Appunti esame Neuroimaging techniques, prof Vallesi

WHAT ARE THE ADAPTATION STUDIES?

Refractory effects, changes in the amplitude and timing of a response based on the characteristics of preceding responses, are not always a problem for fMRI. They could be exploited to understand brain functions through adaptation studies. By changing a stimulus feature (e.g., color, or shape), one could infer whether a given brain region codes for this feature or not. If adaptation is present (decreased response to the repeated presentation of stimuli): the region does not code for the changing feature as it "treats" the stimuli equally. If adaptation is absent (unchanged response or recovery from adaptation): the region codes for the manipulated stimulus feature, since, when a stimulus with a changed feature is presented, it treats that stimulus as new and does not adapt to it.

WHAT IS SNR?

When we present a stimulus, we hope to measure the pure response of the brain. But the signal we hope to detect is mixed with other irrelevant sources of variability.

the relevant measures for functional MRI (fMRI)?

THE MAJOR SOURCES OF NOISE?

There are five major sources of noise: thermal noise, system noise, motion noise, non-task related neural variability and cognitive and behavioural variability.

Thermal noise refers to fluctuations in MR signal intensity over space or time that are caused by thermal motion of electrons within the sample or scanner hardware. Thermal noise linearly increases with field strength (outside the brain). It depends on the voxel's signal amplitude: the higher the amplitude the less the additivity. It can be reduced by averaging and filtering.

System noise refers to variations in the function of the scanner hardware, gradient field variation, static field inhomogeneities due to imperfect shimming. An example of system noise is scanner drift, slow changes in voxel intensity over time, caused by the changes in the resonant frequency of hydrogen protons associated with subtle changes in the strength of the static field.

Head motion noise can be minimized through use of

restraints, such as padding, head mask or it's corrected in preprocessing. Head motion can be also corrected prospectively, shifting slices on the fly so that slice planes follow motion, but it this way you can never get raw data. Non-task related neural variability refers to potential neural processes assumed to be unsynchronized with external stimulus. Behavioural and cognitive variability refers to the fact that cognitive processes are not static and subjects may adopt different strategies.

WHICH ARE THE METHODS/SOLUTION TO IMPROVE SNR?

We can use filtering approaches. We identify the unwanted frequency variation and reduce power around those frequencies through application of filters, but we can also remove frequencies composing response of interest. In the presence of physiological and system scanner noise, the detection of BOLD signal can be improved with ICA or PCA. Trial averaging assumes that the MR data recorded on each trial are composed of a constant signal and a random noise. So

signal remains constant with averaging while noise decreases.

WHAT IS PREPROCESSING?

Preprocessing refers to a series of computational procedures that operate on fMRI data following image reconstruction but prior to statistical analysis. It has two goals: to remove uninteresting variability from the data and to prepare the data for statistical analysis.

One important aspect of preprocessing is data quality assurance testing. The first step is to examine your data. It can be done with visual inspection, looking at raw functional images, but it's not sufficient and researchers should apply other tests to evaluate quality. A frequent test is phantom test. Since phantoms are filled with liquids or gels with known properties, when they go under the scanner, problem with the scanner system can be identified.

During preprocessing, head motion correction is also performed through alignment, in which to image volumes are spatially aligned. For motion correction, subsequent image volumes in the

time-series are coregistered to a single reference volume. Because the brain is the same in every image, a rigid body transformation is used. In the first stage, registration, 3 rotation and 3 translation parameters that describe the rigid body transformation between each image and a reference image are estimated. In the second stage, reslicing, each image is re-sampled according to the determined transformation parameters. Another step of preprocessing is slice timing correction. To collect data from the entire brain, one approach is ascending-descending slice acquisition, with collection of data in consecutive order. Another approach is interleaved slice acquisition, in which collection is in an alternating order. Almost all fMRI scanning takes each slice separately. So each slice is scanned at a slightly different time. Slice timing correction shifts the data as if the whole volume was acquired at exactly the same time.

