Graphs, the Shire, & multimodal imaging

Neuro-X Institute, École polytechnique fédérale de Lausanne, Geneva, Switzerland

Department of Radiology and Medical Informatics (DRIM), Faculty of Medicine, University of Geneva, Geneva, Switzerland

physiopy (https://github.com/physiopy)

	smoia
	@SteMoia
	s.moia.research@gmail.com

Donostia, 21.11.23

Graphs, the Shire, & multimodal imaging

Donostia, 21.11.23

The Shire

Geneva

Graph Signal Processing

Ortega (2022) Introduction to Graph signal Processing & other material courtesy of Dimitri Van De Ville

A-D = \mathcal{L} = U\lambda U^T

Graph Signal Processing: brain applications

Novel approach to signal processing, based on graph embedding of timeseries¹.
Useful for multimodal brain imaging².
GSP applications can improve subject identification³, improve data smoothing⁴, and reveal structural/functional coupling patterns agreeing with cognitive domain hierarchy and genetic expression⁵.

1. Ortega (2022) Introduction to Graph signal Processing,
3. Griffa et al. (2022) Neuroimage,
5. Preti, Van De Ville (2019) Nat. Commun.

2. Huang et al. (2018) Proc. IEEE,
4. Abramian (2021) Neuroimage,

Shuman et al. (2013) IEEE Signal Process Mag

Huang et al. (2018) Proc. IEEE

Huang et al. (2018) Proc. IEEE

Graph Signal Processing

Griffa et al. 2021 (bioRxiv), Preti et al. 2019 (Nat. Commun.)

NiGSP

NiGSP is an open development python based module to apply
Graph Signal Processing, with emphasis on neuroimaging applications (atlas based)

It offers a set of function for filter creations, filtering, visualisation,
and metrics computations

NiGSP: workflow

nigsp -f timeseries.nii.gz -s sc_mtx.tsv -a atlas.nii.gz --informed-surrogates -n 1000

Based on Preti, Van De Ville (2019) Nat. Commun. and Griffa et al. (2022) Neuroimage

Features

Digital Object Identifier
Automatic output of citations
all-contributors authorship
Apache 2.0 licence
Thoroughly tested
Graphic identity

NiGSP: workflow extension

Common data loading
Different central "workflow" for different analyses/publications
Common data visualisation and data export
Wrapper for BIDS-compatible reading
Easy to run at subject or group level
Improved performance

Graph Signal Processing and lesions

1. Egger et al. 2021 (Stroke); 2. Egger et al. in prep.; * Morishita, Inoue, 2016 (Neurol. Med.-Chir.)

Subcortical brain lesions (e.g. stroke) may affect structural connectivity (SC) by disrupting white matter traits.

However, the extent of its impact might be underrated by current tractography methods, especially in the acute phase of stroke¹.

Instead, the portion of SC interested by the lesion only can be computed and removed from the global SC¹. Doing so increases agreement between SC and prognosis¹ or SC and behavioural scores²

GSP-based diffusivity model

\Delta Y = (- I - \tau_1 \mathcal{L}_1 - \tau_2 \mathcal{L}_2) Y + E \quad \leftrightarrow \quad E_n = \Delta Y + (I + \tau_1 \mathcal{L}_1 + \tau_2 \mathcal{L}_2) Y_{n-1}

\Delta Y = -\tau \mathcal{L} Y + E \quad \leftrightarrow \quad E_n = \Delta Y + \tau \mathcal{L} Y_{n-1}

Total

Unaffected

Affected