Perhaps, block design is the simplest and most frequently used experimental design for fMRI study. In a typical block design, there is a control condition block with one or more task blocks, and the control block and task block appear alternatively. The duration of each block is about 10–20 s. The length of these blocks can be equal or unequal. If the length of each condition block is equal, then it has additional advantage, i.e., Fourier analysis method can be applied straightforward because the stimuli change periodically.

Figure 2.1 shows one example of block design for emotional face processing. In this design, there is one white cross fixing control condition and three face/task conditions, and each block appears alternatively. The task conditions include neutral faces, sad faces, and happy faces. There are five faces with the same type of presentation within each block, and the duration of each face is 2 s.

In this experimental design, we can study the brain response to the face stimuli by comparing all the face condition response with the white cross fixing condition. In addition, we are able to address the emotional effect by comparing happy/sad face response with the neutral face condition. Finally, the question of the brain region difference in response to happy and sad faces can also be answered using statistical methods.

Ideally, it is always better to collect as many brain fMRI responses as possible. However, because subjects cannot maintain their positions in the scanner for a long time, we often take each functional run last 6–20 min. If more fMRI time slots are available, we can employ repeated run measure methods to collect more data from the same subject.

It should be mentioned that during the fMRI experiments, subjects are asked to perform certain task which may be irrelevant to stimuli but to help subject to keep awake in the scanner.

Another type of fMRI experimental design commonly in use is ER design. It has been used for many studies including, but not limited to multisensory, visual/auditory psychophysics, perception learn, and neuronal adaptation studies. As the name indicated, in the random design, each event or condition/block will appear randomly so that the subject cannot guess what the next stimulus is during the experiment. It is worthwhile to mention that the rapid ER design with randomly select interstimulus interval belongs to ER design, and it has been becoming popular for fMRI studies. When using this type of experiment design paradigm, because hemodynamic delay of the cerebral blood flow, it should be kept in mind to leave enough time interval between two events [3].

One example of random ER design paradigm which was used for emotional face processing study is shown in Fig. 2.2. In this ER design, we include four conditions, e.g., three task conditions (neutral, sad, happy face stimuli) and one control stimulus (white cross fixation) condition. The presentation of each task condition is randomly selected, and there is a white cross eye fixation control condition after each neutral, sad, and happy faces stimuli condition. The duration of white cross is varied according to the duration of the face stimulus. For instance, if a single face which takes 2TR (TR is the scanner repetition time, in this example experiment, we set TR = 2 s) to present, a white cross stimulus with 4TR duration is displayed subsequently. We represent this task as event block 1 in Table 2.1, and it takes 6TR to finish the task in total.

The second event block in Table 2.1 is a double face stimuli (same type of the stimulus presented two times, i.e., both happy/sad/neutral faces, but different person face) presentation, which needs 4TR (2TR + 2TR) for faces presentation and 6TR for white cross displays.

The third type of event block is a triple stimuli presentation, which includes 6TR (2TR + 2TR + 2TR) for the same type of face presentation and 7TR for white cross display. The role of the white cross is to separate face condition with other types of face stimuli tasks.

For each condition and each case, we have three repetitions. It is easy to see that the design is balanced, i.e., the total number of each condition is the same. We arrange each event from 1 to 27 as shown in Table 2.1, and then we randomly select each number; as a result, each event block will appear randomly. Therefore, the interval between each even block is randomly determined accordingly. The total time for this experiment is 8 min 42 s, as given in Table 2.1.

We may want to study the correlation between brain activation and behavioral task; we need to record the task performance during the MRI scan for the analysis. In addition, we may want to do additional behavioral tests after MRI scan, for example, in this experiment, during the fMRI experiment in the scanner, and we ask the subject to remember the face and corresponding name for a memory test after MRI scan.

To help understand the experimental design, we give one typical fMRI response (Fig. 2.3) from the experimental design paradigm as shown in Fig. 2.2. In Fig. 2.3, the blue thin solid curve denotes the normalized experimental design in Fig. 2.2, the dotted curve represents the fMRI response, and the red thin curve displays the model fitting for the response when comparing faces with white cross fixation condition (data is from Appendix D). It is obvious that the fMRI response follows the experimental design, suggesting that there is activation at this particular brain voxel position.

Another regularly used fMRI experimental deign is phase-encoded design. In contrast to block and ER designs, although phase-encoded design is less frequently in using, it has been applied successfully in visual retinotopic mapping studies [4–6] and proved to be another effective way to study visual cortex.

The idea of this type of design is to present a travelling wave (shift block) type stimulus in which stimulus changes continuously, without obvious interruption. For example, in visual retinotopic expanding ring stimulus, we change the position of the checkerboard pattern slowly over time, a flicker checkerboard color ring stimulus growing from small size (a) to big (d) as shown in Fig. 2.4. Then another small ring appears in the central of the stimulus to replace the largest ring stimulus in Fig. 2.4d, and the whole process repeated itself. This is the eccentricity stimulus for retinotopic mapping study. We also can create a contracting ring, in which the size of the ring changes from biggest (d) to smallest (a). Combining with phase-encoded wedge (polar angle) stimuli (Fig. 2.5), we can calculate the visual sign map and define the boundaries of the early visual cortex accordingly [4–6].

