Utilisateur
Functional Magnetic Resonance Imaging
high spatial resolution, low temporal resolution
measures blood oxygenation (BOLD signal) to infer neural activity. Relies on the hemodynamic response.
Protons spin randomly in the body, the MRI machine creates a strong magnetic field (B0) that aligns the spinning of the protons. Now there are just up and down protons. The
next radio frequency pulse changes the alignment of the protons (makes them all go the same way). When the RF is stopped, the protons realign, and the energy that is released in the realignment can be measured with a second field (B1). B1 has a gradient to selectively target volumes.
- biased sampling account
The differential responses observed at the voxel level reflect the fact that each voxel samples cells arranged in cortical columns. Due to slight irregularities or fluctuations in the columnar maps, each voxel will, by chance, sample a slightly different number of each cell type tuned to specific features.
- macroscopic biases
macroscopic biases instead of local sampling biases.
Spatial preferences are influenced by global biases in processing across different regions.
MVPA analyzes patterns of activity distributed across multiple voxels simultaneously, rather than treating each voxel independently to directly study the link between mental representations and corresponding multivoxel fMRI activity patterns.
- spatial selection (deciding which voxels from the brain will be included in the pattern analysis)
- level of temporal aggregation (e.g. single volumes, trials, blocks, or runs)
1. Whole-brain classification
2. Region of interest (ROI) selection
3. spatial filtering, using wavelet pyramids
4. Searchlight analysis
- interpreting accuracies (under-/ overestimation, comparing brain regions)
- circularity and overfitting (classifier trains and tests on the same data/classifier is too complex)
- Interpreting classification maps (interpreting the effect of a single voxel)
- controlling for nuisance variables
Computational approaches to understand how cognitive representations are encoded in fMRI activity patterns
Decoding models predict aspects of a cognitive state or stimulus property from the measured brain activity pattern
Encoding models use the inverse direction of inference. They predict brain activity from stimulus properties or cognitive features
- small sample sizes
- small effect sizes
- large number of tested effects
- no standardization of design, definitions, outcomes, and analysis methods
- being a "hot" scientific field
- multiple comparison correction
- preregistration (of multiple comparison correction)
- more robust methods for statistical inference
- large shared data sets
- developing formal ontologies
- integration fo multiple neuro imaging techniques
- address ecological validity
- "Forward inference" inferring the presence of activation in each voxel given the presence of a particular term in the paper.
read attention, think PFC
- "reverse inference," where researchers infer a specific mental process based on the activation of a particular brain region, which can be problematic if the region is involved in many processes
read PFC, think attention
- Localizer: to quickly acquire low-resolution images used to position and orient subsequent scans.
- field map: measure magnetic field inhomogeneities (distortions B₀), to correct distortions during preprocessing.
- T1 - anatomical images, to align and overlay functional (BOLD) data.
- T2 - to measure bold signal and infer brain activity.
- Slice-time correction, because there is a time difference between slices
- Unwrapping, use fieldmap to calculate voxel displacement map
- Realignment, correcting for movement
- Coregistration, align anatomical and functional images
- Segmentation, calculatinjg different maps for different tissue types
- Normalization, to find out what is going on on a general level
1. Regressors of interest
2. Nuisance regressors
- slice
- render
- inflated
- layout
Trace white matter tracts, based on the diffusion properties of water in biological tissue
Blood Oxygenation Level Dependent signa
Also known as smoothing.
combining signals across multiple sensors or voxels in a specific way to enhance signal quality or isolate activity from a particular brain region or source.
a multivariate pattern analysis technique where a "searchlight" (a small cluster of voxels) is moved across the brain to identify local patterns that contain information about cognitive states or stimuli.
used for decoding!