to the acquisition. The MRI protocol included the following sequences: a) 3D structural T1-weighted MPRAGE sequence: Repetition Time (TR): 1900ms, Echo Time (TE): 4.44ms, Inversion recovery time (TI): 1050ms, Flip angle: 8◦, FOV: 220×220mm2, isometric 1mm3; b)The parameters for the dwMRI sequences are given in the table 1 (top). We have acquired in total 21 datasets from which 7 DTI, 6 HARDI and 8 DSI. For some of the subjects the scans were repeated in the same scanning session or after one month.

Connectome calculation: We calculated the connectomes using publicly available software, the connectome mapper (cmp). For all imaging modalities we used the default settings (re-sampling the dwMRI data to 1mm isotropic voxel size using trilinear interpolation, tracking stopping criteria at angle=60◦, number of seeds=32, fibre filtering with enabled spline filter and cut-off filter in the interval of [20;500] mm). For DTI fibre tracking, the cmp uses the standard FACT method and for Qball and DSI FACT-alike algorithm implemented in the Diffusion Toolkit. For only two subjects to improve the registration step we performed non-linear registration using the T2 data, and for the rest of the subjects we used linear registration. We employed Lausanne parcellation where from a starting anatomical atlas, it creates a multi-scale parcellation of the cerebral cortex. At the end of the process, each of the five atlases comprises, respectively, a total of 1015, 463, 234, 129 and 83 labels.

Connectome comparison measures: Depending on the imaging modality, DTI, HARDI (Qball) or DSI reconstruction was performed. For network creation, we first apply an absolute threshold in order to discard edges with less than 10 fibres (considered spurious fibres from data observation), and connection matrices are created by either binarizing edge weights, or by normalizing the edge weights with maximum number of found fibres. Indices for connectome comparison and quality assessment: The simplest way of comparing networks is to assess the difference between their overall matrix representations. We computed the correlation between the graphs. We computed several other indices (normalized root-mean-square deviation, dot product of the direct embedding of the matrix into a vector-space representation, Hamming distance and Fleiss’ kappa reliability of agreement) but since they do not give further insight, we omit them for simplicity.


Results

As is apparent in Figure 1, the choice of connectome construction method (binarizing or normalizing edges) influences the used measures. Binarized connectomes show overall smaller correlation indices. As parcellation scale decreases (33 to 500 parcellated regions) another trend can be observed. In binarized connectomes the correlation index decays slower while in the normalized connectomes we observe the opposite effect. We evaluated the variability of the connectome constructed from data acquired in the same imaging session, as well as after some period of time (a month in our case) and compare it among connectomes constructed at different scales of parcellation (33-500). Highest reproducibility is achieved within the same day at 0.95 (for normalized networks). As scale decreases, however, the differences between the different acquisition schemes becomes larger, with DSI keeping the most reproducibility index. This interesting trend is not observed with binarized networks.
To assess the differences between structural connectomes built using different modalities, we calculated the similarity between different modalities within each subject. The agreement between them is kept across scales, however it decreases faster in the case of normalized networks.


Conclusions

In this work we have evaluated the characteristics and mutual differences of the structural connectomes constructed over multi-scale cortical parcellation scheme. We have done this by employing graph based measures to real data that can quantify the information content and the differences between different techniques, at different scales of hierarchy. From these measures we observed that the connectome does not significantly capture richer information by using locally more accurate acquisition schemes such as DSI, if one uses a course parcellation scheme. As scale decreases (more fine parcellation areas) the differences between the different acquisition schemes become bigger. This study shows only preliminary results from 5 subjects. To improve the statistic analysis, larger cohort of subjects should be analyzed.

Figure 1

References

1. Hagmann, P.: From Diffusion MRI to Brain Connectomics. PhD thesis, Ecole Polytechnique Federale de Lausanne (2005)
2. Sporns, O., Tononi, G., Kötter, R., Ko, R.: The human connectome: A structural description of the human brain. PLoS computational biology 1(4) (September 2005) e42

Keywords: connectome, structural connectomics, Diffusion Magnetic Resonance Imaging, cortical parcellation, tractography

Conference: Imaging the brain at different scales: How to integrate multi-scale structural information?, Antwerp, Belgium, 2 Sep - 6 Sep, 2013.

Presentation Type: Poster presentation

Topic: Poster session

Citation: Rodrigues P, Prats A, Gallardo-Pujol D, Villoslada P, Falcon C and Prčkovska V (2013). Evaluating Structural Connectomics: the effect of the cortical parcellation scheme. Front. Neuroinform. Conference Abstract: Imaging the brain at different scales: How to integrate multi-scale structural information?. doi: 10.3389/conf.fninf.2013.10.00030

Received: 31 Aug 2013; Published Online: 31 Aug 2013.

* Correspondence: Dr. Vesna Prčkovska, Hospital Clinic, Service of Neurology, Barcelona, Spain, vesna.prckovska@gmail.com

© 2007 - 2019 Frontiers Media S.A. All Rights Reserved