Assessing Uncertainty and Reliability of Connective Field Estimations From Resting State fMRI Activity at 3T.

Azzurra Invernizzi,Azzurra Invernizzi,Nicolas Gravel,Remco J Renken,Nicolas Gravel,Nicolas Gravel,Koen V Haak,Koen V Haak,Frans W Cornelissen,Frans W Cornelissen

doi:10.3389/fnins.2021.625309

Abstract

Connective Field (CF) modeling estimates the local spatial integration between signals in distinct cortical visual field areas. As we have shown previously using 7T data, CF can reveal the visuotopic organization of visual cortical areas even when applied to BOLD activity recorded in the absence of external stimulation. This indicates that CF modeling can be used to evaluate cortical processing in participants in which the visual input may be compromised. Furthermore, by using Bayesian CF modeling it is possible to estimate the co-variability of the parameter estimates and therefore, apply CF modeling to single cases. However, no previous studies evaluated the (Bayesian) CF model using 3T resting-state fMRI data. This is important since 3T scanners are much more abundant and more often used in clinical research compared to 7T scanners. Therefore in this study, we investigate whether it is possible to obtain meaningful CF estimates from 3T resting state (RS) fMRI data. To do so, we applied the standard and Bayesian CF modeling approaches on two RS scans, which were separated by the acquisition of visual field mapping data in 12 healthy participants. Our results show good agreement between RS- and visual field (VF)- based maps using either the standard or Bayesian CF approach. In addition to quantify the uncertainty associated with each estimate in both RS and VF data, we applied our Bayesian CF framework to provide the underlying marginal distribution of the CF parameters. Finally, we show how an additional CF parameter, beta, can be used as a data-driven threshold on the RS data to further improve CF estimates. We conclude that Bayesian CF modeling can characterize local functional connectivity between visual cortical areas from RS data at 3T. Moreover, observations obtained using 3T scanners were qualitatively similar to those reported for 7T. In particular, we expect the ability to assess parameter uncertainty in individual participants will be important for future clinical studies.

Highlights

Spontaneous blood-oxygen level dependent (BOLD) fluctuations have been used to study the intrinsic functional connectivity of the human brain. Biswal et al (1995) observed, for the first time, the presence of bilateral spatial integration, coherent activity and functional connectivity between distant homotopic brain areas, even in the absence of a task
The connective field (CF) maps obtained from resting state (RS)-based data for eccentricity, polar angle and CF size were qualitatively comparable for the standard and Bayesian CF models
An higher uncertainty from the CF parameters was observed from both RS scans compared to visual field mapping (VFM) data and between RS2 and RS1 scans

Summary

Introduction

Spontaneous blood-oxygen level dependent (BOLD) fluctuations have been used to study the intrinsic functional connectivity of the human brain. Biswal et al (1995) observed, for the first time, the presence of bilateral spatial integration, coherent activity and functional connectivity between distant homotopic brain areas, even in the absence of a task. CF, known as the cortico-cortical population receptive field (cc-pRF), modeling allows to describe the response of a population of neurons in the cortex in terms of the activity in another region of the cortex It translates the concept of the receptive field into the domain of connectivity by assessing the spatial dependency between signals in distinct cortical visual field regions (Haak et al, 2013b). A previous study by Gravel et al (2014) showed that CFs, estimated from RS-fMRI data recorded at a high magnetic field (7T), reflect the visuotopic organization of early visual cortical maps This indicates that even in the absence of any visual stimulation, CF modeling is able to describe the activity of voxels in a target region (e.g., V2 or V3) as a function of the aggregate activity in a source cortical visual area (e.g., V1)

Methods

Results

Discussion

Conclusion