Can we learn from coupling EEG-fMRI to enhance neuro-feedback in EEG only?

Tuesday, 10 de November de 2020

Un nuevo estudio nos muestra las ventajas de usar EEG-fMRI para mejorar nuestras técnicas de neuro-feedback.
 

Neurofeedback (NF) measures brain activation during a task, and gives back to the subject a score reflecting his performance that he tries to improve. Among noninvasive functional brain imaging modalities, the most used in NF, are electroencephalography (EEG) and the functional magnetic resonance imaging (fMRI) For MRI.


EEG measures the electrical activity of the brain through channels located on the scalp, with an excellent temporal resolution (milliseconds), but has a limited spatial resolution due to the well-known ill-posed inverse problem of source reconstruction. Also, NF-EEG (NF session with NF scores extracted from EEG) is not easy to control since it comes from mixtures of propagating electric potential fluctuations.


Blood oxygenation level-dependent (BOLD) fMRI measures neuro-vascular activity, easier to control, with an excellent spatial resolution, making NF-fMRI (NF session with NF scores extracted from BOLD-fMRI) an adequate modality for NF. However its temporal resolution is only of a few seconds, and it is costly, exhausting for subjects and time-consuming modality.

Since those modalities are complementary, their combined acquisition is actively investigated, as well as the methodology to extract information from fMRI with EEG which is the easiest modality to use [EEG fMRI
].
Our challenge is to learn EEG activation patterns from NF-fMRI scores extracted during an NF session using coupled EEG fMRI data (NF-EEG-fMRI) to improve NF scores when using EEG only.

Methods:

The researchers (1) propose an original alternative to source reconstruction in the context of NF. They directly intend to predict NF-fMRI scores, without having to deal with EEG source reconstruction to estimate the BOLD-fMRI signal as proposed by methods reviewed in [Abreu et al. 2018].

They first summarized each time interval of EEG signal in a design matrix, estimated from different frequency bands and channels. They then applied a non-linear transformation to the design matrix to increase the potential linear relationship of EEG with NF-fMRI scores [Meir-Hasson et al 2014]. The model estimation uses an adapted prior, using a mixed norm giving a structured sparsity [Gramfort et al. 2011], to first spatially select electrodes and then smooth the corresponding frequency bands.

Linear regression with the chosen prior is then applied to learn the EEG activation pattern, which takes EEG signal only as input and outputs an estimate of the NF-fMRI score. On the testing set, EEG only is measured and the learned activation pattern is applied to predict NF-fMRI scores before being combined to the NF-EEG scores (Laplacian above C3).


In their experimental setup, EEG and fMRI are acquired and used to estimate reference NF-EEG and NF-fMRI scores via a hybrid neurofeedback platform [Mano et al.]. They used a dataset of 17 subjects acquired with a 64-channel MR-compatible EEG solution from Brain Products and a 3T Verio Siemens scanner with a 12 channels head coil at the NeurInfo platform (Rennes, France) [Perronnet et al. 2017]. Subjects are healthy volunteers, right-handed, and had 3 NF motor imaging sessions of 8 blocks (rest/task) each.


Results:

Estimated pNF-fMRI on the learning set has a median correlation of 0.80 with the ground truth scores, confirming that the model is well adapted to the problem.
For each subject, activation patterns were learned on a NF session and applied to estimate the prediction pNF-fMRI on the 2 other sessions. The correlation of NF-EEG-fMRI with the predicted score combining pNF-fMRI + NF-EEG is significantly better than NF-EEG alone (mean correlation of 0.70 vs 0.67, paired t-test: p = 0.02). The average activation pattern shows spatial sparsity and smoothing in the frequency bands.



 

Conclusions:

The researchers presented an efficient model able to learn from NF-EEG-fMRI sessions and enhancing NF scores estimation when using EEG only. They were able to predict NF-fMRI with EEG signals only. pNF-fMRI was good enough to add consistent information to the NF scores.

 


References

https://www.brainlatam.com/manufacturers/brain-products/-fmri-EEG-cap-braincap-mr-111

1. Claire Cury, Pierre Maurel, Lorraine Perronnet, Rémi Grionval, Christian Barillot. (2019). EEG-fMRI to enhance neuro-feedback.

2. Abreu, R., EEG-Informed fMRI: A Review of Data Analysis Methods. Frontiers in Human Neuroscience, no. 12 (Feb 2018)

3. Gramfort, A., Functional Brain Imaging with M/EEG Using Structured Sparsity in Time-Frequency 
Dictionaries. International conference on Information Processing in Medical Imaging, vol. 6801, pp. 600–611 (2011)

4. Mano, M., How to Build a Hybrid Neurofeedback Platform Combining EEG and fMRI. Frontiers in Neuroscience 11 (2017)
Meir-Hasson, Y., An EEG Finger-Print of fMRI deep regional activation. NeuroImage 102, pp. 128–141 (2014) 

The content published here is the exclusive responsibility of the authors.

Autor: Sebastian Moguilner
#eegerpbci #brainstimulation #eegfmri #eegnirscombined #eeglatam #nirsbcineurofeedback #selfperceptionconsciousness #bienestarwellnessbemestar #neurocognition