Flashcards
Section B
Flashcards do conjunto (25)
- 1Why is statistical analysis of fMRI data challenging?-massive amount of data
-signal is weak and noisy
-doesn't directly measure neural activity - 2What are the fundamentals of data analysis?-Plot the data on same scale for comparison (& remove outliers),
-Data averaging, to enhance signal-noise
-Detrend the data, if there is a slow varying component of the data
-Correlation - 3Explain slice timingAs all brain slices aren't collected at the same time, interpolating will synchronize slice. (fit curve and put data at correct time points)
- 4What two techniques help one tell which side of the image is left?1) place a pill behind the left/right ear of participant and log it
2) conduct a simple motor task with right or left hand - 5The rigid body assumption assumes the spatial difference between 2 brain volume data sets can be a accounted for by a rotation and translation. What does the rotation matrix look like in v_1=Rv_2+TR= ( cos(theta) -sin(theta)
sin(theta) cos(theta) ) - 6What is the minimized error function for re-alignment? How is realignment performed in the framework of multiple sessions?s=f(correlated brain volume -
reference brain volume)
The first volume of each session is re-aligned with the first volume of the first session. - 7Normalizing brain volume to the SPM EPI (echo planar imaging) template allows for inter subject comparison. How is the transformation performed to sets of data?The mean volume of each re-aligned data set is first normalized to the SPM template. The transformation is then applied to all re-aligned volumes within the set.
- 8What is the name of the translation performed in normalization ( a b| c d ), which allows rotation, translation, sheer, stretching, and shrinking?Affine transformation
- 9As a result of hardware a preprocessing the fMRI images are spatially noisy. What kernel performs spatial smoothing of the data?A gaussian kernel of a specified width (full idth at half maximum) for example FWHM=2.35sd
- 10What does the scaling parameter, a, represent?Strength of BOLD response to stimulation at corresponding voxel location. It reflects 'brain activity' in response to a stimulus.
- 11When is a t statistic useful over other statistical tests?-Small sample size
-Unknown population standard deviation - 12T tests can help determine model parameters. They can look at individual, differential, and interaction effects. What is the t-test formula for interaction?t=(a_2 - a_1)-(a_4 - a_3)/
se( (a_2 - a_1)-(a_4 - a_3)) - 13How can one modify the design matrix to take into account that people's reaction times are different for the same stimulus?Add the derivative of the model as a column in the design matrix.
- 14Neighboring voxels are not independent due to continuity in the data. What correction is necessary for a multiple dependent comparisons?Gaussian random field theory
- 15What are 3 possible reasons for different fitted parameters for different subjects?-variation due to measurement noise
-subject variation
-between subject variation - 16What are key equations of summary statistics in mixed effect analysis? What assumption is made.Y=X1(X2 B2+E2)+E1=X1 X2 B2+X1 E2+E1
=X1 X2 B2 +n2,
Cov(n2)=X1 C2 X1^t+C1 where
C1=Cov(E1) and C2=Cov(E2).
So X1 C2 X1^t ∝ C1, is an assumption. - 17What steps do you perform mixed effect analysis with, using SPM? (steps 1 & 2)1) Do 1st level analysis for each participant & obtain estimated model parameters (B)
2) Select appropriate contrast vector and apply to each participant - 18What steps do you perform mixed effect analysis with, using SPM? (steps 3, 4, & 5)3) Model estimated model parameter from 1st level analysis at 2nd level:
Bi^=[1]B^+ei
4) Enter the same contrast image in each subject into the 2nd level
5) Do a one sample t-test - 19What are some issues with all-in-one, mixed effect analysis?- parameter optimization is iterative and computationally expensive
-huge design matrix
-cannot look into individual subject results
-additional subjects means total re-analysis - 20Why is a long box car design not ideal?-human factor: subjects can't concentrate on a task for that long
-high pass filter is usually needed to remove low frequency noise - 21Why is a dense box car design not ideal?With hrf the baseline is never established.
- 22In event relate designs, what greatly increases efficiency?Randomisation, and more stimuli. This allows a baseline to be established, as well as the researcher gets more data.
- 23Can null events improve design efficiency? If so, give an example.Yes.
Event related design variance - 24What are the stages in performing selective averaging in order to estimate a haemodynamic response function?1) include null events randomly spaced with events of interest
2) Calculate mean time series for all events and non-events
3) estimated hrf=av event - av nonevent - 25What are some basis functions that can approximate the an unknown hemodynamic impulse response function?-Fourier basis functions (sines and cosines)
-gamma functions
Informações
- Descrição
- msc Brain imaging FlashCards sobre Section B, criado por samsden em 25-05-2016.
- Criado por
- samsden
- Visualizações
- 2
- Avaliação
Mais sobre fMRI and Functional Neuroimaging