Perelman School of Medicine at the University of Pennsylvania

Section for Biomedical Image Analysis (SBIA)

participating with CBICA

Resting-state Functional MRI (rs-fMRI)

Resting-state Functional MRI (rs-fMRI) is a prominent tool for the analysis of brain function. By capturing the interaction between all brain regions at once, this modality provides a rich description of brain connectivity and a crucial insight on the cognitive change happening through the course of neurodevelopment.

However, resting-state fMRI suffers from several issues. First, the subjects need to be scanned for a long time, at least a dozen of minutes. As a result, rs-fMRI scans count among the largest neuroimaging 4D datasets. This large amount of data significantly complicates the analysis and increases their computational burden. Moreover, rs-fMRI scans usually contain a large amount of noise, for a large part induced by subject motion in the scanner. This critical issue requires a great deal of efforts, such as the use of dedicated motion correction and denoising strategies. In addition to motion and noise, multiple factors such as subject stress or caffeine consumption have been shown to have an impact on rs-fMRI. As a result, the reproducibility of individual rs-fMRI scans is significantly more limited than other MRI modalities.

Our research focuses on the developments of mathematical frameworks for tackling these issues. Our goal is to develop methods able to extract reliable biomarkers from individual rs-fMRI scans, with the aim of providing new tools to the neurologists studying brain development. We have initiated the following crucial preliminary steps toward this end.

(a1) Develop a robust and reproducible parcellation schemes. Parcellating the brain allows to understanding how the brain is organized into functional units and offers the possibility to reduce noise by averaging the rs-fMRI signal in the regions where it is strongly coherent. [1]

Brain parcellations at different scales for the PNC.

(a2) Extract shared robust and interpretable functional networks, with the aim of being able to decompose the connectivity measured from an individual rs-fMRI scan as a combination of large networks shared across a population. This combination is an interesting biomarker, which might be used for describing or diagnosing brain disease. [2]
Shared functional network extracted from the PNC.

(a3)Develop novel denoising and feature extraction approaches, with the aim of improving statistical significance, biomarkers quality, and incorporate other insight such as local brain connectivity.[3]

Mode of variation of the HCP data and potential biomarker.

The methods developed during these projects aims at addressing two transversal topics.

(b1) The study of neurodevelopment, via the analysis of the multimodal data acquired as part of the neurodevelopmental studies such as the Philadelphia Neurodevelopmental Cohort (PNC).

Cortical map summarizing one of the relations between local structure and local function in the PNC.

(b2) The study of the variability of healthy brains by analyzing large datasets, such as the publicly available Human Connectome Project (HCP), with the aim of developing templates and models able to detect the abnormal changes induced by pathological neurodevelopment.

Median myelin map for the hundred unrelated HCP subjects.
  1. Honnorat N, Eavani H, Satterthwaite T, Gur RE, Gur RC, Davatzikos C. GraSP: Geodesic Graph-based Segmentation with Shape Priors for the functional parcellation of the cortex. NeuroImage. 2015;106:207-21.
  2. Eavani H, Satterthwaite TD, Filipovych R, Gur RE, Gur RC, Davatzikos C. Identifying Sparse Connectivity Patterns in the brain using resting-state fMRI. NeuroImage. 2015;105:286-99.
  3. Honnorat N, Satterthwaite TD, Gur RE, Gur RC, Davatzikos C. sGraSP: A graph-based method for the derivation of subject-specific functional parcellations of the brain. Journal of neuroscience methods. 2017;277:1-20.