WHAT IS SPATIAL NORMALIZATION?

Spatial normalization refers to compensations of

differences in brains by mathematical stretching, squeezing and warping the brains to be normalized in order to be as similar as possible to a template brain. So it makes results from different studies comparable by aligning them to a standard stereotaxic space.

WHAT IS SPATIAL SMOOTHING?

Smoothing means blurring. It consists of applying a small blurring kernel across the image to average part of the intensities from neighboring voxels together. The effect is to blur the image and soften the hard edges, lowering the overall spatial frequency and hopefully improving signal to noise ratio. Each voxel after smoothing effectively represents a weighted average over its local region of interest. The advantages of smoothing are the increased SNR, the reduction of the effective number of comparisons and the improvement of comparisons across subjects, anyway it reduces spatial resolution.

WHAT IS SPM?

SPM is a statistical technique for the analysis of fmri data. It refers to the construction of spatially

extended statistical processes to test hypotheses about regionally specific effects. It is a massively univariate approach, meaning that a statistic (e.g. t-value) is calculated for each voxel, using the general linear model. The resulting statistical parameters are assembled into an image. Statistical parametric maps (SPMs) are image processes with voxel values that are, under the null hypothesis, distributed according to a known probability density function, usually the Student's T or F. They are interpreted as spatially extended statistical processes by referring to the probabilistic behavior of Gaussian fields.

WHAT IS THE GENERAL LINEAR MODEL?

The General Linear Model describes a response/data (y), such as the BOLD response in a voxel, in terms of all its contributing factors (xβ) in a linear combination, while also accounting for the contribution of error (e). univariate GLM model analyses each voxel's data separately. The model assumes, at every time point, the HRF is equal

to the summed version of the events active at that point. A researcher creates a design matrix specifying which events are active at any time point. One common way is to create a matrix with one column per event and one row per time point and to mark it with a 1 if a particular event, a stimulus condition, is active at that time point. The design matrix embodies all available knowledge about experimentally controlled factor and potential confounds. So the design matrix models aims to explain as much of the variance in Y as possible, in order to minimise the error of the model. Anyway this model can have some problems. First of all, BOLD responses have a delayed and dispersed form (a solution can be convolution model). Then, the bold signal includes substantial amounts of low-frequency noise. Then, the assumptions about the error, that is that errors are normally distributed, that is the same in each measurement point and that there is no correlation between errors, can be violated. SPM can perform.

An autoregressive model which is able to calculate the correlation between the error at each time point and the error at the previous time point and then remove it from the data. Another problem is related with physiological confounds, that are head movements, arterial pulsations, breathing, eye blinks, adaption effects, fatigue and changes in attention to the task.

Which is the role of contrasts in SPM?

Specific questions about the comparison of activation between 2 conditions may be asked regarding the data through contrasts. The simplest contrasts involves only one explanatory variable, to test activation in the first condition versus baseline, or can regard the difference between the first and second condition. After the definition of the effect via the contrast and the estimation of the variance of the effect, we test whether or not there is any evidence for the effect. That is, whether or not the effect differs from zero, using the ratio of the effect to its standard error, called the T statistic.

Anyway contrast may be tor f contrast depending on the nature of the question. A t-contrast is adirectional test of a one-dimensional quantity. A f-contrast is a non-directionaltest of multidimensional quantity.

WHAT IS THE MULTIPLE COMPARISON PROBLEM? HOW TO CORRECTFOR IT?

The multiple comparisons problem refers to the increase in the number of falsepositives (type I error) with the increase in the number of statistical tests. Thisis of particular consequence for voxel-wise fmri analyses in which thousands oftest are performed. So, it’s important to correct for multiple comparisons. Astudy made by Bennett and colleagues in 2010 on a dead salmon showed theimportance of correcting for multiple comparisons. The dead salmon wasscanned with fmri and no multiple comparison correcton was performed.Perfoming thousands of test it can emerge that some are positive althoughthey are false positive. So it’s important to correct for multiple comparisons.