The MRI prescanning process can sometimes seem random and/or arbitrary. Sometimes some prescan procedures are followed and other times, it can be a completely different set of procedures. The list of steps for a given scan can change based on the scan type or even the order of the scans. In order to help clear up any confusion and the frustration that can follow, I explain below some of the most common prescans that a scanner operator might encounter. I’ll explain what each prescan is, why it is needed, and when it is needed. Hopefully, after understanding these principles, the confusion/frustration you might be experiencing should be reduced at least a little.
Directional Coupler Prescan
Time required to run: ~1 second
In human MRI, the amount of radiofrequency energy transmitted by the RF coil and subsequently absorbed by the subject is a major concern. This metric is closely tracked to ensure that the scan session is conducted safely, and within FDA or IEC safety limits. Most scanner OEMs both calculate and monitor the RF power transmitted throughout the course of the scan session, and if the integrated power rises above the safety limits, the scanner will not run again until sufficient time has passed to allow for cooling.
The power calculation is a predictive model – the scanner software can predict to a very high degree of precision; how much power will be transmitted and when. So, it is possible to see how much power will be used for a given scan and how close to the safety limits the scan will get before the scan even starts. But there is another layer of safety in place to ensure that the predictive model is correct.
The amount of RF power transmitted is monitored and integrated in real time. The measured power is compared to what the model predicted, and if they don’t match, the scanner will shut down and require a service call to troubleshoot the problem. The device used to monitor the power is called a directional coupler which is commonly called the DiCo.
A DiCo is a remarkably simple device. There are many designs, but one of the simplest ones (depicted in the figure below) consists of two parallel loops of wire separated by their radius (aka, a Helmholtz coil) with the wire carrying the RF passing through the center of the two loops. The ends of the looped wire are connected to a voltage recording system. Any RF that passes through the center wire (which is the wire that conducts all the RF to the scanner’s RF transmit coil) induces a voltage in the loops, and is measured & recorded. Since the voltage measured is proportional to the RF amplitude, the RF power can be easily calculated in real time.
The first thing the scanner needs to do is ensure that it can accurately measure the RF power transmitted. If it can’t do that, then there is no point in continuing the scan session. The DiCo prescan is just about as simple as it gets. The scanner transmits a very low amplitude RF pulse, and measures the voltage from the DiCo. If the voltage measured doesn’t exactly match the prediction, then either the DiCo or something in the transmit chain is malfunctioning. Once the scanner has verified that it can correctly measure the RF power, it proceeds.
This prescan is run at the very start of every scan session, and is usually trusted to be valid for the entirety of that session.
Resonant Frequency Prescan
Time required to run: ~2 seconds
Virtually all MRI techniques require that the frequency of the RF used be set to match the resonant frequency of water. (This is assuming, of course, that the image or spectrum desired is a proton spectrum from/of a human subject.) Since the resonant frequency depends on the strength of the magnetic field, it can and does change as the field changes due to various things. Magnets drift over time. Magnetic fields are not perfectly homogeneous. Magnetic field shimming (see below) changes the field. Even physiological processes can affect the magnetic field.
All of these things affect the magnetic field experienced by the water hydrogen nuclei (protons) of interest. So, this is almost always one of the first prescans that need to be run. It is usually the first prescan to take place after the DiCo prescan. There are so many factors that affect the magnetic field that change from day to day, subject to subject, even scan to scan that this is a crucially important prescan to ensure the success of ALL the other scans that follow.
Some imaging and spectroscopic techniques are much more sensitive to errors in this setting than others. So, based on each scan’s susceptibility to produce off-resonant artifacts due to a poor frequency calibration, this scan might be run only once at the beginning of a scan session or for each scan. In general, echo-planar, spiral, and other non-Cartesian k-space trajectory scans tend to be most sensitive to off-resonant problems, and spin-echo-based scans are less sensitive. So, it is quite common for EPI scans to always run a frequency prescan, and fast spin-echo scans to trust a previous setting acquired earlier in the same scan session, if available.
A frequency prescan is typically run by the simplest pulse sequence – called pulse-acquire. A slice-selective RF pulse is transmitted at the last known good frequency at an intensity and duration that is certain to result in less than a 90 flip angle. After the RF pulse, the data are immediately acquired with no gradient activity at all. Those data are Fourier transformed into a frequency spectrum. Sometimes, the resulting plot is shown to the operator:
Typically, the spectrum will look similar to the one shown above. Water is usually the largest peak, and a set of peaks from fat will appear about 3.5 parts per million of the absolute frequency away. That is to say that if the resonant frequency of water is found to be exactly 123.0MHz (a very common frequency for water on human 3.0T scanners), the fat will appear 3.5 x 123,000,000 Hz x 10-6, or about 430Hz away. Note that by convention, NMR spectra are plotted with frequency increasing from right to left. As seen in the figure above, fat resonates at a lower frequency than water due to the relative differences of magnetic field shielding from chemical bonding experienced by the protons in water compared to fat. The analytical algorithm usually looks for the largest peak, verifies that there is a complex of peaks about 3.5ppm away in order to guarantee that the peak is indeed water, and sets the RF transmitter to use that frequency as the center.
