Neuroinformatics Workgroup

What is Neuroinformatics: In a nutshell, Neuroinformatics is a transdisciplinary field encompassing neuroscientific knowledge and questions and the applicaton of computational tools and models. The growth of this area in the past few years reflects its crescent contribution to experimental designs, image processing, and theoretical developments in the field of neuroscience, including clinical neuroscience. Another important contribution of neuroinformatics involves neurotechnological developments, such as software and hardware integration tools (for example, Brain-Computer Interfaces, BCI).

Our Neuroinformatics Workgroup, is closely associated with the Translational Cognitive Neuroscience of Affective Disorders Lab (, and includes professionals from multiple areas, including neuroscience, mathematics, statistics, clinical neuropsychiatry, computer engineering and software development. Among several projects involving neuroinformatics (fMRI, DTI, VBM, spectroscopy, cortical thickness, applied to cognitive neuroscience, neuroplasticity, neurological and psychiatric disorders), we have been dedicating special efforts to the development of methods, experimental designs and software tools for neuromodulation using real-time functional MRI (rt-fMRI) neurofeedback. A core part of this project is the development of the FRIEND toolbox (Functional Real-time Interactive Endogenous Neuromodulation and Decoding).

This was motivated by the observaton that people can be taught to control their own neural activity when they are given feedback about their ongoing neural states. Earlier neurofeedback experiments relied primarily on electroenchephalography (EEG), with subjects aiming at increasing amplitudes in a given frequency range. However, fMRI provides the unique benefits of higher spatial accuracy and importantly, the ability to monitor neural activity in deep brain regions which are virtually inaccessible by EEG methods (see groundwork on fMRI neurofeedback). This is crucial for the ability to perform neuromodulation of emotional and motivational states, a cornerstone for applications in the fields of psychology and neuropsychiatry. In the near future, our rt-fMRI tool will be integrated with quantitative EEG tools, in order to combine the benefits of high spatial accuracy of fMRI and high temporal accuracy of EEG.

FRIEND: Functional Real-time Interactive Endogenous Neuromodulation and Decoding

FRIEND is a graphic-oriented user-friendly software for real-time fMRI processing, multivoxel pattern decoding and neurofeedback. The package integrates routines for image preprocessing in real-time, ROI-based feedback (single-ROI percent signal change and functional connectivity) and brain decoding-based feedback using the FSL and libSVM libraries. FRIEND enables users to create or employ pre-specified visual stimuli for neurofeedback experiments. FRIEND also allows extensive control of processing parameters and includes quality control features.

FRIEND currently comes in two flavours: FRIEND for Windows and FRIEND Engine. The standalone Windows® version runs embedded FSL and libSVM functions and provides a simple, straightforward solution. The FRIEND Engine is a multiplatform (Mac/Linux) toolbox that incorporates the same processing capabilities as noted above, but in which the frontend is separated from the processing module. This allows users to develop their own frontend GUIs while employing FRIEND processing capabilities through TCP/IP communication services. Advanced users can develop plugins to extend the original pipeline provided by the engine (e.g., Matlab functions for data preprocessing and multivariate classification, or stimulus presentation features from standard packages such as E-Prime, for customized feedback). A fully functional frontend is also provided along with the FRIEND Engine distribution.

In this clip, the initial segment shows the first version of the rt-fMRI display of a volunteer thinking about affiliative emotional memories. Note the increasing activity in the septal region and especially in the orbitofrontal cortex, in the frontal part of the brain. The second part of the clip shows the control panel of the FRIEND toolbox, while a volunteer was trained to control his own brain activity (positive or negative emotional affiliative states) while being given a feedback signal (distorted rings, which became progressively smoother as the SVM weight vector projections increased). High accuracies were obtained when classifying positive vs. negative affiliative emotional states (AGRAD or DESAGR cues, which are inverted on the screen because they were projected to the participant in the MRI scanner via a mirror).



| Mais