Introduction to Magnetic Resonance Microscopy (MRM)

Home Introduction to MRM NIH Image Reconstruction Cross Sectional Analysis Data Sets and Reconstructions Additional Scan Information

Nuclear Magnetic Resonance (NMR)

NMR is a phenomenon associated primarily with subatomic particles, especially the nucleus which contains protons and neutrons. In this account, the behavior of the nucleus is considered in greatly simplified detail with a concentration on those aspects that can be directly related to magnetic resonance imaging (MRI) and magnetic resonance microscopy (MRM). It is the association with the nucleus that accounts for the "nuclear" part of nuclear magnetic resonance. "Nuclear" is often omitted for simplicity and to alleviate the fears it may create for those who do not understand its meaning. Magnetic resonance (MR) is equivalent to NMR.

To understand the "magnetic" and "resonance" parts of the process, one needs to know how the nucleus of an atom creates a magnetic field and how many fields in a volume of tissue can be manipulated to synchronously release energy that can be detected and analyzed. Hydrogen will be used as an example because it is the predominant element in water and thus, in biological tissues. It is the element on which MRI is dependent.

The nucleus of a hydrogen atom has a single positively charged proton that spins (precesses) about an axis; this motion is often likened to a toy top or gyroscope spinning on a pedestal. The motion of this positively charged particle creates an electromagnetic current. The magnetic properties of the nucleus and its behavior in relation to externally applied electromagnetic fields account for the magnetic part of nuclear magnetic resonance.

The tiny magnetic field associated with each proton is similar to a magnetic dipole. Under normal conditions, the billions of magnetic fields produced by precessing protons are randomly oriented and therefore distributed in such a way that their magnetic moments cancel each other and no useful signal can be detected. If, however, the tissue is exposed to a strong, stable magnetic field the precessing protons and the fields they create become aligned with the fixed magnetic field. The stronger the static magnetic field the more fields line up and the greater the potential energy. In a magnetic field, protons line up in one direction or the other, and many of these fields also cancel each other. There are, however, significant numbers of "unmatched" ("unpaired") protons, and although relatively small in terms of total number, they can be manipulated by the injection of pulsed energy (see below) to synchronously release signals that can be detected and analyzed.

The frequency of the released energy corresponds to the frequency at which the hydrogen protons precess. This is called the Larmor frequency; it is directly related to the magnetic field strength and a constant called the gyromagentic ratio. With a stable magnetic field of 0.5 Tesla the spin frequency is 21.28 MHz. If the field strength is 1.0 T the Larmor frequency increases to 42.5 MHz.

See http://bmrl.med.uiuc.edu:8080/cgi-bin/cgiwrap/~dmorris/larmor.pl

Once the fields are aligned, resonance can be created by introducing radio frequency (RF) pulses that permeate the specimen. When this energy is applied at 90 or 180 degree angles to the aligned spins of the protons it creates a torque effect and changes the spins and thus, the magnetic moments. After each pulse ends, there is a return of the spins and magnetic moments to their initial state, a process called relaxation. This is when energy is released in the form of radiated photons (small quanta of electromagnetic radiation). This is a resonance phenomenon and as such is dependent on the frequency content of the RF pulse. The RF that best drives the system and the resonance frequency are identical and both are at the Larmor frequency.

The production of RF energy and the detection of the resonance energy are done by coils that encircle the specimen. The received energy is amplified and computer analyzed, especially by fast Fourier transform (FFT) which gives precise information about signal phase, strength and frequency spectrum as a function of time. Important data that may be gleaned include Larmor frequency shifts, relaxation times that describe the rate that nuclear spins return to equilibrium (T1 relaxation time) and time constants that describes the rate of signal decay (T2 relaxation time).

The following web sites are highly recommended for more detailed considerations of magnetic resonance.

Joseph P. Hornak, Ph.D. The Basics of NMR

Wm. Faulkner, B.S.,R.T.(R)(MR)(CT)

Return to Top

Magnetic Resonance Imaging (MRI).

MRI is a method comonly used to produce section-like images of humans for the diagnosis of medical problems. It is based on the detection and analysis of magnetic resonance energy from specific points in a volume of tissue and the subsequent construction of gray-scale images. In digital images the 2D picture elements are called pixels; in MRI, where small volumes are dealt with, the pixels have three dimensions and are called voxels. When seen on a computer screen they appear as 2D pixels but their 3D nature can be easily appreciated when a computer reslices and reconstructs the volume in different planes.

The shades of gray in an MRI slice typically depict the abundance of hydrogen protons within field of a voxel. If part of the image is black (filled with black voxels) it indicates that there is little signal (as in bone or air); if the voxels are white, there is an abundance of hydrogen, as in fluid-filled compartments. Intermediate levels, as would occur in connective tissue and muscle, produce voxels that are intermediate and in the gray scale.

