White Papers

Spatial Resolution of Flat Panel Detector CT Machines


Flat panel detector CT systems (also referred to as Cone Beam CT) have inherently high spatial resolution. This comes from the use of detectors with small pixels, sometimes down to 100 microns, and a small focal spot (sometimes down to 0.3mm). The typical resolution of e.g. the Fidex scanner is about 0.2mm isotropic or 25 lp/cm, and this is valuable in scanning small animals with sub-millimeter bones. It is far better than what human spiral CT scanners can deliver which is about 7 lp/cm – see e.g. ‘Comparison of Cone Beam CT vs. Multiple Detector CT by Plexar, February 2015’. They typically cannot go beyond about 0.65mm. Nevertheless, at a podium discussion at the recent 2015 ACVR Meeting, it was mentioned that FDCT machines have poor spatial resolution. So it is worthwhile to take a closer look.

Factors influencing spatial resolution in CT scanners.

We differentiate between intrinsic, or machine-related factors and extrinsic, or patient-related, factors.

  1. Intrinsic factors
    • detector and focus specification
    • scan modes setting detector binning
    • geometry factors given by focus radius and detector radius
    • omographic image reconstruction algorithms
  2. Patient related factors
    • patient motion
    • patient size
    • scan field and image reconstruction field of view
    • display field of view

Let’s examine the machine-related factors first. This is what the designer keeps in mind foremost, and Fidex has been designed to exploit the superior spatial resolution of flat panel detectors, as opposed to custom made detector elements as found in spiral scanners.

Simple geometry covers most of it.

Pixel Size: Flat panel detectors have pixel sizes of 100 to 150 microns. Projected to isocenter, this is 67 to 100 microns. It is fair to say that the detector is normally not the limiting factor for spatial resolution

Detector binning: this means summing up the signals coming from the small individual pixels. Fidex uses 2x2 binning giving us a real pixel size of 200 microns. In some modes (see below), we use 4x4 binning to 400 microns. Detector binning is done for many reasons, including readout speed, noise reduction and date size reduction. In most practical situations, 2x2 or 4x4 binning is applied.

Focus size: This is a different story. In most FDCT machines, spot size is 0.6mm optical which translates to 200 microns in isocenter. But note that this is only true for rays going through the center of the image at cone angle 0. As soon as one moves out radially, one has to take into account the real size of the focus when viewed from an angle. And going out in the cone direction, this becomes slightly asymmetric with an effective widening of the spot. This has been pointed out by Plexar. However, the cone angle in Fidex is only about 5 degrees, so the effect is not substantial.

Note that spiral scanners require much higher x-ray power due to the small solid angle extended by the detector arc and the short exposure time during rapid rotation. Higher power necessitates larger focal spots with width of about 1mm.This is another important factor limiting the spatial resolution of spiral scanners.

X-ray geometry: the two main geometric factors are focus radius and detector radius, meaning how far away from isocenter these components are mounted. Large distances allow higher spatial resolution but require more power and larger detectors. Shorter distances will limit spatial resolution. Fidex is a compact scanner with relatively short distances. In both cases (large gantry or compact gantry), the magnification (ratio focus-detector distance/focus radius) is roughly the same at about 1.5.

Tomographic image reconstruction: this is often the most important factor governing spatial resolution. The ‘algorithm’ or filter kernel defines the trade-off between spatial resolution and image noise. As long as the cut-off is within the mechanical limits discussed above, it is up to the user to define this trade-off. If high spatial resolution is the objective, one must use the ‘sharpest’ algorithm available, usually called ‘bones’. If soft tissue is being imaged, one is better off applying a soft algorithm, and that dramatically reduces spatial resolution. Therefore, if a measurement of spatial resolution is made, one has to pay close attention to the reconstruction algorithm.

So far, FDCT looks pretty good in comparison to spiral scanners. Fidex reaches 0.2mm resolution or 25 lp/cm without going to extremes in terms of focal spot size, detector binning and reconstruction kernel. However, how does this translate into practical animal scanning?

Patient-related factors. There are several factors coming from the patient that severely reduce spatial resolution. One is patient motion.

