Minicourse-21

Synopsis: NIRSTORM is a plugin dedicated for fNIRS data analysis, built upon Brainstor, an internationally recognized software for EEG/MEG processing, featuring advanced databasing, visualization, signal processing, source localization and statistical analysis methods. The purpose of this mini-course is to introduce NIRSTORM as a user-friendly and fully complete environment dedicated to fNIRS statistical analysis. The first section will be dedicated to beginners, introducing NIRSTORM database, data importation and classical channel-space fNIRS processing (band pass filtering, Modified Beer-Lambert Law, motion correction and window averaging). Most recent updates will then be presented: General Linear Model based statistical analyses (auto- regressive/precoloring model, mixed-effect group level analysis) to provide statistics of the hemodynamic response either in the channel space or along the cortical surface after 3D reconstruction. Finally, we will present the most advanced NIRSTORM features, such as the integration of MCXLab software [Fang and Boas Opt. Express 2009] to estimate light sensitivity profiles within anatomical head models, our method allowing personalized optimal montage design targeting a predefined brain region [Machado et al JNS-Methods 2018] and advanced 3D reconstructions using Maximum Entropy on the Mean. 

Rationale : Despite measuring physiological signals different origins, EEG and fNIRS are sharing several similarities: (i) they consist in scalp measurements, (ii) they offer an excellent temporal resolution and access to long duration recordings, (iii) their spatial resolution is limited and 3D reconstruction of the generators of these scalp recordings requires solving an ill-posed inverse problem. This was the main reason for us to choose Brainstorm software environment [Tadel et al. Comp Intell Neurosci 2011] to develop a fNIRS data analysis platform inspired by electrophysiology. NIRSTORM allows ideal 3D visualizations and interactions features involving multi-channel signals in the time domain, co-registration of fNIRS sensors along an anatomical model and eventual 3D reconstructions of hemodynamic responses within the brain, along the cortical surface. In addition to fNIRS specific features we implemented, NIRSTORM also benefits from Brainstorm complete library of signal processing methods that can be directly applied to fNIRS data (filtering, time-frequency based analysis, non parametric statistics, estimation of functional connectivity patterns), while offering the possibility to implement new specific scripts and pipelines. NIRSTORM is an open-source initiative, developed in Matlab and it welcomes any contribution. It is currently hosted on github (https://github.com/Nirstorm/nirstorm), where the wiki pages of our first training session organized in Montreal in May 2018 are available. 

Course structure: The course will consist in hands-on sessions, fNIRS data sets dedicated for the training will be made available to the participants.

Part 1

• Database organization in Brainstorm and fNIRS data importation 
• Standard fNIRS preprocessing and quality check (co-registration, filtering, Modified Beer Lambert law, motion correction, block averaging) 
• Statistical analysis of the hemodynamic response: General Linear Model at the single subject level and at the group level, at the level of the sensors and after 3D reconstruction along the cortical surface 
• Measurement of functional connectivity patterns

Part 2

• fNIRS forward model through MCXLab, using head models derived either from a standard template MRI (Colin 27) or a subject-specific MRI. 
• Personalized optimal montage design targeting a predefined brain region. This method consists in maximizing light sensitivity to the target region, while ensuring spatial overlap between sensors to allow local 3D reconstruction [Machado et al JNS-Meth. 2018, Pellegrino et al Front. Neurosc. 2016]. 
• Advanced 3D reconstruction methods, inspired from methods developed for EEG/MEG source imaging, notably within the Maximum Entropy on the Mean framework.

Learning objectives: At the end of the session, participants will be able to use efficiently the GUI of Brainstorm and NIRSTORM to perform standard fNIRS processing, statistical analysis through GLM approaches and more advanced features such as tomographic reconstructions and optimal montage design.

Requirements: You are expected to bring a laptop with Matlab and NIRSTORM installed. Detailed instructions and training datasets will be provided before the course. 

Please follow the links below to download NIRSSTORM Plug-in and Brainstorm software.

NIRSTORM Plug-in (link)

Brainstorm software (link)

You can find previous training sessions at this link.