General FAQ

 

What is microCT?

MicroCT uses X-ray imaging and computed tomography to produce 3D images of very high resolution, with voxel sizes down to 1µm or smaller. MicroCT, which utilizes differences in X-ray attenuation properties of materials to reconstruct 3D structure, differs from conventional CT by combining a much smaller field-of-view with a high resolution detector.

MicroCT is used to study diverse materials including bone, teeth, medical implants, snow, textiles, concrete and precious stones. MicroCT reveals in great detail the internal structure of these materials, such as the trabecular architecture within bone or grain within wood, allowing quantitative analysis of properties such as density and strength.

For more in-depth information about microCT imaging, please refer to the publication list.

 

What is resolution?

There are several different definitions of resolution including spatial resolution, also called 10% MTF resolution, and nominal resolution. Moreover, terms such as voxel size and detectability are also used. 

What is resolution?

Resolution is the distance between objects (or cavities) at which they can still be identified as independent from each other in an image (Figure 1). Resolution is usually given as a characteristic size (µm) or a spatial frequency (line pairs per mm). 

Figure 1
Lines in patterns with ridges and grooves smaller than the resolution can no longer be distinguished from each other as seen towards the right (center) in the pattern above.

What is a Point Spread Function (PSF)?

A PSF describes how a single infinitesimally small point in the source object is spread out (blurred, smeared) by the imaging apparatus. Typically PSFs have a gauss-bell like shape.

If the PSF is narrow, this indicates less blurring and thus that small objects can still be discerned (better resolution) As two points are brought closer together, their PSFs begin to overlap and eventually they become difficult to discern (Figure 2 and Figure 3).

Figure 2
Two Point Spread Functions (PSF) with little overlap. The two source points would still be distinguishable in a final stage.

 
 

Figure 3
 Two Point Spread Functions (PSF) almost entirely overlapping. The two source points would probably no longer be distinguishable in a final image when this close together.

What is the Modular Transfer Function (MTF)?

This is the Fourier transform of the PSF. It shows how well different spatial frequencies (line pairs per mm) are transferred to the final image (Figure 4). Typically, MTFs drop off with increasing spatial frequency e.g. as objects get smaller and closer together.

Figure 4. Measurement Transfer Function (MTF) with the 10% MTF level shown as a dotted line. All spatial frequencies with intensities below 10% will be below the resolution of the system.

How is spatial resolution determined?

Generally, two different methods exist:

  1. Measure the PSF of the instrument, determine its Fourier transform (MTF) and see at which spatial frequency it drops below a given threshold. Usually 10% of the maximal value is chosen which correlates well with average visual impression by eye (Figure 4).
  2. Image objects of known geometry such as a series of lattices with decreasing size. By eye, an operator determines the smallest recognizable object. Its characteristic size is then related to resolution. Although visually more intuitive, this method is also more subjective. 

What about the Signal to Noise Ratio (SNR)?

There are two general sets of factors determining image resolution.

  1. The scanner’s build
    Examples: detector resolution, X-ray source focal spot size, overall scanner geometry relative to the measurement object, precision of scanner movement during image acquisition
  2. Scan protocol and image reconstruction software
    The reconstruction influences the quality of what is visible to the user. For example, different image reconstruction kernels will smear features trading off resolution for noise suppression or using a finer image grid might give more detail but increase computation time and apparent noise. Also, not providing enough acquisition angles will hinder the reconstruction of high quality images.

What does voxel size mean?

 Voxel size is the size of a 3D pixel in the rendered image. This can be equated to the nominal resolution of the image.

What is nominal resolution?

Nominal resolution is the smallest possible voxel size of a scanner’s reconstructed image and not to be confused with image resolution or resolution at 10% MTF (see below).

What is the connection between image size and image resolution?

Obviously, the final image resolution can never be better than the voxel size. The voxel size is often chosen in correspondence with the X-ray detector’s pixel size. However, because of other factors, such as the finite spot size of the X-ray source or image reconstruction, resolution is usually further reduced with some information smeared out over neighboring voxels.

How does Scanco determine a scanner's resolution?

A solid aluminum cylinder is scanned and from the reconstructed image, the PSF is assembled from the gradient of the cylinder’s edge. From this PSF the MTF is calculated and the 10% level of maximum MTF criterion is applied. This provides a reproducible measurement of the entire imaging process which is independent of the operator’s level of experience. This method is analogous to ISO recommendations for medical CT.

What is detectability?

Detectability is the smallest size of an object that can still be seen in an image and is not a measure of resolution. When determining detectablility, the imaged object usually has a very high contrast compared to the empty background. The object may appear even if it is smaller than the resolution of the scanner or the voxel size. However, location in the image grid will affect the visibility and estimate of X-ray absorption (Figure 5). The higher the X-ray absorption of the object, the smaller the object can be and still be detected at the same SNR (Figure 6). For instance, a gold particle (high absorption) might be detectable while a glass particle (medium absorption) of the same size might no longer be detectable. 

