1.      What is preprocessing? How is it distinct from experimental analysis?

 Preprocessing is the computational procedures that are applied to fMRI data following image reconstruction but before statistical analysis. Preprocessing steps are intended to reduce variability in the data that is not associated with the experimental task and to prepare the data for statistical testing. Experimental analysis is the analysis done after preprocessing steps are taken to reduce variability in the data such as head movement and other artifacts in the image.  

 

2.   What is the first rule of quality assurance? 

The first rule is to examine your data. Many common artifacts are readily visible in the raw images even under cursory examination. You may do this as a time-series movie. This method of viewing the data may seem crude, but the human visual system is well equipped at picking up change over time, and with the time-series movie of MR data, one would be able to notice any glitches and big errors.

 

3.   What are phantoms? 

Phantoms are objects used for testing MR systems. Most phantoms are filled with liquids or gels with known properties, so that problems with the scanner system can be readily identified.The MR lab is supposed to scan a phantom before every testing procedure to verify that the MR machine is running properly. The team can compare the current scan with a previous one to make sure they are the same.

 

4.   Why is correction for time of acquisition within a TR important? 

Each slice is acquired at a different time point within the TR. Through small plane motions, as when a voxel moves from slice 12 to slice 13, the timing of activity will be off by one half of the TR value at affected time points. Thus slice timing correction first for interleaved sequences with long TR's will minimize timing errors.

 

5.   If you saw an fMRI activation map that was highly positively active along the edge of the right hemisphere, but highly negatively active along the edge of the left hemisphere, what would you conclude? 

You would conclude that there may be possible head movement of a millimeter or two. A movement toward the right would mean the edge of the right hemisphere would be flush in the gray area while the left edge of the left hemisphere would be in the CSF fluid.

 

6.   What can be done to help prevent head motion, in terms of designing the experiment and setting up the subject? 

One can partition the fMRI experiment into a number of relatively short runs to reduce subject fatigue. Also, you can set the experiment up so that the two conditions being analyzed are not at the very beginning and the very end. Plus, the experimenter can put the subject into a mock scanner and play real scanner noises to the subject and record the head movement to let the subject know how they did and also to gauge whether it is feasible to test this person in terms of reliable data acquisition due to head movement. Lastly, to prevent data loss due to head motion, onecan take additional slices on the adges of the area of interest. 

 

7.   What are some forms of head restraints used in fMRI experiments? 

Some head restsraint examples are: bite bars, masks, vacuum packs, padding, or taping. Bite bars are not popular among subjects, but they reduce head movement substantially. Subjects like vacuum packs, but the head is able to move forward. Taping does not restrain the subject very well, but provides a reference point for the subject to return to the original position depending on the tension in the tape.

 

8.   How should researchers interact with subjects to minimize head motion? 

Allowing the subject to be comfortable by talking to them or by having them participate in training sessions within a mock scanner may reduce head movement too. Researchers should ask questions such as “How are you doing?” after each run to reduce the subjects anxiety. Oftentimes between runs, the researcher will talk to the subject, but during this time, head motion occurs. If the subject is taped, then they will be able to return to the correct position based upon tape tensions, but otherwise this conversation may cause head motion.  However, this small bit of relaxing time helps the subject stay calm and may reduce head motion during the course of data collection, so there are pros and cons of subject resesarcher interaction.

 

 

9.   How do researchers correct for head motion? 

For motion correction, successive image volumes in the time series are coregistered to a single reference volume. Because the brain is the same in every image of the time series, a rigid-body transformation is used. Rigid body transformation assumes that the size and shape of the two objects to be coregistered are identical and superimposes them onto each other through translations and rotations. Spatial interpolation, the estimation of intensity of an image at a special location that was not originally sampled using data from nearby locations, would also be done. This is done so that you can estimate the values that would have been obtained had there been no head movement.    

 

10. What is the difference between magnetic field mapping and bias field estimation? 

Under what circumstances would one or the other be done? Magnetic field mapping is the collection of explicit information about the strength of the magnetic field at different spatial locations. Bias field estimation estimates the inhomogeneity using the distorted image itself. From the distored image itself, the magnetic field is estimated and a correction factor is applied to the original image. The low-signal region is then corrected in the final image. In short, magnetic field mapping uses a phantom to provide a field map while bias field estimation uses the distorted image itself to estimate the inhomogeneity.   

 

11. Why do researchers coregister images? 

Coregestering allows us to map our functional data onto high-resolution and high-contrast structural images from the same subject. We would want to do this for 2 reasons. The first occurs when the two types of images were acquired at different locations, either because different slices were wanted for each or because the subject moved slightly between their acquisitions. A second reason for function-structural coregistration is image distortion.

 

12. How large is the average adult human brain, in cubic centimeters?  

1300 cubic centimeters in volume.

 

13. What is segmentation?

Segmentation refers to the process of partitioning a digital image  into multiple regions (sets of pixels). These regions are usually GM, WM, and CSF.

 

14. What benefits are provided by spatial normalization? 

Spatial normalization allows us to compensate for shape differences of the brain by mathematically stretching, squeezing, and warping each brain so that it is the same as every other brain. This allows combination of data across individuals. 

 

15. Name two commonly used stereotaxic spaces for human MRI.

 Talairach space and the MNI template from the Montreal Neurological Institute (MNI)

 

16. What circumstances cause problems for spatial normalization? 

Normalization is based upon gross morphological features of brains and these features vary among individuals. Furthermore, these gross features do not necessarily indicate functional divisions between brain areas. Even when using cytoarchitecture instead of gross anatomy, boundaries between cytoarchitectonically distinct regions are highly variable.

 

17. Why might spatial smoothing increase the SNR of fMRI data? 

Because all subjects’ brain differ from one another in shape and size, and potentially functional organization, areas of activity are rarely represented in exactly the same voxels. Instead, combining data across many subjects distributes activity across a range of voxels and improves SNR. Smoothing not only increases the signal to noise ratio of each voxel, it also reduces the number of resolution elements that are assumed to be independent and used for correction of multiple testing.Spatial smoothing and signal averaging both serve to increase the SNR. Spatial smoothing, however, combines data from multiple voxels, instead of data from multiple trials, in order to get a more robust SNR.

 

18. What is the multiple comparison problem? Why is this more of an issue for fMRI than for behavioral studies? 

The multiple comparison problem is the increase in the number of false-positive results with increasing number of statistical tests. It is of particular consequence for voxelwise fMRI analyses, which may have many thousands of statistical tests. In short, if you set the significance at alph < .05, then there should be more than 5000 voxels active due to chance alone because a typical functional imaging volume has more than 100,000 voxels. If you don’t correct for multiple comparison, you will get random significance.

 

19. State and explain the principle of matched filters. What is its relevance for fMRI?

 Matched filters is the use of filters of the same frequency as the signal of interest. This provides maximal signal to noise ratio as you are filtering out only the frequency you want. 

 

20. What approaches have been developed to compensate for the multiple comparison problem?

 You can run a Bonferonni test or use some kind of filter to increase SNR.  

 

21. What is temporal filtering? Under what circumstances would it be useful? 

Temporal filtering allows one to selectively attenuate frequencies around a certain range while keeping intact other frequencies. This is useful when one wants to eliminate certain variabilies such as heart rates. Typical heart rates during an fMRI experiment varies, but are often between 1 to 1.5 Hz. For comparison, a typical experimental design might present alternating blocks of 12s of task and 12 s of rest, for a total presentation rate of .04 Hz. One can use a low pass filter to exclude frequencies above .2 Hz to remove physiological oscillations such as heart rate.