Patient motion. If scanning to 0.2mm is desired, the patient must not move more than that during the gantry rotation (about 10 sec minimum). Even in full anesthesia and with controlled breathing (e.g. using the ABC synchronized breathing developed by Animage), many patients will move. Only boney parts may be fixed tightly enough to make use of the full resolution capability of the FDCT machine. Heart motion, bowel movement, involuntary muscle spasms etc. will diminish it and in some cases lead to double contouring and streak artifacts. This is of great practical importance, and we train users to do their best in keeping the patient motionless during scanning. A spiral scanner may be less affected because of its faster data acquisition speed, but in the end motion artifacts are quite common here as well.

Another factor is more for the physics community to appreciate: Scatter.

Scatter. This happens when x-rays travel through the material of the patient. The larger the patient, the more scatter is produced, and the ratio of primary signal to (unwanted) scatter gets worse. This influences mainly image contrast which we will treat in a separate paper. But it will also negatively affect spatial resolution. It is like trying to see small stars in the night sky when there is light pollution from a city. How important is this effect?

By way of example, below we show 3 images of the resolution insert of a CT phantom taken with increasing amount of material around the phantom.

Figure 1: Phantom

The phantom itself is a Shelly Medical Imaging Technologies mCTP 610. It has a diameter of 70mm. Shown above is the slice containing the spatial resolution spirals.

We took 3 scans:

  1. with the phantom in air
  2. with the phantom placed inside a 13cm diameter Plexiglas ring
  3. with the phantom placed inside a 20cm diameter Plexiglas ring.

While we kept the algorithm the same (Bone), we increased the dose to partially compensate for absorption of additional material around the phantom. This happens also in animal scanning in Fidex: larger animals will be scanned with higher dose.

With very high attenuation, the spirals get lost in the noise and spatial resolution is compromised. This is what one will observe e.g. in a spine scan of a large dog: one can, with careful positioning, use the small field of view, high resolution mode to image the spine, but massive tissue around the boney structure will reduce spatial resolution.

Image 2 below shows an image of the phantom in air. The second spiral (0.2mm – upper left) is barely visible, the next two (0.3mm and 0.5mm) are clearly resolved. This resembles a head scan of a small dog or cat.

Figure 2: Phantom in Air

The next image shows the same phantom embedded in 13cm diameter of Plexiglas. This resembles a small dog spine scan. The dose has been increased by a factor of 2 over the first image to keep the pixel noise about the same. It appears visually unchanged, but measurement of the MTF shows a slight decline from the image taken in air.

Figure 3: Phantom in 13cm diameter Plexiglas

The next image shows the same phantom embedded in 20cm of Plexiglas. This resembles a spine scan of a 100 pound dog. Image noise is high (85 HU), and contrast is reduced.

Figure 4: Phantom in 20cm diameter Plexiglas

This is not surprising. As scatter increases, contrast is reduced, and the MTF goes down. However, quantitatively, the spatial resolution is still very good with about 0.3mm. Therefore, it is fair to state that while spatial resolution of FDCT machines suffers from scatter when scanning large animals, the effect is not big. Spatial resolution – even for large dogs - is still very good in Fidex, and better than in a spiral scanner.

Scan field and viewing field. I mention this under the category of patient-related parameters, because large patients are typically scanned and displayed with large field of view. It is possible to reconstruct e.g. a spine scan of a larger dog into a small field of view, both the scan field and the display field. In this case, spatial resolution is basically preserved and quite good at 0.2 to 0.3mm. However, if the user decides to image a larger field of view of 15cm or 23cm diameter, the convention of displaying 512 pixels across the field of view will limit the spatial resolution severely as shown below.

Figure 5: Phantom reconstructed into a 23cm field of view using 0.45mm voxel size. Resolution of 0.3mm structure is not possible

Note that this is not necessary to reconstruct into 0.45mm voxels, it is just a practical way to show an overview of the patient body. One can always reconstruct into a smaller field of view and thus preserve the spatial resolution.

In conclusion: FDCT scanners offer higher spatial resolution than spiral scanners. The practical limit is about 0.3mm, even for large fields of view and presence of scatter. Patient motion is most detrimental to image quality and can easily affect image sharpness. Care must be taken by the radiographer to set the imaging and viewing parameters correctly for high spatial resolution.

Comments welcome, please email to hbruning@animage.com