GENERAL IMAGE

CHARACTERISTIC
S, AND IMAGE
RECONSTRUCTIO
N
AGENDA

 Quantitative aspects of assessing image quality: SNR, CNR, and spatial

resolution.
 Data acquisition: analogue-to-digital converters
 Dynamic range

 Image Artifacts

 Two essential algorithms for the reconstruction of medical images:

 Fourier transform

 Filtered back-projection (will be covered later in CT imaging)

2
QUANTITATIVE MEASURES FOR IMAGE
QUALITY
 A clinician making a diagnosis based on medical images looks for a number
of different types of indication. These could be changes in shape, changes
in image intensity within that structure compared to normal tissue, and/or
the appearance of features such as lesions that are normally not seen.
 Quantitative measures for assessing image quality are important to a
clinician in making his diagnosis.
 The three most important of these criteria are:

 Spatial resolution

 Signal-to-noise ratio (SNR)

 Contrast-to-noise ratio (CNR).

3
 Figure (a) shows a magnetic resonance
image with two very small white-matter
lesions indicated by the arrows. The
spatial resolution in this image is high
enough to be able to detect and resolve
the two lesions.
 If the spatial resolution was four times
worse, as shown in Figure (b), then only
the larger of the two lesions is now
visible.
 If the image SNR was four times lower,
illustrated in Figure (c), then only the
brighter of the two lesions is, barely,
visible.
 Finally, if the CNR between the lesions
and the surrounding white matter is
reduced, as shown in Figure (d), then 4
SPATIAL RESOLUTION

 The spatial resolution of an imaging system is related to the smallest

feature that can be visualized or, more specifically, the smallest distance
between two features such that the features can be individually resolved
rather than appearing as one larger shape.
 The concept of spatial frequency is very useful in characterizing spatial
resolution.
 A common measure in the spatial domain is the line spread function (LSF).

5
SPATIAL FREQUENCIES

 The spatial frequency of a particular grid of lines is measured as the

number of lines/mm, for example, 5 mm-1 for lines spaced 200 µm apart.
 The closer the lines are, the higher the spatial frequency, and the better
the spatial resolution the imaging system needs to be able to resolve each
line.

Grid patterns with increasing spatial frequencies going

from (a) to (c). 6
THE LINE SPREAD FUNCTION (LSF)

 The simplest method to measure the spatial resolution of an imaging

system is to perform the equivalent of the optician’s test. A single thin line
or set of lines is imaged made of the appropriate material for each
different imaging modality.
 Since the imaging system is not perfect, it introduces some degree of
blurring into the image, and so the line in the image does not appear as
sharp as its actual physical shape. The degree of blurring can be
represented mathematically by a line-spread function (LSF).
 The width of the LSF is usually defined by a parameter known as the full-
width-at-half maximum (FWHM).

7
THE LINE SPREAD FUNCTION (LSF)

 If two small structures in the body are separated by a distance greater

than the FWHM of the LSF, then they can be resolved as separate
structures as opposed to one larger structure.

Imaging results produced by

two different systems with a
relatively narrow (left) and
broad (right) LSF.

8
SPATIAL FREQUENCIES

For each medical imaging modality, a number of factors affect the resolving
power. Mainly, two general components:
 First: the instrumentation used to form the image,

 Second: the quantity of data that is acquired, i.e. the image data matrix
size.
o example for a 7 x 7 mm FOV:
1
mm

9
SIGNAL-TO-NOISE RATIO (SNR)

 In all measured or recorded signals there is some contribution from noise.

 Noise refers to any signal that is recorded, but which is not related to the
actual signal that one is trying to measure.
 In the simplest cases, noise can be considered as a random signal which is
superimposed on top of the real signal.
 As the noise level increases, the information content and diagnostic utility
of the image are reduced significantly.
 It is important in designing medical imaging instrumentation that the
recorded signal is as large as possible in order to get the highest signal-to-
noise ratio (SNR).
10
The effects of noise on image quality for an MR image. As the standard deviation
(σ) of the noise is increased (from top-to-bottom), features within the image
become indistinguishable. 11
SIGNAL-TO-NOISE RATIO
Two general factors that affect SNR are:
 If the noise is random, as in MRI, then the image SNR can be increased by
performing ‘signal averaging’ where a scan is repeated a number of times
and then adding the scans together. The overall SNR increases as the
square root of the number of scans. The trade-off is the additional time
required for data acquisition.

 The SNR in X-ray and nuclear imaging is proportional to the square root
of the number of X-rays and γ-rays detected, respectively. The trade-off is
the radiation dose to the patient, whose limits are controlled by various
government guidelines throughout the world.
12
Signal averaging to improve the image SNR. (a) MR image acquired in a single
scan, (b) two identical scans averaged together, (c) four scans, and (d) sixteen
scans.

