Hello! Thank you for everyone who studying and sharing the knowledge. I’m currently conducting multi-voxel pattern analysis. I extract data at 6 seconds post-stimulus peak from BOLD data and use it as a feature. I’m wondering if this approach is appropriate.
Since I’m not yet well-versed in fMRI analysis, my knowledge might be a bit confusing. I’ve come across using GLM and beta values as features after it, is my understanding correct?
Or there is any other way?(I heard about deconvolution.)
For reference, stimuli were presented at intervals of 8-14 seconds, and the duration was set to 4.5 seconds.
MVPA is a relatively advanced technique in fMRI analysis, so it might help if you learn about the more basics of fMRI first. The “Principles of fMRI” course on Youtube taught by Tor Wager and Martin Lindquist (Part 1, Part 2) is a popular resource for getting started. The MVPA specific videos are Part 2 modules 26-30, but you should review any videos before that will help you in understanding fMRI basics.
There is also a Nilearn tutorial in MVPA here: 5. Decoding and MVPA: predicting from brain images - Nilearn but you should first make sure you understand the basics of fMRI before starting it.
Thank you for sharing that!
Actually, my major is in machine learning, so I am attempting to perform machine learning using fMRI data. However, it is complicated and challenging to understand how to select features.
Most of the MVPA tutorials involve conducting machine learning using example data, making it difficult for me to comprehend the data structure. (like Is it the beta value of GLM, or the raw value after 4~6 seconds after onset?)