Abstract
There is a recent increase in the use of multivariate analysis and pattern classification in prediction and real-time feedback of brain states from functional imaging signals and mapping of spatio-temporal patterns of brain activity. Here we present MANAS, a generalized software toolbox for performing online and offline classification of fMRI signals. MANAS has been developed using MATLAB, LIBSVM, and SVMlight packages to achieve a cross-platform environment. MANAS is targeted for neuroscience investigations and brain rehabilitation applications, based on neurofeedback and brain-computer interface (BCI) paradigms. MANAS provides two different approaches for real-time classification: subject dependent and subject independent classification. In this article, we present the methodology of real-time subject dependent and subject independent pattern classification of fMRI signals; the MANAS software architecture and subsystems; and finally demonstrate the use of the system with experimental results.
Highlights
PATTERN CLASSIFICATION AND RATIONALE BEHIND CHOOSING Support Vector Machines (SVM)Prediction of brain states from functional imaging constitutes a major scope of neuroscience with applications ranging from neuro-rehabilitation, brain-computer interfacing, neurofeedback in understanding brain function during cognition, perception, and detection of deception (Haynes and Rees, 2006)
Univariate statistical parametric mapping, traditionally used in functional imaging, compares activation levels at each voxel between a task state and a baseline state, or another task state, in order to determine whether the voxel is involved in a particular task or not
Given that the toolbox is mainly targeted for use by researchers and experimenters, it is expected that they will apply this tool for different experiment by creating new classifier models and test its performance in decoding brain states
Summary
Prediction of brain states from functional imaging constitutes a major scope of neuroscience with applications ranging from neuro-rehabilitation, brain-computer interfacing, neurofeedback in understanding brain function during cognition, perception, and detection of deception (Haynes and Rees, 2006). (5) To perform real-time pre-processing of fMRI images, such as realignment, co-registration, segmentation and normalization, for use in classification of brain states or brain mapping. Pre-processing This is the first step toward analysis of the brain signals It involves reading the DICOM files, converting them into Analyze format which is understood by the SPM2-based preprocessing scripts used in this toolbox, removing the noise in the acquired images, realigning them with each other, normalizing them (if required, as in the case of subject-independent analysis) to a standard space, and smoothing them. We implemented this sub-system by building around the C-language implementation of the core engine from SVM-Light (Joachims et al, 1999)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