To obtain MRI images in humans, the subject is placed inside the bore of a strong cylindrical magnet (typically 0.5 - 2.0 Tesla) to align the magnetic fields of the precessing protons. A controlled series of radiofrequency pulses are then presented at 90 or 180 degrees to the aligned proton spins, the spins are changed (magnetic moments are flipped) and quanta of energy are released as they return to their pre-RF pulse state.

MR imaging is dependent on the detection of this resonance from very restricted areas within a large mass, and the assignment of the detected energy to a specific voxel in x, y and z planes. The key to understanding how this is done is to recall that the Larmor frequency is dependent on the magnetic field strength which can be varied to produce changes in the resonance frequency. Magnetic gradients also change the phase of the resonant signals arriving at the detectors, especially when they are introduced during the intervals between RF pulses. Thus, it is by the introduction of gradients at specific points in time and the analysis of phase and frequency data that makes spatial encoding possible. MRI machines have coils that apply qradients in three planes. The injection of electromagnetic energy into a single plane is used to produce a slice through the volume. To produce consecutive slices, the body is advanced in small increments so that each voxel can be assigned an x, y and z position within the volume.

The RF pulses that are used to apply torque to the precessing nuclei can have many different properties (duration, shape, and bandwidth), but the key is the inclusion of frequencies that will match the different resonance frequencies.

The voxel size (resolution) varies according to the volume of the tissue being scanned. The volume that a MRI scanner deals with is called the field of view; it is a 3D grid-like area assigned to the volume being scanned; it has grid element dimension such as 256 x 256 x 256. Voxel dimensions in each direction are determined by dividing the field of view by the number of grid elements in each direction. If the view were 256 x 256 x 256 mm and the array was 256 x 256 x 256, then the voxels would be isotropic, 1 mm x 1 mm x 1 mm. Clinical MRI systems are currently able to achieve voxel dimensions of less than 1.0 mm. Slice thickness can be varied such that voxels are not always isotopic.

The time required to make an MRI series for a portion of the body varies from about 20 - 90 minutes and during this time the patient must remain very still. The acquisition time is dependent on sequence repetition time, phase encoding steps, and number of signals to average which affects the signal-to-noise ratio.
Recommended web sites for more information on MRI.

The Basics of MRI.

The MRI Tutor Web Site
Ray Ballinger

How Magnetic Resonance Imaging (MRI) Works
by Todd A. Gould, RT-(R)(MR)(ARRT)

Return to Top

Magnetic Resonance Microscopy (MRM)

MRM is basically MRI at a microscopic level. MRI can be accomplished at almost any hospital but for acquiring MRM data sets of the type we have worked with, there are only a few facilities world wide. All of our scans have been done at the Duke University Center for In Vivo Microscopy.

There are a number of differences between MRI and MRM. Two major ones are specimen size and resolution. Tissue can be living or fixed, but living animals are difficult because they must be still during long periods of time or special techniques must be used to acquire images between breaths or heart beats. To obtain the best possible resolution (smallest voxels), the structure must be small enough to fit into a custom-made holders no larger than 6.4 mm x 6.4 mm x 6.4 mm. With a cubic, 6.4 mm field of view and a 256 x 256 x 256 array, the isotropic voxels are 25 µm on each side. Larger specimens can be used, but the larger the field of view, the larger the voxels. Fields of view smaller than 6.4 mm and grids of 512 x 512 x 512 have been used to try to improve resolution, but the results have been dissappointing.

Much stronger magnets are used with MRM than clinical MRI. As noted above, clinical MRI units have magnets in the 0.5 - 2.0 T range. In our studies, magnets were in the 7 - 9 Tesla range. The stronger magnets make more protons line up and create more potential energy.

Custom made coils encircle the specimen and function to produce the RF pulses and gradients. The RF pulses are broadband to cover a range of Larmor frequencies and the gradients are applied in three planes simultaneously. The gradients are injected immediately after each RF pulse so that phase changes are created for spatial encoding of voxels. The time required for a complete serial scan is much longer than the 20 - 90 minutes needed for clinical MRI. Our MRM scans of the temporal bone were done overnight and the scan times were on the order of 12 - 13 hours.

Return to Top

MRM of the temporal bone and ear


Our work on the ear has been done almost entirely with formalin fixed material. The temporal bones or isolated elements were removed from the skull and trimmed to fit into a small enclosure (ideally a 6.4 mm well in a plastic depression slide, as shown on right). Sealed plastic tubes made from 10 - 50 cc plastic syringes, and small seal-a-meal bags were also used to hold temporal bones. The middle ear cavity was filled with a contrast agent (Magnevist) and the tissues were also soaked overnight in this agent. When the air in the middle ear cavity is replaced with Magnivist, the cavity stands out in sharp contrast to the adjacent bony ossicles and walls of the tympanic cavity. Prior to scanning,the tissue was soaked in Magnevist overnight so it would permeate the fluid spaces of the inner.ear.


