Chapter 20 Digital Image Processing

Historical Development
 Discovery of computed tomography (CT) in the 1970s
 Introduction of computed tomography scanning created the concept of electronic data
sets for medical images
 The explosion of the imaging sciences that took place between the early 1970s and the
1990s was caused by the first wave of digital computerization in CT and diagnostic
medical sonography (DMS)
o Computed tomography captured the imagination of the public
 Digital processing has now been successfully applied to magnetic resonance imaging
(MRI), nuclear medicine, cardiovascular imaging, and mammography, and
predominates in diagnostic imaging
 Digital radiography imaging systems replace traditional film with a reusable detector

Two Types of Digital Radiography Systems

 These systems are divided into two types:

1.) computed radiography (CR): A type of indirect digital radiography; the
radiographer must usually move the detector, which is most often housed in a cassette,
between image acquisition and display
 Computed radiography systems use a photostimulable storage phosphor imaging
plate (PSP or IP), typically inside a cassette
 This cassette can be used in a Bucky or for mobile exams, similar to traditional
film-screen systems
 Once the PSP is exposed, the cassette is taken to a reader to process the plate and
create the image
 With these systems, there is a two-step process because the radiographer must
move the detector between image acquisition and display
 some PSP systems that are permanently fixed in units that do not require this two-
step process; however, they are not the norm
 2. digital radiography (DR): Imaging systems that replace traditional film with a
reusable detector
 Digital radiography (DR) systems typically have the detector and reader that are
a permanent part of a table or wall unit; therefore, a cassette is not needed
 Newer technology has now made some of these systems wireless
 and they can resemble a sealed cassette that can be moved around a room, from a
table Bucky to an upright unit
 With these systems, great care must be taken in moving the wireless DR unit as it
is a costly device
 With DR systems, the image is acquired and sent directly to the display monitor,
without the need for the radiographer to move the detector for the image to be
processed physically
 makes these systems faster and more efficient at creating an image
 DR systems can be categorized as either those that employ :
1. direct conversion
\indirect conversion
 Indirect conversion detectors use a two-part process involving a scintillator
(which converts incoming X-ray photons to light) and a photodetector (which
converts light into an electronic signal)
 Indirect systems include those used in CR (PSP technology)
 and those used in DR, including either a charge-coupled device (CCD) or
amorphous silicon with a thin film transistor (TFT) array
 The transformation of radiography from a film-based, analog medium to digital
imaging has necessarily brought about a redefinition of many of the basic principles of
imaging, changing the role of the radiographer
 The migration from film-based imaging to the digital technologies has been
hastened by changes in procedure reimbursements,
 which were implemented in 2017 by the Centers for Medicare and Medicaid
Services (CMS)
 Beginning in 2017, reimbursements for film-based x-ray procedures were cut
by 20 percent
 CR-based procedures will be reduced annually by 7 percent through 2018
and 10 percent thereafter
 DR procedures will receive full reimbursements and these payment
incentives encourage the transformation to true digital radiography using
flat-panel technology

2
  One standard that has been central to the development of modern radiological imaging
is the Digital Imaging and Communication in Medicine (DICOM)
 The radiologic community has promulgated this digital standard to ensure that all
equipment from all manufacturers who choose to adhere to the standard is
speaking the same computer language that can be understood across
manufacturer and device lines

Digital Image Formation

 Digital radiography imaging systems have replaced

traditional film with a reusable detector that produces a
digital image
 A digital image has been converted into numerical
values for transmission or processing
 To understand digital systems and the resulting digital
image, you must have some understanding of how
analog signal is converted to digital and how
computers manage data
 The electronic devices that produce radiographic images,
by design, produce analog signals
 These analog signals consist of a series of voltage
fluctuations on a continuous waveform with infinite values between an upper
and lower value
 Computers operate from a binary machine language; analog signals must be
converted to a binary language through a process called analog-to-digital
conversion (ADC)
 ADC consists of two distinct steps
1. sampling
 During sampling, the analog voltage values are measured at a chosen sampling
frequency on the analog waveform
2. quantification

