12-13 April 2017
Brainhack Ghent
Project 1: Construct an ecologically valid meta-analysis based on a large data set (Freya Acar)
To be able to test meta-analytical tools and methods it is important to have a meta-analysis set that resembles a real meta-analysis in within- and between "study" variance but in the meantime does not suffer from the same lack of data as fMRI meta-analyses. In this project we aim to develop such a meta-analysis set and test it's within- and between "study" variance and the influence of family ties on the variability.
Project 2: Running SPM on high performance computers (Freya Acar)
As High Performance Computers are more commonly used it is valuable asset to perform neuroimaging studies on them. However, there have been some issues with running the SPM-toolbox on the HPC of Ghent University. In this project we aim to compile the SPM toolbox so that it can be used on HPC's.
Project 3: Combining blind-deconvolution and DCM for resting state fMRI (Hannes Almgren)
Dynamic causal modeling (DCM) is a method used to infer effective connectivity among neural populations. Recently, a DCM for resting state fMRI has been developed which parametrises neuronal fluctuations in frequency domain ('spectral DCM'). The present project would explore whether it is feasible to use estimates of neuronal states inferred using blind-deconvolution as 'events' that can be plugged in to the DCM model.
Project 4: Is resting state fMRI a valid tool as null data for inference in fMRI? (Han Bossier)
Last year, Eklund et al. (2016) have demonstrated how clusterwise inference is invalid for functional magnetic resonance imaging. Using resting state fMRI, the researchers modeled task based paradigms to compute the empirical familywise error rate. Especially for the clusterwise inference, the researchers observed false positive rates up to 70%. Subsequently, this result was highly debated (see e.g. Cox and Reynolds 2016; Slotnick 2017). However, their results highly depend on the assumption that resting state fMRI is purely noise (white noise when serial correlation is removed in the first level analysis). Only then can we expect zero-valued estimates when modeling task based paradigms.
In this project, we would like to further investigate this assumption. We can consider the spatial distribution of the obtained false positives over all simulations, as well as the impact of the chosen task based paradigm, dataset, etc. If resting state can be used as null data, we expect there will be no spatial pattern in the results or influence of the paradigm and dataset.
Project 5: Developing a consensus approach to analyze your diffusion MRI data using MRtrix/FSL (Hannelore Aerts)
Diffusion-weighted magnetic resonance imaging (DWI) is becoming increasingly popular to study the structural connections within the brain. Since different preprocessing strategies can influence structural connectivity/tractography results, a consensus approach is required. In this project, we aim to develop such a consensus approach on how to preprocess your DWI data using MRtrix and FSL software.
Project 6: Prepare your neuroimaging data for sharing (Helena Verhelst)
Data sharing is more and more common practice in the world of research. In light of collaborations across different research labs, publicly available datasets and the growing demand for replication studies, it is important to prepare your data so that it is ready to be shared in an efficient but also ethical way. Informed consent forms reassure our participants that their data will be anonymized, so we need to live up to these promises. In this project we would like to establish a pipeline to prepare your neuroimaging data for sharing, including anonymization of filenames, removal of identifying MRI metadata, deidentification (defacing) of brain images and a clear data structure to store scans and behavioral data.
Project 7: Dynamic causal modeling: scripting and parallel estimation (Frederik Van de Steen)
In this project, template scripts will be provided for automatically specifying dynamic causal models (DCM's, both fMRI and EEG). More importantly matlab routines will be presented for estimating DCM's in parallel using the matlab parallel computing toolbox. In addition, linux bash scripts and compiled matlab code will be provided that allows to run DCM estimation in parallel on the Ugent High Performance Computing infrastructure (only for those with access to the HPC, please check http://www.ugent.be/hpc/en/policy for more information).
Project 8: Automatic BIDS-organisation of the data and processing with SPM (Christophe Phillips)
BIDS, Brain Imaging Data Structure (http://bids.neuroimaging.io/), is an open format to organize your neuroimaging data, all of your data: MRI (functional, structural, diffusion, spectroscopic, etc.), PET, EEG, MEG, behavioural, etc. The idea is to gather all the information (=data + metadata) about an experiment in a standard structure (with some flexibility) such that 1) its processing can be automatized, 2) its data integrity fully checked, 3) one could share it with his colleague or the world. There is a need to convert or bring in this format already existing data or those we are currently acquiring. There is a bit of manual intervention needed but as little as possible. For example just selecting the data manually (this is my subject sMRI, these are the fMRI of session 1 and those of session 2) along stating the subject #Id and type should be enough to 1) reorganize and rename those data and 2) add an entry in the subject summary table (the 'participants.tsv' file). All the metadata are stored in text or .json files, i.e. text-based, could thus be easily manipulated in Matlab (or any other language). In this project we would thus build these routines to help the end user setup his BIDS-ified data structure. Then we could create a “function” where one only needs to define a processing pipeline for a single subject and apply it on all the other subjects. Quality control would be ensured by signally when things ran smoothly (green light), ran but the data don’t seem to be ok (orange light), or crashed (red light).
Project 8: Automatic BIDS-organisation of the data and processing with SPM (Christophe Phillips)
BIDS, Brain Imaging Data Structure (http://bids.neuroimaging.io/), is an open format to organize your neuroimaging data, all of your data: MRI (functional, structural, diffusion, spectroscopic, etc.), PET, EEG, MEG, behavioural, etc. The idea is to gather all the information (=data + metadata) about an experiment in a standard structure (with some flexibility) such that 1) its processing can be automatized, 2) its data integrity fully checked, 3) one could share it with his colleague or the world. There is a need to convert or bring in this format already existing data or those we are currently acquiring. There is a bit of manual intervention needed but as little as possible. For example just selecting the data manually (this is my subject sMRI, these are the fMRI of session 1 and those of session 2) along stating the subject #Id and type should be enough to 1) reorganize and rename those data and 2) add an entry in the subject summary table (the 'participants.tsv' file). All the metadata are stored in text or .json files, i.e. text-based, could thus be easily manipulated in Matlab (or any other language). In this project we would thus build these routines to help the end user setup his BIDS-ified data structure. Then we could create a “function” where one only needs to define a processing pipeline for a single subject and apply it on all the other subjects. Quality control would be ensured by signally when things ran smoothly (green light), ran but the data don’t seem to be ok (orange light), or crashed (red light).
Project 9: Sensitivity of fNIRS to cognitive state changes in TPJ (Roma Siugzdaite)
Functional near infrared spectroscopy (fNIRS) is a low-cost noninvasive neuroimaging technique known as a potential alternative for fMRI in studies of executive function, particularly in pediatric and clinical populations that are not amenable to fMRI (Fishburn et al 2014). We noticed that motor task related activity is always very strong and clear, but looking at other cognitive functions that are not spatially focal such as language HRF is not so clear (Gerrits preliminary results). In this project we have three questions to address: 1) How reliable can TPJ activation be measured with a Theory of mind (ToM) localizer by means of fNIRS? 2) Are these localized channels also more sensitive in the ToM tasks? 3) How high is subject inter-variability for the Theory of mind localizer task?
Project 10: fMRI beginners facing SPM challenges (Mario Bonato)
The group will be open to unexperienced users of SPM who are interested in doing basic fMRI analysis. The idea is to take your own fMRI data and try to receive tips from the several talented geeks who will be attending the meeting.