Figure 5: Blurring of beads slightly smaller than the voxel size: The voxel volumes are only partially filled leading to an underestimation of the attenuation coefficient (partial volume effect) and inaccurate rendering of sizes (positions and real shapes left, rendered image right).

 
 

Figure 6: Beads of different attenuation and significantly smaller than the voxel size: Even though smaller than the voxel size, the bead with the higher absorption (black) is still detectable in the image. The bead with lower absorption (gray) gives a feeble signal and might be lost in noise (positions and real shapes left, rendered image right).

How does the SNR influence image quality?

Overall image quality is influenced by both signal-to-noise ratio (SNR) and resolution. Image quality is furthermore influenced by the original object contrast, on which the SNR depends: For example, bone and water will be easier to distinguish than two soft tissues with almost the same X-ray attenuation e.g. if the signal (tissue contrast) is smaller, SNR is smaller. Also, if there is a lot of noise in the image, object details might be obscured by the image noise e.g. if the noise is higher, SNR is smaller.

What influences SNR?

The signal is essentially the X-ray intensity detected by the image sensor. A stronger X-ray source or moving the detector closer to the source will increase signal intensity. Also, the difference in X-ray attenuation between the background and the object to be imaged will define the relevant signal level. Blood in vessels will have a stronger signal if a contrast agent is added. The noise in the image stems mainly from the statistical physical processes involved: X-ray generation, interaction with the imaged object, and detection. With longer observations (integration time) and more measurements (frame averaging) we become more certain of our detected values (mean value) and the noise is reduced (standard error). Instead of averaging over time and acquired frames one can average over image pixels. During the acquisition neighboring pixels can be binned together – 4 into 1, for example. On the down side, we would have a loss of resolution. Another option is to filter the image noise out during reconstruction. This often results in smeared or fuzzier boarders between objects representing a loss of image resolution but reducing the noise locally.

If you still have a question on this topic, please don't hesitate to e-mail us for more information.

 

What determines scan time?

Total scan time can be broken down into the time required for the scanner to collect data from the sample (scanning time) and the time required for the computers to reconstruct a 3D image from the raw data (reconstruction time). Several factors influence scanning and reconstruction time.

What factors determine the required scanning time?

  1. The resolution: standard, medium or high
  2. The number of full rotations to scan the desired sample length
  3. The integration time

What factors determine the required reconstruction time?

  1. The number of voxels within the volume-of-interest (VOI), which is determined by the resolution
  2. The computing power available to perform the reconstruction (# of and type of CPUs, available RAM, if the computer is part of a cluster)

A comprehensive document on this subject will be available for download soon.

If you still have a question about this topic, please don't hesitate to e-mail us for more information.

 

Explanation of structural indices of trabecular bone analysis.

Prefixes

VOX

based on counting voxels

DT

based on distance transformation (filling structure with spheres)

TRI

based on triangularization of surface (thus one more interpolation step in comparison to VOX)

Indices

TV

total volume [mm^3]

BV

bone volume [mm^3]

BV/TV

relative bone volume [1] (Percent)

Conn.D.

connectivity density, normed by TV [1/mm^3]

SMI

structure model index: 0 for parallel plates, 3 for cylindrical rods

DT-Tb.N

trabecular number [1/mm]

DT-Tb.Th

trabecular thickness [mm]

DT-Tb.Sp

trabecular separation = marrow thickness [mm]

DT-Tb.1/N.SD

standard deviation of local inverse number [mm]

DT-Tb.Th.SD

standard deviation of local thicknesses [mm]

DT.Tb.Sp.SD

standard deviation of local separations [mm]

DT indices are calculated using distance transformation without assuming anything about the shape of the bone (i.e. without plate model assumption). SDs: with the DT operation, a local thickness/separation for every voxel within bone is calculated. A histogram of local thickness/separation values can be obtained, and a mean and SD of this distribution is calculated. [Explanation for Tb.1/N.SD: First answer: forget about it, take Tb.Sp.SD. Detailed answer: For DT-Tb.N, the histogram is actually of the local separation of the skeletonized structure, thus 1/N. The mean value can be inverted to give Tb.N, but the SD only makes sense as Tb.1/N.SD]

Mean1

Mean voxel values of everything within volume of interest (mixture of bone and background). If scan was calibrated for bone, then the mean voxel value is in units of Hydroxyapatite density [mg HA/ccm], otherwise in linear attenuation coefficient [1/cm], Hounsfield [HU] or native file number [1] units

Mean2

Mean of segmented region, thus lin.att. only of what was considered bone

Mean3

Mean of additional region (if applicable)

TRI-BS

bone surface [mm^2]

TRI-Tb.N
TRI-Tb.Th
TRI-Tb.Sp

trab. number, thickness, separation, this time derived from the surface ratio, assuming that the bone is made of parallel plates (MIL method). Corresponds to traditional 2D histomorphometry, but this plate-model assumption leads to a bias in most cases. Scanco recommends to use DT-Tb.N,Th,Sp for truly 3D results.

