Want to blindly get robust estimates of neuronal-related activity on your resting-state, naturalistic or clinical fMRI data?
I am excited to share our latest preprint with you👇
Whole-brain multivariate hemodynamic deconvolution for multi-echo fMRI with stability selection
It has been a pleasure to collaborate with @fMRI_today, @javiergcas, Charles Zheng, and @CaballeroGaudes.
You can read the preprint in the following link: biorxiv.org/content/10.1101/2022.09.30.510299v2
Or you can keep reading to learn about the highlights👇
Context:
Deconvolution algorithms can blindly estimate neuronal-related activity without any information about the timings of the BOLD events.
This makes them especially useful to analyze resting-state, naturalistic, and clinical fMRI data, where timings are unavailable.
twitter.com/eurunuela/status/1420036314066341888
‼️ However, these methods have 3 main limitations ‼️
• they don't take advantage of the spatial information in the data
• their efficacy strongly depends on the selection of the regularization parameter
• they offer no measure of statistical certainty of the estimates
Our novel approach:
• a multivariate formulation that operates at the whole-brain level and adds spatial information via a mixed-norm regularization
• the use of stability selection to avoid the choice of the regularization parameter
• the AUC of the stability paths
What makes the AUC so interesting?
• depicts the probability of having a neuronal-related event at each voxel and time-point
• we make our estimation of activity based on these probability values
• robust estimates against the selection of the regularization parameters
The estimation of neuronal-related activity with this multivariate formulation and stability selection is
👍 MORE ACCURATE 👍
than that of multi-echo Paradigm Free Mapping (3dMEPFM) as shown by the dice coefficient, sensitivity, and specificity metrics.
The Python library will be soon available as part of the splora package.
In the meantime, you can use the univariate (voxel-wise) version that is already available in pySPFM 👇
github.com/eurunuela/pySPFM