 Each sampled piece of analog data is then computed and assigned a discrete
analog value, a process called quantification
 These analog values are then converted to a binary digit, using binary counting
methods

3
 Just as English has a 26-letter-symbol alphabet, the binary system operates with a two-
symbol alphabet
 Because electrical currents are most easily understood as being either on or off, the
binary system consists of information recorded as either a 0 for off or a 1 for on
 Each binary number is called a bit, for binary digit
 Nominal numbers function on a base 10 system instead of base 2
 An 8-bit word is required to form the 26 letters of the English alphabet
 an 8-bit word is called a byte
 Computer memory is often rated in terms of the total byte memory (rounded off)
- Example: a 10-megabyte magnetic hard disk will store over 10,000,000
bytes (80,000,000 bits) of information

Digital Image Characteristics

 Computerized digital images are described in terms of the number of

values displayed per image
 A matrix is a series of boxes laid out in rows and columns that
gives form to the image
 Each box of an image matrix will display a numerical value
that can be transformed into a visual brightness or gray levels
 The individual matrix boxes are known as picture elements or
pixels
 Each pixel location is determined by its address
 The total number of pixels in the matrix would be calculated by
multiplying the number of boxes in the row by the number of
boxes in the column
 Example: a matrix of 512 × 512 would have a total of 262,144
pixels that form the image
 In radiography, each pixel represents a two-dimensional data point
 These dimensions are along an x- and y-orientation
 In other imaging modalities (e.g. MRI), the data point has a
third dimension, creating a volume data point along the z-
axis

4
  That three-dimensional data point is referred to as a volume element or a voxel
 The overall dimension of the image matrix is called the field of view (FOV)
 In digital radiography imaging systems, the FOV is determined by the size of
the detector
 In other modalities, such as CT and MRI, the operator can select the FOV for a
particular study
 If the FOV remains the same, then as matrix size increases, the pixels must get
smaller, creating a sharper image
 If the overall dimensions of the FOV are 400 × 400 mm and the matrix size is
200 × 200, then every pixel would be 2 mm in size
 If the overall dimensions of the FOV remain
unchanged at 400 × 400 mm and the matrix
size is increased to 400 × 400, then every pixel
would be 1 mm in size
 The greater the matrix size for the same FOV,
the better the spatial resolution because a
larger matrix provides smaller pixels
 Spatial resolution or image sharpness is a critical image
quality factor that is determined by pixel size
 Spatial resolution is measured in line pairs per
millimeter (lp/mm)
 In digital imaging it can be no better than the size
of the pixel
 Each pixel in the matrix is capable of representing a wide
range of different shades of gray from pure white through
total black
 The number of shades of gray is determined by the
grayscale bit depth
 Grayscale bit depth ranges from 8 bits to 32 bits
 A grayscale bit depth of 8–32 equals a range of 1–4 bytes of storage required per
pixel in the image matrix
 A grayscale bit depth of 12 produces 2¹² gray levels
 This represents 4,096 different shades of gray that are available
 This is also known as density resolution

5
 Spatial Domains

 Digital images are represented in two domains, based on how the image is acquired
 Images in the spatial location domain are based on a matrix that has specific locations
for each pixel
 The spatial location is usually described as the pixel location
 the x coordinate representing the location of the pixel on a horizontal line from
left to right
 followed by the y coordinate representing the location of the pixel on a vertical
line from top to bottom
 Images in the spatial frequency domain are based on the number of cycles per unit
length
 which is the definition of frequency
 The spatial frequency is usually described as the resolution (object size) and the
contrast
 Small structures produce high frequency and high contrast
 Low frequency is produced by larger structures and low contrast
 The Fourier transformation (FT) is the mathematical algorithm that is applied to
change an image from the spatial location domain to the spatial frequency domain
 and an inverse FT can be used to return an image back to the spatial location
domain
 Physicians and radiographers view images in the spatial location domain
o whereas physicists and engineers can extract information from the
mathematical representations of the image in the spatial frequency domain