B1+ Calibration Prescan
Time required to run: ~2-5 seconds, but in some niche applications, up to 2 minutes
After running the DiCo prescan (see above), the scanner knows how much RF power is transmitted, but it does not know how “effective” that power is at rotating (i.e., flipping, or nutating) the bulk magnetization. In other words, the scanner still does not know how much power is needed to affect a prescribed flip angle. For instance, if the sequence needs a 9 degree flip angle and it uses a 5ms long Hamming-apodized sinc pulse, how much power should it use?
The only way to effectively calibrate that parameter is to conduct an empirical measurement. This is called a B1+ calibration or sometimes referred to as a transmitter or RF calibration. There are many ways to perform this type of calibration. Most scanner OEMs provide different methods that can be used, depending on the application and the accuracy needed. But one of the fastest and most reliable methods, and the one used most frequently by default, is called the double-echo method.
The double echo process can be executed in different ways, and it varies across scanner OEMs. But they all can be boiled down to the acquisition of two different gradient echoes using two RF pulses of different intensities; usually the second is double the amplitude of the first. The simplest variant of this experiment takes the ratio of the two gradient echo signals. This procedure tends to cancel out complicating factors like T1, T2, and off-resonant effects. The ratio of the signal intensities is proportional to cos(a), where a is the flip angle. This gives the scanner a point of reference to use going forward.
It is now possible to further calculate what RF amplifier gain setting should be used for any RF pulse of any length to deliver any desired flip angle. This prescan is usually run once at the beginning of a scan session, and that calibration is trusted to be valid for the entirety of that session.
Shimming Prescan
Time required to run: 1 second to 3 minutes, depending on which method is used
Shimming is the act of homogenizing the magnetic field. It is of course beneficial to have the most homogeneous magnetic field possible for MR imaging. Magnets are shimmed using very thin strips of steel in the factory, and typically again at the time of installation if needed. But, that field is disrupted when a subject (or anything else) is placed in the bore. So, more fine-tuning is sometimes/usually necessary.
This fine-tuning is accomplished through the use of extra electromagnets wound into the case of the pulsed field gradients (aka, the gradients), and even the gradients themselves. If the field happens to have a distortion that slightly “tilts” the field from the front to the back of the scanner, that can be corrected easily. This so-called tilt amounts to a magnetic field gradient that results in a higher field in the front of the scanner (and lower in the back). The Z-gradient coil is designed to create a gradient in that direction, so pushing a little bit of current through it such that it generates a gradient of the same slope but the opposite direction, that standing gradient can be canceled out. The same goes for the X and Y directions.
Using only the pulsed field gradient windings to shim the field is called “linear shimming” or “low order shimming.” This method can be used to greatly improve image quality for a great number of imaging techniques. It can take less than a second to map the field in all three directions, and calculate the currents necessary to run through the gradient coils to cancel out the unwanted field gradients.
But some imaging sequences (e.g., EPI) are particularly sensitive to even small magnetic field inhomogeneities, and unrecoverable image quality degradation results when they are not corrected. In those cases, it is necessary to cancel out not just linear field gradients, but non-linear ones as well. The X, Y, and Z gradient coils are only capable of generating linear gradients, so a second set of non-linear gradient coils is needed. These are sometimes called the “high order shims.” These coil windings have some very unusual patterns, and are shown here:
Of course, setting the currents through these coils is not as simple as it is for the linear coils. A 3D map of the magnetic field must be acquired and analyzed to compute the currents for all coils. The 3D magnetic field map is low resolution but can still take more than two minutes to acquire.
Once a 3D fieldmap has been acquired, the shim settings should be usable for all identical scans that follow unless the subject moves a significant amount.
Parallel Imaging Prescan
Time required to run: ~1-2 seconds
Most modern imaging sequences make use of multiple receive coil elements to increase SNR and take advantage of an ever-increasing list of image acquisition acceleration techniques. One class of acceleration is colloquially referred to as parallel imaging. These methods speed the acquisition times by skipping parts of k-space. With less data to collect, the image takes less time to acquire.
Some of the most popular methods are well-known to most scanner operators. GRAPPA, SENSE, and CAIPIRINHA can all reduce image acquisition times by significant factors. They will all skip portions of k-space, and use different computational methods to reconstruct full FoV images. But they all need reference data to aid the reconstruction process.