After placing an ear in a sealed containers and removing all air bubbles, the tissue was surrounded by the secondary coils (as shown on left) secondary coilthat produced RF pulses, applied electromagnetic gradients and detected the signals. These were then inserted into the bore of the huge magnets (as shown below) and overnight scans were made. The raw images and detailed information about the scan parameters were then imported and saved as Tiff files in NIH Image, a freeware program running on a Macintosh Computer. After all images were collected as "windows," they were placed in "stacks" and cropped. We used Macintosh computers because of their superior graphics and because the source code was available for making changes that allowed us (Art Keating) to write programs for quantitative analysis of the cross-sectional areas of spiraling structures.

Resolution can best be expressed by the dimensions of the voxel elements. In our scans, the isotropic voxels were as small as 25µm. It is important to realize that this does not mean that the resolution was 25 µm. If a membrane were only a few µm thick it would be represented by a single voxel or even 2 voxels if it crossed boundaries. Thus, The orientation and position of a thin membrane can be determine with good accuracy but the thickness of such thin structures cannot be assessed.


One of the real values of the serial images obtained with MRM is the perfect registry from one slice to the next; this makes reconstructions relatively smooth and asthetically appealing. MRM images are, however subject to some distortion, the further the voxels are from the center the field of view, the greater the error. This graph was obtained by calculating cross-sectional area measurements of an MRM scan of a tube with a known constant diameter.

There are several advantages of using MRM over other methods for the study of serial images through a body or organ. MRM requires little specimen preparation to analyze and generates pictures with reasonably good resolution. The non-destructivenature of the procedure makes it useful for studying structures whose anatomical location makes them difficult to visualize by other methods, for example, structures encased in bone. MRM allows the user to visualize anatomical relationships in a more complete fashion than conventional histology, especially because of a computer's ability to reslice the volume in any direstion. Also convential histological methods (especially dehydration) cause more shrinkage and distortion.


MRM scans can generate many types of information. 3-D reconstructions of objects can be generated by color coding the structure(s) of interest in individual slices of the stack. This is described more fully on the NIH Image page. Once three dimensional structures are identified, the length, cross sectional area and volume of these structures can be measured. Using MRM to measure volume is easier than using serial sectioning where it is difficult to align adjacent slices and membrane running at oblique angles are hard to evaluate.

Return to Top

MRM images of the ear

An MRM image of the cochlea is similar to a low power histological section where different structures stand out because of differences in density and staining properties. In the MRM image of the cat cochlea (above) the dense bone appears black because there are few hydrogen protons to give a signal. Less dense material, such as the nerve in the modiolus and the spiral ligament have voxels with different shades of gray. The contrast medium in the cochlear labyrinth has light voxels because of higher numbers of hydrogen protons. The voxels can be one of 255 shades of grey depending on signal strength.

Return to Top

MRM data sets

Cat Data Set Click picture to see QuickTime movie

A complete data set is the entire series of images extending from one side of the field of view to the other. Most of our tempoiral bone data sets had 256 images because the field of view was 256 x 256 x 256; the file size was generally over 10 MB . Once the raw data windows are put into stacks the the slices with no tissue images can be thrown away to reduce the size of the file. This explains why some of our data sets have fewer than 256 images in each plane. The QuickTime movie shows all of the images in the data set. These images are in the "original plane of acquisition" but slices in any plane can be reconstructed. All of our scans have isotropic voxels which means they are the same dimension in all three planes. This makes measurements and reslicing more straightforward.

Return to Top

Available public domain scans and reconstructions

The data sets available for non commercial use include a wide variety of mammals and a frog. They can be accessed by downloading the image sets from our Scans and Reconstructions for Downloading pages. Descriptions of the scans can be found on the Additional Information web page.

Return to Top

MRM scan parameters

With the MRM raw files (the images) are four files that give scan parameters. These are called:

1). Headfile - gives information on the species scanned, the field of view (fov), the date of the scan, and coments, such as the type of container;

2). Convert Info_auto - gives image dimensions, the name of the software program, and the number of slices;

3). Convert Info_fixed - Voxel size, image dimensions, and the convert program and

4). Dimensions - Software (usually VoxelView) and the Descriptor File (field format).

Return to Top

Computer and software requirements for displaying and analyzing MRM scans

See the Techniques for 3-Dimensional Reconstruction using NIH Image web page.

Return to Top