Image Processing Operations

 There are a number of computer operations that can occur to process the image

 These image processing operations are primarily intended to change the input values
of the pixels to improve the diagnostic quality of the output image
 They include point processing, local processing, and geometric processing

6
  Of these, point processing operations are the most important to understand for digital
radiography imaging systems

Point Processing Operations

 Point processing operations are those that are performed between the receipt of the input
image from the image receptor and the output image that is viewed on the monitor
 Point processing operations involve
adjusting the value of an input pixel
(point) to the corresponding output
pixel
 They are the most common
processing operations in digital
radiography imaging systems
 One common point processing
operation is called grayscale
processing, which allows for
adjustments to image brightness and
contrast
 This process involves the creation of a
histogram and the application of the
look-up table (LUT) and windowing

 A histogram is generated during initial

processing from the image data that allows
the digital system to find the useful signal
 by locating the minimum and
maximum signal within the anatomical regions of interest in the image
 To generate a histogram, the scanned area is divided into pixels and the signal
intensity for each pixel is determined
 The shape of a histogram will correspond to the specific anatomy and technique
used for an exam
 The system will perform a histogram analysis to determine the values of interest
(VOI) and the exposure indicator
 The VOI is used to locate the minimum and maximum exposure values for
the body part
 This histogram is compared
to a reference histogram
stored in the computer
 Each body part has a
reference histogram
that was obtained
using the correct
exposure factors

7
 If the exposure is outside the range from underexposure or overexposure, then the
computer will correct the image by shifting or rescaling the histogram to the correct area
 This is referred to as histogram modification or stretching
 A wide histogram demonstrates higher contrast
 A narrow histogram shows lower contrast
 If the histogram values are concentrated in lower comparative values, the image
appears dark or dim
 If the histogram values are
concentrated in higher
comparative values, the
image appears bright or
light

 The next step is to adjust the

contrast
 The digital detector has a
linear response that would
result in a low contrast
image if displayed without this step
 An LUT is applied to the data that has the standard contrast for that exam to give
the desired image contrast for display
 Selecting the proper projection ensures that the proper LUT will be applied when
the image is processed
 The LUT contains stored data to substitute new values for each pixel during
processing
 The proper LUT will provide the proper grayscale, regardless of variations in kVp
and mAs, resulting in consistent images
 If the exposure values are far outside the normal range, the system cannot
compensate and produce a diagnostic image
 Radiographers must set exposure factors that will produce images within normal
exposure ranges if a digital system is to function properly

 Windowing is a point processing operation that changes the contrast and brightness of the
image on the monitor
 The brightness
and contrast of the
digital image
depend on the
shades of gray,
which are
controlled by
varying the
numerical values
of each pixel
 The human visual
range
encompasses 32 or
fewer shades of

8
 gray, whereas the photon beam exiting the patient encompasses over 1,000
variations
 A 12-bit depth has 4,096 shades of gray
 Because the stored shades exceed the visual range, any digital image is only a
window of the total data
 The window width (WW) is the range of shades of gray that will be displayed
 A narrow width → few shades, high contrast
 A large width → many shades, low contrast
 The window level is the center of the window width and controls brightness
 A low level → image will be light
 A high level → image will be darker
 The computer can easily change the level and width by mathematical recalculations

 The quantity of information stored for each pixel varies depending on the grayscale (bit
depth) imaging abilities of the system
 The radiographer controls which portion of this information will be seen by
manipulating the data prior to sending it to PACS
 Radiologists can manipulate the data sent to them
 However, the more the technologist changes the original data before it is sent
to PACS, the less information the radiologist has to work with
 It is critical that radiographers be familiar with how the computer determines what
portion of the image information will be displayed
 Careful image processing is critical to avoid obscuring diagnostically critical
information

 Grayscale processing, both image subtraction and temporal averaging are considered
point processing operations
o These functions are used primarily in digital subtraction angiography