The most common GRAPPA reference data prescan acquires some lines near the center of k-space, known as the Auto-Calibration Signal (ACS). SENSE makes use of receive coil sensitivity maps. But either way, some reference data needs to be acquired in order to reconstruct full FoV images from partial k-space datasets.
This type of prescan needs to be run for each unique set of slices or volume (for 3D scans) prescriptions. Once those data are acquired, they can be reused for all following scans in the session provided they are of the same scan type and have the exact same slices. However, some scanner OEMs will reacquire these reference data for every single scan that uses parallel imaging.
SMS Prescan
Time required to run: 5-10 seconds
Simultaneous multislice imaging, sometimes known as multiband imaging excites and acquires multiple slices at once. The resulting data, without specialized processing, results in what is called a “pancake image” which is essentially, the sum of all the images acquired. If the SMS factor is 3, then 3 slices are acquired simultaneously, and the pancake would include all three slices.
In order to “unfold” the data into discrete slices, the algorithm needs reference data very similar to that required for GRAPPA acceleration (see above). Reference data are acquired that are used to separate the images.
While in theory, it should be possible to re-use these reference data for successive identical scans, practically, it is better to reacquire these data for every scan, and that’s how all the OEMs (that I am aware of) actually perform this prescan.
Single-Band Reference Prescan
Time required to run: TR x SMS factor
When SMS is used for dynamic or quantitative imaging (e.g., fMRI), it is frequently helpful to have a set of the images acquired without the use of SMS acceleration for QA, comparison, and artifact characterization. That is exactly what is done in this prescan – perform the same scan exactly as prescribed, but without SMS. In that sense, this is not a prescan so much as it is a separate full scan used for checking. It is never run without the SMS feature selected.
In the example of fMRI, SMS is almost exclusively used to reduce the TR and thereby increase the temporal resolution. In that application, the TR selected is nearly if not exactly the minimum possible. So, the only way to acquire all the slices with the same TR but without SMS is to do so over several repetitions. For instance, if the SMS factor is set to 3, then one third of the slices (i.e., slices 1, 4, 7, …) are acquired in the first repetition. The second third (i.e., slice 2, 5, 8, …) are acquired in the second repetition, and the last third (i.e., slices 3, 6, 9, …) in a third repetition.
This type of prescan is optional, but when selected, it is run for each and every scan individually.
Normalization Prescan
Time required to run: 3-10 seconds
As noted above under the parallel imaging prescan section, the use of multiple coil elements in MRI is now pervasive. Nowhere is this more true than in the field of neuroimaging. Some scanner OEMs no longer even manufacture birdcage (single coil/channel) head coils anymore, except for niche and rare applications. In order to extract the most benefit from using multiple elements it becomes desirable to use as many elements as is practical. This pushes the physical sizes of each individual coil element smaller. Since a coil element rapidly loses sensitivity to signals more than a radius distance from the center of its coil, smaller elements reduce the sensitivity to deep structures. In neuroimaging, the consequence is that there tends to be higher SNR near the surface of the scalp than near deep brain structures. This leads to a shading artifact that can be visually distracting, and can complicate some anatomical analyses.
Prescan normalization helps to remove that artifact. It is accomplished by acquiring a low resolution image using the body coil (usually reserved for RF transmit only, built into the bore of the scanner). The body coil is constructed using a design very similar to the old birdcage head coils that did not suffer from this problem. The result is an image without the shading.
This image, combined with the results of the full scan that follows, is used to create a signal bias field map. This bias field is then divided out of the images of the full scan to create a normalized image virtually free of the shading artifact. A side-by-side comparison can be seen here:
The normalization prescan must be run for every scan. There is no sharing of normalization data from scan to scan, even if they are identical.
Dummy Prescan
Time required to run: 1-10 seconds, depending on tissue imaged, TR, and flip angle
For dynamic scans and functional scans, acquiring the same image over and over again should create a timecourse of images that demonstrate dynamic changes of a physiological process of interest. But there is a complication in the imaging itself. Beginning with the first scan, the bulk magnetization has not achieved a steady state. And since the signal is proportional to the bulk magnetization, the signal will change (falling) outside of any physiological process.
More senior neuroimaging researchers will be familiar with this phenomenon. The first several images of a functional scan would have higher signal than the rest, and each successive repetition would have slightly lower signal. It was common practice to either systematically remove that bulk signal drop from those images, or eliminate them altogether. Some scanner OEMs allowed users to set a number of dummy scans – repetitions at the start of the scan that were automatically eliminated and not saved as part of the series.
Most modern scanners will automatically calculate an optimal number of dummy scans based on the tissue being scanned, the TR, and the flip angle. So, the length of the dummy scanning process changes depending on these parameters. But typically, it should last more than 1-10 seconds, with the average being around 5 seconds. It will always be a multiple of the TR in length though.
This prescan will be run every time for all dynamic scans.