13
CONTRAST-TO-NOISE RATIO (CNR)
 There should be a high enough CNR to be able to distinguish between
different tissues, and in particular between healthy and pathological
tissue.
 Image contrast is commonly defined as:

where CAB is the contrast between tissues A and B, and SA and SB are
the signals from tissues A and B, respectively.
 The CNR between tissues A and B is defined in terms of the respective
SNRs of the two tissues:

where σN is the standard deviation of the noise. 14

CONTRAST-TO-NOISE RATIO (CNR)

15
 IMAGE ARTIFACTS
 The term ‘artifact’ refers to any signal in an image that is caused by a
phenomenon related to the imaging process, but which distorts the image
or introduces an apparent feature that has no physical counterpart.
 There are many examples specific to each imaging modality: for example,
motion artifact in MRI, multiple reflections in ultrasound, and metal-
induced artifacts ‘streaking artifact’ in both CT and MRI from implants.

(a) Motion in MRI causes extra lines to appear in the image (arrowed), (b) acoustic
shadowing in ultrasound produces a black hole in the image (arrowed), and (c) a metal
16
implant causes ‘streaking artifacts’ in a CT image.
DATA ACQUISITION

 Nowadays, all modern hospitals are in the ‘digital age’, in which images
are acquired and stored digitally.
 They can be post-processed and filtered to maximize the diagnostic quality,
 Images acquired at different times can be quantitatively compared to
determine the effectiveness of a therapy,
 Images can also be transferred automatically to another hospital.

 The common requirement for producing digital images is an analogue-to-

digital converter, which takes an analogue input voltage from the medical
imaging system and produces a digital output.

17
DATA ACQUISITION: ANALOGUE-TO-DIGITAL
CONVERTERS

Data flow from different medical imaging modalities to produce a digital output.
18
ANALOGUE-TO-DIGITAL CONVERTERS: DYNAMIC
RANGE AND RESOLUTION
 The dynamic range of a photograph can be expressed as the number of
color levels or the number of graytone levels. The greater the number, the
more accurately subtle differences in color or graytone can be reproduced.
 The greater the number of levels, the higher the quality of the image.

 For an ADC, the dynamic range is measured in bits, and specifies the
number of different values that the output of the ADC can have. For an N-
bit ADC, the number of different output values is given by 2N.
 For example, an 8-bit ADC can have values from 0 to 255 (2 8=256),
whereas a 16-bit ADC can give values from 0 to 65535.

19
ANALOGUE-TO-DIGITAL CONVERTERS:
DYNAMIC RANGE AND RESOLUTION

Representation of an MR image with a maximum of 256, 16 and 4 graytone

levels, corresponding to 8, 4 and 2-bit ADCs.
20
ANALOGUE-TO-DIGITAL CONVERTERS:
DYNAMIC RANGE AND RESOLUTION
analogue signal 3-bit ADC
 The difference between these levels is
called the resolution, and its value is
given by the voltage range of the ADC
divided by the number of levels.
 The difference between the true 5-bit ADC 6-bit ADC
analogue input signal and the digitized
output is called the ‘quantization’
error, and this error becomes smaller
the greater the number of bits.

21
ANALOGUE-TO-DIGITAL CONVERTERS:
DYNAMIC RANGE AND RESOLUTION
What is the minimum voltage difference that can be measured by a 5
volt, 12-bit ADC?
 There are 4096 different levels that can be measured by the ADC,
with values from -5 to 5 volts (note that the maximum voltage of the
ADC refers to positive and negative values). Therefore, the
minimum voltage difference (resolution) is given by 10/4096 = 2.44
mV.

22
FOURIER TRANSFORMS
 The Fourier transform is an integral part of image processing for many image
modalities.
 In CT, filtered back-projection algorithms are implemented using Fourier
transforms.
 In MRI, the signal is acquired in the spatial frequency-domain, and the signals
undergo a multi-dimensional inverse Fourier transform to produce the image.
 In ultrasonic imaging, spectral Doppler plots are the result of Fourier
transformation of the time-domain demodulated Doppler signals.
 This section summarizes the basis mathematics and properties of the Fourier
transform, as relevant to the imaging modalities covered.
23
FOURIER TRANSFORMATION OF TIME- AND
SPATIAL FREQUENCY-DOMAIN SIGNALS
 The forward Fourier transform of a time-domain signal, s(t), is given by:

 The inverse Fourier transform of a frequency-domain signal, S(f), is given by:

 The forward Fourier transform of a spatial-domain signal, s(x), has the form:

 The corresponding inverse Fourier transform of a spatial frequency-domain signal or k-

space signal, S(k), is given by:

24
FOURIER TRANSFORM PAIRS COMMONLY USED IN
IMAGE AND SPECTRAL RECONSTRUCTION.

26