Local Processing Operations

 Local processing operations are those in which mathematical calculations are applied
to only a small group of pixels, although the operation will often be continued until all
pixels have been subjected to it
o This process is often referred to as a kernel
o A kernel is processing code that is mandatory and common to the computer
system
 It is normally applied over and over to the entire set of data being
processed
o Local processing is also called area or group processing

9
  Spatial frequency filtering is considered a type of local processing operation
o It is used to sharpen, smooth, blur, reduce noise, or pull elements of interest
from an image
o It can occur in the spatial location domain or in the spatial frequency domain
o When spatial frequency filtering is done in the spatial frequency domain, the
Fourier transform is used
o When it is done in the spatial location domain, the pixel values themselves are
used

 High-pass filtering (also referred to as edge enhancement or sharpness)

 uses an algorithm to convert the image into the spatial frequency domain
 then a high-pass filter is applied to remove low spatial frequency and produce a
sharper output image
 When frequencies that represent structures of interest can be identified, as in
contrast-medium–filled vessels, amplification of their frequencies can produce
an edge enhancement effect
 while other frequencies are actually suppressed
 The result is an image with edge enhancement and greatly increased contrast
 This type of high-pass
filtering is useful in
digital vascular
imaging and in digital
mammography
 extremely small
structures can
sometimes be buried
in an edge, and
undesirable edges are
usually enhanced as well
 Low-pass filtering (also referred to as smoothing)
 uses a similar process to intentionally blur the image, thus reducing noise and the
displayed brightness level of the pixels
 This process also decreases image detail, which is not desirable
 Therefore, low-pass filtered images should be viewed in a normal or high-pass
mode
 Unsharp masking (also referred to as blurring)

10
  subtracts a low-pass filtered (blurred) image from the original image, thus
producing a new subtracted and sharper image
 Spatial location filtering is another type of local processing operation, and is also
known as convolution
 A kernel is applied repeatedly to each pixel in a matrix in order to weigh the
values or apply a coefficient across the matrix
 Application of convolution kernels usually works better when array processors
can be used to rapidly apply them

Geometric Processing Operations

 Geometric processing operations are used to change the position or orientation of the
pixels in the image
 This allows rotation, magnification, and other operations similar to those used in
digital photographs

Digital Image Quality

 Digital image quality can be described in terms of:

 Spatial resolution
 Noise
 detective quantum efficiency
 artifacts

Spatial Resolution

 Spatial resolution is controlled by the matrix size and how many pixels can be
displayed by the monitor
 There is a direct relationship between matrix and spatial resolution
 When the matrix size is increased and pixel size goes down, spatial resolution
increases
 Matrix size is not typically variable within a system

11
  It is controlled by the design of image detectors and the accompanying
electronics built into the system

 Density resolution is the effect of the bit depth

 Bit depth refers to how much information can be held by each pixel in terms of
the number of grayscale levels that can be displayed
 Greater bit depth provides better density resolution
- Example: high-resolution mammography systems with a matrix of 4096 ×
4096 can provide 12-bit depth
 Contrast resolution is the ability of a detector to resolve different energy differences
striking it and transfer these energies into signal values, which are expressed as varying
grayscale levels
 Low-contrast resolution is a type of contrast resolution that deals with the
ability to visualize subtle energy differences, particularly in soft tissues
 These soft tissues are visualized as shades of gray and are often a visual
indication of underlying pathology
 CR and DR have superior low-contrast resolution compared to film-screen
imaging
 Improved low-contrast resolution is due to a greater dynamic range of the
image receptors
 This results in more gray shades and the improved visualization of
pathology on digital images

12
 Noise

 Noise can be classified as either:

 system noise
 ambient noise
 quantum mottle noise
 System noise is a result of undesirable signals from the digital system itself,
often through normal functioning
 System noise is random background information that is detected but does not
contribute to image quality
 It is congruent with audio noise, such as static white noise heard between radio
stations
 Background radiation also contributes to image noise and results in ambient
noise

