Journal of Medical Imaging

and Radiation Sciences

Journal de l’imagerie médicale

et des sciences de la radiation
Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41
www.elsevier.com/locate/jmir

Image Postprocessing in Digital RadiologydA Primer for Technologists

Euclid Seeram, RTR, BSc, MSc, FCAMRTa,* and David Seeramb
a
British Columbia Institute of Technology, Vancouver, British Columbia
b
Photography BB.com, Vancouver, British Columbia

LEARNING OBJECTIVES
On completion of this directed reading article, readers should be able  Histogram
to:  Look-up table
 Windowing
1. Define the term ‘‘digital image processing’’  Window width
2. State the motivation for developing digital imaging modalities  Window level
such as computed radiography (CR) and computed tomogra- 10. Explain the characteristics of spatial frequency filtering using
phy (CT) high-pass and low-pass filters
3. Describe the rationale for understanding the use of image post- 11. Describe the basic features of the commercial image processing
processing techniques in digital radiology tool Photoshop and explain how it can be used in medical
4. List examples of image postprocessing techniques currently used imaging
in digital imaging modalities 12. Define the term ‘‘compression ratio’’
5. Briefly explain the meaning of the term ‘‘image domain’’ and 13. Distinguish between reversible (lossless) compression and irre-
identify the most conspicuous difference between the spatial versible (lossy) compression
location domain and the spatial frequency domain 14. State the problems and advantages of irreversible compression for
6. Describe the characteristics of a digital image use in digital imaging
7. Identify and briefly describe five classes of image postprocessing
operations as stated by author Baxes *This article is a directed reading and provides the equivalent of
8. State the major difference between point, local, and global pro- 2 hours of continuing education that may be applied to your
cessing algorithms provincial credit system.
9. Explain each of the following:

ABSTRACT RÉSUMÉ
This article deals with several image postprocessing concepts that are Cet article traite de plusieurs notions de post-traitement d’images qui
now commonplace in digital imaging in medicine. First, the motiva- sont maintenant courantes dans l’imagerie numérique en médecine.
tion for the development of digital imaging modalities is described, fol- D’abord, la motivation pour le développement des modalités d’image-
lowed by a rationale for understanding image postprocessing operations rie numériques sont décrites, suivi par un raisonnement pour compren-
that have become common in radiology. dre les fonctionnements des post-traitements d’images qui sont devenus
courants en radiologie.
Second, the image domain concept is outlined with a focus on the char-
acteristics of a spatial location domain image including the matrix, Deuxièmement, la notion de domaine d’image est décrite avec une con-
pixels, and the bit depth. In addition, the transformation of the spatial centration sur les caractéristiques de la localisation spatiale du domaine
location domain image into the spatial frequency domain is described. de l’image, y compris la matrice, les pixels et la profondeur de bits. En
plus, la transformation de la localisation spatiale du domaine de l’image
The third topic addresses five classes of image processing algorithms
en domaine de fréquence spatiale est décrite.
including image restoration, image analysis, image synthesis, image
enhancement, and image compression.

* Corresponding author. British Columbia Institute of Technology, Medical Imaging, 3700 Willingdon Avenue, Burnaby, British Columbia, V5G 3H2, Tel.: 604-432-
8231, fax: 604-430-5443.
E-mail address: [email protected] (E. Seeram).

