This week, I continued the “Principles of fMRI 1” course, on coursera.org. The lecture discussed “Phyisiology, Signal, and Noise”.
On this lecture, Martin Lindquist and Tor Wager tackled “Signal, noise, and BOLD physiology”, “fMRI Artifacts & Noise”, “Spatial and Temporal Resolution of BOLD fMRI”, “Experimental Design – Kinds of Designs” and “Pre-processing”.
It started with some definitions and differences between MRI, fMRI and BOLD fMRI, and how BOLD fMRI allows us to measure the metabolic demands of active neurons. It also pointed out that not all BOLD signals reflect neuronal activity, and that BOLD fMRI require some acquisition, analysis and pre-processing techniques to be valid for a specific context.
One of the biggest problems in acquiring, analysing and pre-processing data is spatial and temporal resolutions, which need to be taken into consideration prior to the experiments, so it can be included in their design in order to validate the outcomes.
Another common issues is group experiments, which require more complex alignment and normalisation techniques that allows us to compare and match the outcomes with the expected results.
Finally, they discussed the kinds of designs that we can create and how data is handled by the General(ized) Linear Model, the trade-offs between each design, the characteristics variables, and how all comes down together with pre-processing techniques that correct for noise and errors (i.e., design errors, experiment errors, result deviations).
Without a doubt, it was a somewhat difficult lecture to understand, mostly, because the examples are too abstract for me to understand the underlying processes that undergo these three phases (i.e., acquisition, analysis, pre-processing). However, it was enlightening and clear enough for me to understand better how fMRI is modelled to each experiment and how to design for general or specific results.