 Noise is measured as the signal-to-noise ratio (SNR)

 A high SNR indicates little noise in the image
 Image noise has an inverse relationship to contrast
 Increased noise decreas es image contrast
 Increased image contrast tends to obscure or decrease noise
 The effect of image noise on brightness is irrelevant in digital imaging
 because the computer can easily compensate for a lack of exposure as
long as sufficient contrast exists to differentiate diagnostic data from
noise

 Quantum mottle noise results from an insufficient quantity of photons from

improperly set exposure factors (mAs and kVp)
 This produces a grainy image
 It is easily corrected by increasing mAs or kVp, depending on the clinical
situation
 Although these increases produce better-quality images, they also increase
patient exposure
 The correct exposure is a combination of image quality achieved with
reasonable patient exposure

13
 Detective Quantum Efficiency (DQE)

 A measure of the sensitivity and accuracy by which the image receptor converts the
incoming data to the output viewing device is known as the detective quantum
efficiency (DQE)
 A perfect device would perform this task with 100 percent efficiency (no loss of
information) or with a DQE of 1
 Digital radiography imaging systems have DQEs of 0.3–0.7 or 30–70 percent
 DQE is affected by the:
 SNR
 quantum noise
 system noise
 DQE can be described as the ratio of the squared
output signal-to-noise ratio (SNRo)² to the squared
input signal-to-noise ratio (SNRI)² of the imaging
detector
 It can be calculated using the formula
 DQE = (SNRo)² / (SNRI)²
 In addition to considering SNR, DQE is dependent on
 modulation transfer function (MTF)
 incident x-ray energy
 detector material
 spatial frequency
 Modulation transfer function (MTF) is used to measure the capacity of the detector to
pass its spatial resolution characteristics to the final image
 A perfect detector would have an MTF of 1
 DQE and MTF of a digital detector are directly proportional
 DQE and SNR have an inverse relationship
 High DQE values indicate that less radiation is necessary to achieve identical image
quality that a lower DQE detector would produce
o a digital receptor with a higher DQE does not always achieve improved image
quality at a lower radiation dose
o Many other factors affect image quality:

14
  incident beam energy
 signal sampling and processing
 quality of image display

Receptor Exposure
Exposure Indicators

 Exposure indicators reflect the amount of x-ray exposure to an image receptor, which
looks at the quantity of photons that strike the detector
 Exposure indicator numbers can be used as a
quality control tool for image quality and
radiation protection
 Departments have established target exposure
indicators (EIt), which reflect optimum
exposure and acceptable noise levels
 The correct exposure factors will deliver an
exposure indicator within an acceptable range and
result in the best image quality
 It is important for the radiographer to understand
the exposure indicator numbers for each image
and how to use this data to assess image quality

 Different manufacturers calibrate their systems

differently
 They use different beam qualities and different
parameters
 This makes it extremely difficult to compare the
sensitivity of one system to another
 When manufacturers update systems, new image receptors and software
enhancements can also cause sensitivity measures to change

 The exposure sensitivity of imaging detectors ranges from 0.1 mR to 100 mR, a range of
approximately 1,000:1
 If soft tissue and bone densities are imaged, one exposure will give the necessary
information due to the ability to change window levels
 To evaluate exposure, most manufacturers have an exposure indicator that provides
information on the average amount of radiation used to produce an ideal diagnostic
image

 Digital radiography processing systems use exposure indicators to provide information

regarding the exposure to the image receptor
 This information is useful to determine patient exposure and ALARA compliance
 There is no universal system at this time

15
  Different manufacturers use different systems and acceptable noise levels

Fuji, Konica, Philips CR Systems (S Number)

 These systems use a sensitivity number (S number)

 It is inversely proportional to the exposure reaching the imaging plate
 Example:
 An S value of 200 ≈ 1 mR exposure
 A higher S value = underexposure
 A lower S value = overexposure
 Fuji’s formula: S = 200 / (exposure in mR)
 S values relate back to film-screen speed systems
 A 400-speed system used lower exposure than 200-speed
 So an S number of 400 means less exposure was used
 Properly exposed imaging plates should produce S values of 150–250