1939-8654/08/$ - see front matter Ó 2008 Elsevier Inc. All rights reserved.
doi: 10.1016/j.jmir.2008.01.004
The article continues with a detailed description of point processing Le troisième sujet examinera les cinq classes d’algorithmes du traite-
operations as well as local processing operations. The former is dis- ment d’images, y compris la restauration d’image, l’analyse d’image,
cussed in terms of the histogram, look-up table (LUT), and window- la synthèse d’image, l’accentuation d’image et la compression d’image.
ing. The latter describes spatial location filtering (convolution) and
L’article continue avec une description détaillée des fonctionnements
spatial frequency filtering using high- and low-pass digital filters, fol-
des processus ponctuels ainsi que des fonctionnements des traitements
lowed by a brief description of the unsharp masking technique.
locaux. Tandis que la première est examinée au moyen d’histogramme,
The fifth major topic presents essential features of the commercially de table de recherche et de fenêtrage, la deuxième décrit le filtrage de la
available image postprocessing tool Photoshop, with applications in localisation spatiale (filtrage numérique) et le filtrage de la fréquence
medical imaging and an emphasis on how this tool can be used by spatiale en utilisant des filtres numériques passe-haut et passe-bas, suivi
teachers and students alike in an educational environment. par une courte description de la technique de masquage flou.
Finally, a technical overview of image compression is reviewed with Le cinquième sujet principal a présenté les caractéristiques essentielles
a discussion of compression ratio and types of image compression tech- d’un outil de post-traitement d’images offertes sur le marché, Photo-
niques. In particular, irreversible compression is outlined briefly, and its shop avec des applications dans l’imagerie médicale et a mis l’accent
effect on the visual quality of images is demonstrated. sur comment cet outil peut être utilisé par des professeurs et de
même que des étudiants dans un environnement éducationnel.
Furthermore, a statement from the Canadian Association of Radiolo-
gists (CAR) on the use of irreversible compression in digital radiology En conclusion, un aperçu technique de compression d’image est exam-
is provided. The article concludes with a summary of image postpro- iné avec une discussion du taux de compression et des types de tech-
cessing as an essential tool for those who work in a digital imaging niques de compression d’image. En particulier, la compression
environment. irréversible est décrite brièvement et son effet sur la qualité visuelle
des images est démontré.
En outre, un énoncé de l’Association canadienne des radiologistes
(CAR) sur l’utilisation de la compression irréversible en radiologie nu-
mérique est fourni. L’article conclut avec un énoncé sur le post-traite-
ment d’images comme outil essentiel pour ceux qui travaillent dans un
environnement d’imagerie numérique.

Introduction Systems Inc, San Jose, CA). This software is a powerful tool that
can be used to illustrate image postprocessing to students study-
Image postprocessing belongs to the domain of digital image
ing digital imaging technologies.
processing, which is simply the processing of images using a dig-
The purpose of this directed reading is to outline the major
ital computer. Film-based radiology is now obsolete and has
characteristics of digital image postprocessing as it relates to dig-
evolved into various digital imaging modalities, including com-
ital imaging systems used in radiology. First the limitations of
puted radiography, flat-panel digital radiography, digital fluo-
film-based imaging will be described; and, second, digital image
roscopy, digital mammography, computed tomography (CT),
postprocessing operations that are commonly used in digital im-
magnetic resonance imaging (MRI), nuclear medicine, and di-
aging modalities will be discussed. Third, the essential features
agnostic medical sonography. Thus, digital image processing
of Photoshop will be highlighted for the purpose of illustrating
in radiology has become one of the routine skills of technolo-
selected features of this tool and how it can be used for teaching
gists and radiologists alike. In addition, the use of the CT
and learning. Fourth, a technical overview of image compres-
and MRI scanners has become an integral imaging component
sion and its importance in digital imaging will be presented.
in radiation treatment planning. Therefore, it is important that
technologists understand the nature and scope not only of dig-
Motivation for Developing Digital Imaging Modalities
ital images but also of digital image processing, to become effec-
tive and efficient users of the new technologies that have made Film-based imaging, or film-screen radiography, as it is often
a significant impact on the care and management of patients. referred to, has been used routinely ever since the discovery of
The major goal of digital image postprocessing in medical x-rays in 1895. A radiographic image is created using a number
imaging is to alter or change an image to enhance diagnostic in- of steps. First, the patient is exposed to a predetermined amount
terpretation. For example, images can be postprocessed for the of radiation to provide the required diagnostic image quality.
purpose of image enhancement, image restoration, image anal- The latent image formed on the film is rendered visible through
ysis, and image compression. These operations are intended to chemical processing. The processed image is a hard-copy film
transform an input image into an output image to suit the view- image with a fixed optical density, used by the radiologist to
ing needs of the observer in making a diagnosis. In addition, make a diagnosis of the patient’s medical condition.
these operations can also be used in digital photography and Problems in obtaining the optimum image are often a chal-
have become available in commercial image processing software. lenge in film-screen radiography. If the radiation exposure is too
One such popular software package is Adobe Photoshop (Adobe high, the film is overexposed and the processed image appears

