Outline of the module
Recent progress in computer hard- and software enables real-time analysis of fMRI data, providing the basis for neurofeedback and brain-computer Interface (BCI) applications. This real-time analysis approach is in contrast to standard offline analysis in which fMRI signals are recorded first and analysed (hours or days) after the experiment has been completed. In the first part of this module, modifications of methods that are specific to real-time analysis will be introduced, including incremental filtering approaches, recursive general linear model (GLM) analysis, online support vector machine (SVM) classification, windowed functional connectivity measures, and windowed independent component analysis (ICA). The second part of this module discusses several applications of real-time fMRI analysis, including neurofeedback and (communication) brain computer interfaces (BCIs).
The module covers the following topics:
• Difference between real-time fMRI analysis and conventional offline analysis
• Operating an incremental GLM
• Use of windowed ICA to study dynamic changes in networks
• Theory behind online pattern classifiers
• Status of fMRI neurofeedback for clinical applications
• Using real-time fMRI for communication BCIs
• Comparison of real-time hemodynamic BCIs to EEG-based BCIs
Learning objectives
At the end of this module, students will have knowledge of:
• Designing, planning and analysing real-time fMRI studies
• Running neurofeedback studies including selection of target regions-of-interest
• Running online support vector machine classifiers including selection of learning and test phases
• Understanding the methodological details of incremental GLM, online SVM, windowed partial correlation and windowed ICA and their respective advantages and disadvantages for neurofeedback and BCI applications
• Most important neurofeedback studies, and their potential as new therapeutic tools
• fMRI-based communication BCIs and their strengths and weaknesses in comparison to fNIRS and EEG BCIs
Content
Real-time fMRI can be used as a basic tool for quality assurance, i.e. to check signal quality, head motion and brain activity online. More advanced and challenging applications of real-time data analysis include neurofeedback studies and brain computer interfaces (BCIs). Before understanding such exciting applications, details of real-time analysis methods will be taught in this module. This includes calculation of incremental (recursive) general linear models (GLM) with dynamically growing design matrices that ensure constant calculation time per data point and to add or drop (confound) predictors during an online experiment. The distinction between windowed and whole (past) time module estimates will be described in the context of GLM, ICA and partial correlation measures. The principles of support vector machine classifiers will be taught as a method for online classification of mental states (online “brain reading”) and how to best preprocess the data online and how to reduce the number of features (voxels) to ensure high generalisation performance.
The second part of the module focuses on advanced real-time fMRI applications, including neurofeedback and communication brain computer interfaces (BCIs).
In fMRI neurofeedback studies, subjects see (or hear/feel) representations of their own brain activity from selected brain areas or networks during an ongoing fMRI measurement in the scanner. Such feedback studies have shown that participants are able to learn to modulate activity in certain brain areas. These results are extremely important for basic neuroscience research, because they enable to study the degree to which the brain can modulate its own activity and to potentially unravel the function of hitherto unknown brain areas. Furthermore, fMRI neurofeedback will be discussed as a potential new therapeutic tool for some disorders; promising results have been recently obtained with Parkinson patients (Subramanian et al., 2011) and patients suffering from depression.
In fMRI-based communication BCI studies, activation patterns evoked by participants are ‘decoded’ and interpreted online, e.g. as letters of the alphabet, offering the possibility for people with severe motor impairments to ‘write’ letters purely controlled by mental imagery. In this module, a number of online analysis strategies will be discussed for decoding mental states, including analysis of the mean signal of regions-of-interest (ROIs) and the use of pattern classifiers operating at the voxel level. At the end of this module, students will have basic understanding of how real-time fMRI operates and how it can be applied to novel neuroscientific and clinical applications.
Overview of tasks and lectures
There will be 10 lectures of 2 hours distributed over 5 days.
• Introduction into real-time functional MRI data analysis
• Principles of incremental GLM and dynamic design matrices
• Windowed correlation and windowed ICA
• Online training and application of pattern classifiers
• Neurofeedback applications I: Feasibility studies
• Neurofeedback applications II: Patient studies
• BCI applications I: “Brain Pong” real-time hyperscanning
• BCI applications II: From “brain reading” to “brain writing”
• Functional near-infrared spectroscopy (fNIRS): A mobile alternative?
• Summary and Introduction of post-module assignments
Position within the programme
This is a unique module in this Master programme with both a strong methodological and applied focus. The methodological knowledge builds on advanced data analysis modules moving state of the art analysis techniques into a real-time setting. It also builds on module ‘Modelling: from neurons’ to fMRI as a basis to judge the interpretability of single-trial fMRI responses.
Teaching format
Structure
The module is a one week-long residential module consisting of 10 lectures of 2 hours. On most days, students will use real-time fMRI analysis software (Turbo-BrainVoyager) to perform incremental (quasi real-time) data analysis on recorded data sets and to implement small C++ routines guided by tutors. In addition, at least one real-time measurement on a scanner will be performed.
Grading
Passing the module requires an 85% attendance to the lectures and practical sessions, and a satisfactory completion of the practical sessions and the module assignments. The module assignments will be summarised by the students in a written form, which will be evaluated by the module coordinator(s).