Carestream (Kodak) CR – Exposure Index (EI)

 The Carestream system uses the exposure index (EI)

 EI is directly proportional to radiation striking the imaging plate
 0.1 mR = EI of 1,000
 1 mR = EI of 2,000
 10 mR = EI of 3,000
 Formula: EI = Log(exposure in mR) × 2,000
 Proper exposure = EI between 1,800 and 2,200
 If exposure changes by a factor of 2, EI changes by 300
 EI 2,500 → reducing mAs by 1/2 → EI becomes 2,200
 EI 1,500 → doubling mAs → EI becomes 1,800

Agfa – Log Median Exposure (LgM)

 Agfa uses log median exposure (LgM)

 Compares the exposure level of an image to a baseline
 A change of 0.3 LgM = dose changed by a factor of 2
 Uses scanned average level (SAL)
 Formula: SAL 200 = 1,214 × (exposure in mR)
 Agfa has speed classes similar to film-screen: 50, 100, 200, 400
 The selected speed class affects exposure factors
 Choosing the speed directs the system to prepare for the relative amount of light to
be read from the plate
 This allows the electronics to be optimized and results in a better image data set

 It is clear that there is no universal system for exposure indicators

 Different manufacturers use different systems
 The American Association of Physicists in Medicine (AAPM) and the
International Electrotechnical Commission (IEC) have released preliminary
information to standardize exposure indicators

16
  The most important concept is the deviation index (DI)
 Defined as an indicator of whether the detector response agrees with KTGT
 KIND = indicated equivalent air Kerma (radiation dose at detector surface)
 KTGT = target equivalent air Kerma (radiation dose for proper exposure)
 DI compares actual exposure vs proper exposure

 Exposure indicators provide a numerical and quantifiable measure of exposure

 These values are displayed as part of the DICOM data
 Most vendors can record and display both exposure indicator values and DI values
 Regardless of the system used, technologists must understand exposure indicators
and their relationship to exposure
 The ASRT White Paper recommends radiographers become familiar with
 the specific EI standards for their equipment
 and the new standardized EI and DI as they become available

Summary

 Digital radiography imaging systems replace traditional film with a reusable detector
 These systems are divided into computed radiography (CR) and digital
radiography (DR)
 A digital image is one that has been converted into numerical values for transmission
or processing
 A matrix is a series of boxes laid out in rows and columns that gives form to the
image
 The individual matrix boxes are known as picture elements or pixels
 Each pixel can represent a wide range of different shades of gray
 The number of shades of gray is determined by the grayscale bit depth

17
  There are a number of computer operations that process the image
 These operations change the input values of the pixels to improve diagnostic
quality
 They include point processing, local processing, and geometric processing
 One common processing operation is grayscale processing
 Allows adjustments of image brightness and contrast
 Involves creation of a histogram, use of a look-up table (LUT), and windowing
 A histogram is generated during initial processing
 It allows the digital system to find the useful signal by locating the minimum and
maximum signal within the anatomical region of interest
 An LUT is applied to the data that has the standard contrast for that exam
 Gives the desired image contrast for display
 Windowing changes the contrast and brightness of the image on the monitor
 Digital image quality factors include
 Spatial resolution
 Noise
 DQE
 Artifacts

 When matrix size is increased and pixel size goes down, spatial resolution increases
 Low-contrast resolution is a type of contrast resolution
 Deals with the ability to visualize subtle energy differences in soft tissues
 Noise can be classified as
 Electronic system noise
 Ambient noise
 Quantum mottle noise
 Detective quantum efficiency (DQE) is a measure of the sensitivity and accuracy by
which the image receptor converts incoming data to the output viewing device
 Exposure indicators reflect the amount of x-ray exposure to an image receptor
 Look at the quantity of photons that strike the detector
 Exposure indicator numbers can be used as a quality control tool for image
quality and radiation protection

18