TRI-DA

degree of anisotropy, 1= isotropic, › 1 anisotropic by definition DA = length of longest divided by shortest H-vector

TRI-H1

shortest vector of the MIL tensor, H1x, H1y H1z its components.

TRI-H2

longest vector of the MIL tensor

TRI-H3

intermediate vector of the MIL tensor

TRI-|H1| etc

length of these vectors in [mm]

El-Size-mm

voxel size in mm, in x, y and z direction

 

Hounsfield units differences

Why are Hounsfield Units [HU] of Bone so Different?

Bone in standard clinical CT's typically exhibits a CT number of 3000 [HU]. In microCT (and XtremeCT) the values are closer to 5000-7000 HU. Why?

Usual Hounsfield Unit Definition

HU is defined by measuring the absorption of air and water and assigning them -1000 HU and 0 HU, respectively. These points anchor a linear look-up table for converting the scanner's individual absorption measurements (physical absorption coefficient in [1/cm]) into the HU units.


Implications

As long as the absorption values are between (or close enough) to absorption readings to those two points of definition (air and water), the absorption values expressed in HU will be very similar across different scanners. e.g. soft tissues will deliver very similar HU values no matter what the CT scanner is since all their absorption values are close to water.


Where the Problems Occur

However, physical absorption and the scanners readings depend not only on the material composition, but also on the x-ray spectrum (eg voltage, target, filter), the detector's non-linearity (eg scintillator, CCD), and to a certain extent the geometry of the scanner (cone beam collimation, etc). These are determined by the material's chemical properties, measurement protocol settings, and the manufacturers choices. A synchrotron beam CT @ 20kV without filter or collimation and a home built CCD will very likely give different readings than a full body cone beam clinical CT operating at 120kV with Cu/Al filtering and a very different detector array.
Using HU to match two scanners' outputs will work just fine around air and water in practice. The absorption readings of materials lying between (or very near) the absorption values of air and water will not be identical due to spectral differences, but the differences will be small enough to enable practical work once applying the HU scale.
However, when we go to materials with very different chemical composition and much higher levels of x-ray absorption, the spectral properties in particular will lead to very different absorption behavior. The linear HU look-up needs to be extrapolated way beyond its range of definition for these higher absorbing tissues. It turns out, that the mapping of one scanner's absorption readings behaves very differently to that of another scanner and that the HU for bone can therefore be anywhere between 2500 and 7000 HU depending on the scanner.

Solution

The solution is to apply a scaling defined within the range of interest when possible. This does not have to be a linear function. For bone, correlating the absorption with mgHA/ccm is very practical since the HA density is a measure of prime interest in bone research and most other bone materials (collagen, cells, water) are very low absorbing and therefore do not disturb the density measure significantly. In fact, with beam hardening and scatter correction (non-linear) applied, the absorption reading to mgHA/ccm can be linearized with little error over the typical range used.

Note

The geometry of the scanner is very important: a large cone beam angle scanner with poor source collimation will still capture a fair amount of the scattered x-rays. This creates a material and geometry dependent background in the imaging further altering the readings if uncorrected. We do not know to what extent this explains a lot of the differences in HU between single slice clinical and CBCT scanners, but skimming through some publications would indicate that this and spectrum are the two most significant factors. 

 

How high is the radiation dose in my in vivo scans?

 

The vivaCT's beam is well collimated to ensure that the scanner only exposes areas of the animal that are used in imaging. The shutter is closed whenever possible; no exposure is to take place without an image being acquired. You should pre-calibrate without the animal in the scanner before the scout-view and scan to make sure no unnecessary stray radiation exposure takes place.

In all scanner models we currently offer (including the vivaCT 40 and 75), radiation leaks are stopped right at the source by placing shielding already inside the scanner over critical areas (such as the sides of the x-ray tube) to minimize stray radiation. A number of parts are consider to have double function i.e. they are structural and the materials, thickness, and shape are purposely chosen to stop radiation. These measures minimize the stray radiation that might reach the animal inside and the operators outside the scanner.

The local radiation dose inside of the vivaCT 40 are available for all our customers in a presentation that can be found in the section that requires a customer log in under Technical Notes -> Optimization of Dose.


There have been a few publications mentioning the effects of dose on studies and they have come to the conclusion that no changes in bone architecture can be deteced. Please find the references below.

Julienne E.M. Brouwers, Bert van Rietbergen*, R. H. (2007).
No effects of in vivo micro-CT radiation on structural parameters and bone marrow cells in proximal tibia of wistar rats detected after eight weekly scans. 
Journal of Orthopaedic Research, 25(10), 1325-1332. 


Klinck, R. J., Campbell, G. M., & Boyd, S. K. (2008). 
Radiation effects on bone architecture in mice and rats resulting from in vivo micro-computed tomography scanning. 
Medical engineering & physics, 30(7), 888-95. 

 
 
 
SCANCO Medical AG | Fabrikweg 2 | CH-8306 Brüttisellen | Phone +41 (0)44 805 98 00 | Fax +41 (0)44 805 98 01 | info@scanco.ch | www.scanco.ch