Noise and artifacts in CT

1. Unit – 4
Artifacts and Noise in CT
Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 1

2. Beam Hardening
• CT uses a poly energetic x-ray spectrum, with energies ranging from
about 25 to 120 keV
• Furthermore, x-ray attenuation coefficients are energy dependent:
• After passing through a given thickness of patient, lower-energy x-
rays are attenuated to a greater extent than higher-energy x-rays are.
• Therefore, as the x-ray beam propagates through a thickness of tissue
and bone, the shape of the spectrum becomes skewed toward the
higher energies (Fig. 13-38).
• Consequently, the average energy of the x-ray beam becomes greater
("harder") as it passes through tissue. Because the attenuation of
bone is greater than that of soft tissue, bone causes more beam
hardening than an equivalent thickness of soft tissue
Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 2

3. • The beam-hardening phenomenon induces artifacts in CT because rays
from some projection angles are hardened to a differing extent than rays
from other angles, and this confuses the reconstruction algorithm.
• Most CT scanners include a simple beam-hardening correction algorithm
that corrects for beam hardening based on the relative attenuation of each
ray, and this helps to some extent.
• In a two pass algorithm, the image is reconstructed normally in the first
pass, the path length that each ray transits through bone and soft tissue is
determined from the first-pass image, and then each ray is compensated
for beam hardening based on the known bone and soft tissue thicknesses it
has traversed.
• A second-pass image reconstruction is performed using the corrected ray
values.
Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 3

4. Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 4

5. Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 5

6. Motion artifacts
• Motion artifacts occur when the patient moves during the acquisition.
• Small motions cause image blurring, and larger physical displacements
during CT image acquisition produce artifacts that appear as double images
or image ghosting.
• If motion is suspected in a CT image, the adjacent CT scans in the study
may be evaluated to distinguish fact from artifact.
• In some cases, the patient needs to be rescanned.
• An example of a motion artifact is shown in Fig. 13-40B.
Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 6

7. Partial Volume averaging
• The CT number in each pixel is proportional to the average µ in the
corresponding voxel.
• For voxels containing all one tissue type (e.g., all bone, all liver), µis
representative of that tissue.
• Some voxels in the image, however, contain a mixture of different tissue
types. When this occurs, for example with bone and soft tissue, the µ is not
representative of either tissue but instead is a weighted average of the two
different µ values.
• Partial volume averaging is most pronounced for softly rounded structures
that are almost parallel to the CT slice.
• The most evident example is near the top of the head, where the cranium
shares a substantial number of voxels with brain tissue, causing details of
the brain parenchyma to be lost because the large µ of bone dominates.
• This situation is easily recognizable and therefore seldom leads to
misdiagnosis. Partial volume artifacts can lead to misdiagnosis when the
presence of adjacent anatomic structures is not suspected (Fig. 13-41).
Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 7

8. Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 8

9. • There are several approaches to reducing partial volume artifacts, and
the obvious a priori approach is to use thinner CT slices.
• When a suspected partial volume artifact occurs with a helical study
and the raw scan data is still available, it is sometimes necessary to
use the raw data to reconstruct additional CT images at different
positions.
• Many institutions make use of interleaved reconstructions (e.g., 5-
mm slices every 2.5 mm) as a matter of routine protocol.
Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 9

10. Noise in CT
Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 10

11. • Compared with x-ray radiography, CT has significantly worse spatial
resolution and significantly better contrast resolution.
• Whereas the limiting spatial frequency for screen-film radiography is
about 7 line pairs (lp) per millimeter and for digital radiography it is 5
Ip/mm, the limiting spatial frequency for CT is approximately 1
Ip/mm.
Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 11

12. • CT has the best contrast resolution of any clinical x-ray modaliry.
• Contrast resolution refers to the abilty of an imaging procedure to reliably depict
very subtle differences in contrast.
• It is generally accepted that the contrast resolution of screen-film radiography is
approximately 5%, whereas CT demonstrates contrast resolution of about 0.5%.
• A classic clinical example in which the contrast resolution capability of CT excels is
distinguishing subtle soft tissue tumors:
• The difference in CT number between the tumor and the surrounding tissue may
be small (e.g., 20 CT numbers), but because the noise in the CT numbers is
smaller (e.g., 3 CT numbers), the tumor is visible on the display to the trained
human observer.
• As is apparent from this example, contrast resolution is fundamentally tied to the
SNR.
• The SNR is also very much related to the number of x-ray quanta used per pixel in
the image. If one attempts to reduce the pixel size (and thereby increase spatial
resolution) and the dose levels are kept the same, the number of x-rays per pixel
is reduced. For example, for the same FOV and dose, changing to a 1,024 X 1,024
CT image from a 512 X 512 image would result in fewer x-ray photons passing
through each voxel, and therefore the SNR per pixel would drop.
• It should be clear from this example that there is a compromise between spatial
resolution and contrast resolution.
Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 12

13. Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 13

14. Thank You!
Dr. S. P. Angeline Kirubha, Biomedical Engineering Department,
SRMIST
12/5/2021 14