24 E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41
too dark. Alternatively, if the radiation exposure is too low, the gained from NASA’s space program were applied in several
processed image appears too light and cannot be used to make other fields such as biology, defense, forensics, photography,
a diagnosis. Specifically, these images lack the proper image remote sensing, and medical imaging, to mention only a few
density and contrast, and would have to be repeated in an effort [2,3].
to obtain acceptable image quality needed to make a diagnosis. Today, digital photography is commonplace, and several
In this situation, the patient receives an increase in the radiation types of image postprocessing software are now available com-
dose [1]. mercially for users to become actively engaged in enhancing
There are additional shortcomings associated with film- their photographs [4]. One such software is Adobe Photoshop
screen radiography. The film-screen image receptor is not the Essential, the features of which will be highlighted in this
ideal medium to perform the functions of radiation detection, paper.
image display, and image archiving. As a radiation detector,
film-screen cannot show differences in tissue contrast that are Rationale for Understanding Digital Image Postprocessing
less than 10%. This means that contrast resolution is limited. in Radiology
However, film-screen radiography offers the best spatial resolu-
tion compared to any other imaging modality such as CT, MRI, Why is it important for technologists to understand the na-
or nuclear medicine, and this is one of the main reasons why ture and scope of digital image postprocessing? The answer will
radiography has played a significant role in imaging patients [1]. become clear in this section of the directed reading. First, it is
As a display medium, the optical range and contrast for film important to understand the components of a typical digital
image are fixed and limited. Film can only display once, and the imaging system, and then to appreciate the different types of
optical range and contrast are determined by the exposure tech- specific image postprocessing operations used in the various
nique factors used to produce the image. To change the image imaging modalities.
display (optical range and contrast), one has to repeat the expo-
Components of a Digital Imaging System
sure using a different set of technical factors (mAs and kVp),
thus increasing the dose to the patient [1]. As an archive me- The major components of a generic digital imaging system
dium, film is usually stored in envelopes and housed in a large for use in radiology are shown in Figure 1. The essential
room. It requires manual handling for archiving and physical components include data acquisition, image processing, image
retrieval by an individual. display/storage/archiving, and picture archiving and communi-
These limitations provide the motivation for developing dig- cation systems (PACS).
ital imaging systems for use in digital radiology because of their It is not within the scope of this article to describe these com-
potential for image postprocessing, which can overcome these ponents in any detail; however, the following points are in
problems. order:
 In data acquisition, raw digital data are collected from the
patient.
Historical Perspectives
 The raw digital data are processed by a digital computer for
Digital image processing underwent significant develop- subsequent image display to a human observer. Such pro-
ments at the Jet Propulsion Laboratory at the California Insti- cessing is referred to as digital image processing.
tute of Technology (Caltech) and was subsequently used by  Digital image processing is divided into two parts: prepro-
the National Aeronautics and Space Administration (NASA) cessing and postprocessing. While preprocessing operations
in a wide range of space exploration applications. The history apply appropriate corrections to the raw data, postprocess-
of digital image processing dates back several decades when ing is intended to change the image contrast, reduce image
NASA was working on its lunar and planetary exploration pro- noise and enhance the sharpness of the image displayed in
grams. For example, the organization used computers to process an effort to enhance diagnostic interpretation. This paper
images beamed back from the Ranger spacecraft to improve the will focus on image postprocessing operations and
visualization of details of the moon’s surface. Later, the benefits techniques.

Figure 1. Major Components of a Digital Imaging System. Image processing consists of preprocessing, in which corrections are made to the raw data, and postpro-
cessing, in which displayed images can be manipulated for the purpose of enhancing diagnostic interpretation capability.

