5 EEG Source Imaging and Multimodal Neuroimaging
119
portable nature of fNIRS have made it an intriguing counterpart to EEG, particularly
in BCI applications [2].
As a relatively young modality, signal processing for fNIRS alone has been fairly
simple. Methods have often included simple observation or T-tests/ANOVAs based
on average signals. Fourier analysis has also been applied, assuming that hemodynamic peaks occurring with the same frequency as a stimulus should be attributable
to the task. Statistical Parametric Mapping (SPM) and the GLM were also adopted
for fNIRS in a similar manner to their use in fMRI. Full integration between EEG and
fNIRS has yet to be fully explored, in part due to the relative youth of the technology.
Current methods adopted procedures such as using the β coefficients do determine
informative EEG channels for BCI classification [48], while others have derived
the independent band powers from both EEG and fNIRS frequency bands for use
in linear discriminant analysis and subsequent classification in a hybrid BCI [23].
Bayesian approaches have also appeared here, wherein fNIRS data is incorporated
as a hierarchical prior [3]. Though the field is still young, we can see here the incorporation of methods established earlier (Bayesian methods, GLM-based analyses,
etc.). These represent common themes that pervade many multimodal methods that
will continue to serve EEG combinations in the future, even as new approaches and
imaging modalities are developed.
5.4 Conclusion
EEG-based source localization and multimodal imaging stand as burgeoning topics
within the field of biomedical imaging, and research into both topics has remarkable
breadth and depth. New approaches are constantly arising in this field, pushing it further as more accurate (and often more complex) methods arise. We have sought here
to provide a functional basis from which the general principles and seminal methods of source localization can be understood. This necessarily has driven through
discussions of cortical modeling, types of models, and the forward and inverse calculations used to connect EEG to its potential sources. Continuing beyond this, we
have explored a variety of the possible multimodal combinations that EEG features
in and provided both algorithmic and practical examples for how these fusions are
performed. It is important at this stage to remember that none of this discussion
is comprehensive; even the topics explored more heavily in this chapter have not
achieved their full depth. The readers are instead encouraged to look through the
cited materials on their own, for many of the individual topics that we have touched
on could be the subject of full books.
119
portable nature of fNIRS have made it an intriguing counterpart to EEG, particularly
in BCI applications [2].
As a relatively young modality, signal processing for fNIRS alone has been fairly
simple. Methods have often included simple observation or T-tests/ANOVAs based
on average signals. Fourier analysis has also been applied, assuming that hemodynamic peaks occurring with the same frequency as a stimulus should be attributable
to the task. Statistical Parametric Mapping (SPM) and the GLM were also adopted
for fNIRS in a similar manner to their use in fMRI. Full integration between EEG and
fNIRS has yet to be fully explored, in part due to the relative youth of the technology.
Current methods adopted procedures such as using the β coefficients do determine
informative EEG channels for BCI classification [48], while others have derived
the independent band powers from both EEG and fNIRS frequency bands for use
in linear discriminant analysis and subsequent classification in a hybrid BCI [23].
Bayesian approaches have also appeared here, wherein fNIRS data is incorporated
as a hierarchical prior [3]. Though the field is still young, we can see here the incorporation of methods established earlier (Bayesian methods, GLM-based analyses,
etc.). These represent common themes that pervade many multimodal methods that
will continue to serve EEG combinations in the future, even as new approaches and
imaging modalities are developed.
5.4 Conclusion
EEG-based source localization and multimodal imaging stand as burgeoning topics
within the field of biomedical imaging, and research into both topics has remarkable
breadth and depth. New approaches are constantly arising in this field, pushing it further as more accurate (and often more complex) methods arise. We have sought here
to provide a functional basis from which the general principles and seminal methods of source localization can be understood. This necessarily has driven through
discussions of cortical modeling, types of models, and the forward and inverse calculations used to connect EEG to its potential sources. Continuing beyond this, we
have explored a variety of the possible multimodal combinations that EEG features
in and provided both algorithmic and practical examples for how these fusions are
performed. It is important at this stage to remember that none of this discussion
is comprehensive; even the topics explored more heavily in this chapter have not
achieved their full depth. The readers are instead encouraged to look through the
cited materials on their own, for many of the individual topics that we have touched
on could be the subject of full books.