Figure 2.5 shows the polar angle stimulus for the retinotopic mapping study. Typically, it consists of a rotated wedge, whether clockwise or anti-clockwise rotation. When doing the experimental in the scanner, the subjects are asked to keep eye fixation at the center of the stimulus and perform certain task, i.e., detecting the regular black/white (Fig. 2.5a), blue/yellow (Fig. 2.4c), and red/green (Fig. 2.5b) patterns in the stimulus.

Based on the idea of phase-encoded design, we extended the ring stimuli to spatial frequency [7] and contrast sensitivity perception [8] stimuli for visual cortex studies. For instance, in the phase-encoded spatial frequency design, the spatial frequency changed periodically either from medium to low or from low to medium (Fig. 2.6) in which the spatial frequency of a sinusoidal checkerboard stimulus was gradually varied from 0.5 to 6 c/d over a 1 min period. The temporal frequency of the checkerboard stimulus was 8 Hz. This involved a smooth and gradual change in the spatial frequency of the sinusoidal checkerboard evenly throughout the field. A central fixation point was provided. The stimulus and design matrix are depicted in Fig. 2.6c. The attention of the subjects was controlled using a target detection task as described above for the block design. The advantage of this design is that the stimulus changes smoothly, rather than dramatically changes as in block design, and the fMRI response will change smoothly accordingly which is easier to deal with using Fourier analysis method. Because from Fourier analysis, we know that the periodical triangle stimuli (Fig. 2.6c) can be approximated by a set of sinusoidal functions. The fundamental frequency sinusoidal function of the triangle stimulus can be employed to model the fMRI response (Fig. 2.7). In Fig. 2.7, the dotted curve indicates the fMRI response (data is from Appendix E), and the solid curve denotes the fundamental frequency of the triangle wave (Fig. 2.6c) as a model for hemodynamic response function (HRF).

In a typical fMRI experiment, we collect not only functional data but also structural MRI data for brain activation localization, visualization, and comparison with other subjects. We may also want to compare different subject from different image modalities; as a result, we need to register different subjects to a standard template for comparison. To achieve these goals, we need to register functional data to a high spatial resolution structural image, which needs to be matched to the standard template for group comparison. Image registration methods [9, 10] for these purposes have been developed, and there is a lot of software packages freely available.

For functional images, we perform image registration to correct head movement during data collection within each run and across different runs. In the literature of image registration, many methods have been developed. For example, correlation method, linear and nonlinear least squares methods, and mutual information methods have been developed. In our study for fMRI data preprocessing, dynamic motion correction for functional image time series for each run and for different runs was realigned at the same time by using the fmr_preprocess within MINCtools (http:​/​/​www.​bic.​mni.​mcgill.​ca/​ServicesSoftware​/​HomePage) with default parameters for three-dimensional Gaussian low-pass filtering [11].

After motion correction, the next step is to normalize the fMRI time series to exclude the baseline shift in the analysis. There are two ways to normalize fMRI time series, i.e., temporal and spatial normalization methods. For the spatial normalization, we need to segment the brain region from the image background and using the voxels within the brain region for normalization. For the temporal normalization, we normalize the fMRI response longitudinally, i.e., if we employ x _i to denote the fMRI signal at time point (fMRI image frame) i, then we use
to normalize the signal, where is the mean value of the time series and is the standard deviation of the time series.

Figure 2.8a shows one example of fMRI image intensity time series from previous study [12] which adopted a typical standard block design (see Appendix F for data collection). Figure 2.8b shows the corresponding temporal normalization results using Eq. (2.1). The thin solid curve is the fundamental frequency model for HRF, and the thick solid curve denotes the fitted results. Comparing Fig. 2.8a with dotted curve in Figure 2.8b, it is easy to see that although the magnitude of these two curves is different (the baseline in Fig. 2.8b has been removed), the shape is similar, suggesting fMRI normalizing step maintains the temporal information of the signal changes. In Figs. 2.3 and 2.7, the solid curves represent the temporal normalized fMRI responses.

If the stimuli in the experimental design are changing periodically and the duration of each block condition is equal (Figs. 2.7, 2.8b, and 2.9), then fast Fourier transformation (FFT) method can be applied to model the HRF from fMRI response [15]. This is because the periodical square block wave can be approximated by the fundamental frequency of the sinusoidal wave (e.g., Fig. 2.8), and it is a powerful way to study experimental design with periodical change stimuli. The sine wave function has the form

where θ is the delay/onset or phase of the response that is estimated using the FFT method; ω is the angular frequency; ω = 2πf, where f is the frequency of the stimulus; and a is the magnitude. The algorithm to estimate parameters θ and a in Eq. (2.5) is as follows:

Based on this algorithm, we obtain the estimated HRF model displayed as solid red curve in Fig. 2.7 (six cycles), Fig. 2.8b (ten cycles), and Fig. 2.9 (six cycles).

Another widely used function to model the BOLD-fMRI HRF is two-gamma function (Fig. 2.10a, b). To estimate the two-gamma function adaptively, we need to estimate the delay/onset of the function by the FFT analysis (Eq. (2.5)) or cross correlation analysis. Then, a two-gamma function is built according to the following equation [16]:

Stay updated, free articles. Join our Telegram channel

Join