E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41 25
 In radiology, digital image acquisition systems include the important to our understanding of digital image processing. A
following: computed tomography (CT), computed radiography digital image is made up of a 2-dimensional array of numbers
(CR), digital radiography (DR), digital mammography (DM), called a matrix. The matrix consists of columns (M) and rows
and digital fluoroscopy (DF) for both routine gastrointestinal (N) that define small square regions called picture elements or
fluoroscopy and vascular imaging; and magnetic resonance pixels, as illustrated in Figure 2. The dimensions of the image
imaging (MRI), nuclear medicine, and medical sonography. can be described by M, N, and the relationship M  N  k
These technologies are described in detail by Brennan et al bits give the size of the image. The spatial location 10, 16 will
[5]. Furthermore, CT and MRI scanners are now integral imag- describe a pixel that is located 10 pixels to the right of the
ing equipment in radiation oncology [6,7]. All of these digital left-hand side (L) of the image and 16 lines down from the
imaging systems use digital image postprocessing operations top of the image. The pixel value in this digital image is 144
to alter the displayed digital image, with the goal of enhancing (Figure 2).
diagnosis. In the relationship M  N  k bits, the term k bits implies
that every pixel in the digital image matrix M  N is repre-
Specific Image Postprocessing Operations sented by k binary digits. The number of bits per pixel is the
Common image processing operations in CT include those bit depth. Because the binary number system uses the base 2,
for image reformatting, windowing, region of interest (ROI), k bits ¼ 2k; therefore, each pixel will have 2k gray levels. For
magnification, shaded surface display, volume rendering, pro- example, in a digital image with a bit depth of 2, each pixel
file, histogram, and collage [7]. will have 22 (4) gray levels (density). Similarly, a bit depth of
In CR and DR, image postprocessing includes gray scale pro- 8 implies that each pixel will have 28 (256) gray levels or shades
cessing (windowing), spatial filtering, and dynamic range control. of gray. The effect of matrix size and bit depth on the spatial
In digital subtracting angiography or digital fluoroscopy, resolution and density resolution, respectively, are clearly seen
common image postprocessing operations include analytic pro- in Figure 3. In addition, the pixel size can be calculated using
cessing, subtraction of images out of a sequence, gray scale pro- the algebraic expression: pixel size ¼ field of view (FOV)/
cessing, temporal frame averaging, edge enhancement, and pixel matrix size.
shifting [8]. Typical matrix sizes and bit depths for nuclear medicine,
In digital mammography, image postprocessing techniques MRI, CT, DSA, CR, DR and digital mammography are 128
include manual intensity windowing, histogram-based intensity  128  12; 256  256  12; 512  512  12; 1024 
windowing, mixture model intensity windowing, contrast-lim- 1024  10; 2048  2048  12; 2048  2048  12; and
ited adaptive histogram equalization, unsharp masking, periph- 4096  4096  12, respectively.
eral equalization, and text processing [9]. These are described in Images can also be acquired in the spatial frequency domain,
detail by Pisano et al [10]. such as those acquired in MRI. The term frequency refers to the
In radiation oncology, image postprocessing operations in- number of cycles per unit length: that is, the number of times
clude all those mentioned for CT, as well as segmentation, a signal changes per unit length. Whereas small structures
which includes classification, edge detection, boundary tracking,
geodesic active contours, extraction of tubular objects, atlas reg-
istration, and interactive segmentation [6]. All concepts men-
tioned above will be briefly explained in this reading.

The Image Domain Concept

To appreciate image postprocessing operations, it is essential
to have a firm grasp of what is meant by the term ‘‘image do-
main’’. Images can be represented in two domains, based on
how they are acquired [11]. These domains include the spatial
location domain, and the spatial frequency domain. All images
displayed for viewing are in the spatial location domain. Radi-
ography and CT, for example, acquire images in the spatial lo-
cation domain. MRI acquires images in the spatial frequency
domain; and, in this case, these images must be converted
into the spatial location domain for viewing by technologists
and the radiologist. Medical physicists, however, appreciate
the spatial frequency domain images and can use them to deter-
Figure 2. Digital Image Format. The format of a digital image includes columns
mine and assess physical characteristics of the imaging system.
(M) and rows (N) that define small, square regions called picture elements, or
A spatial location domain digital image is a numeric image pixels. The field-of-view (FOV) is one dimension of the matrix and is used to
described as a matrix of pixels. Apart from being a numeric calculate the size of the pixel. A right-handed coordinate system is used to describe
image, there are other elements of a digital image that are digital images in the spatial location domain. See text for further explanation.

26 E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41
Figure 3. Effects of Matrix Sizes and Bit Depths. Visual effect of different matrix sizes and bit depths on spatial and contrast resolution of an image, respectively.
(Reprinted with permission from Bruno Jaggi, PEng–Biomedical Engineering, British Columbia Institute of Technology.)

within an object (patient) produce high frequencies that repre- rigorous and will not be covered in this article. The FT converts
sent detail in the image, large structures produce low frequencies a function in the time domain (say, signal intensity versus time)
that represent contrast information in the image. to a function in frequency domain (say, signal intensity versus
Digital image processing can transform one image domain frequency). The inverse FT denoted by FT1 is used to trans-
into another image domain. For example, an image in the spa- form an image in the frequency domain back to the spatial
tial location domain can be transformed into a spatial frequency location domain (Figure 4) for viewing by radiologists and tech-
domain image, as illustrated in Figure 4. The Fourier transform nologists. Physicists and engineers, on the other hand, would
(FT) is used to perform this task. The FT is mathematically probably prefer to view images in the frequency domain.

Figure 4. Fourier Transform. The Fourier transform (FT) is used to convert an image in the spatial location domain into an image in the spatial frequency domain for
processing by a computer. The inverse FT (FT1) is used to convert the spatial frequency domain image back into a spatial location image for viewing by human
observers.

E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41 27
 Spatial frequencies can be used to alter the appearance of im- are used to isolate and highlight the objects of interest. Then the
ages by using high- or low-spatial frequency filters to enhance or features of the objects are extracted resulting in object outlines
suppress certain features of the image. For example, an image or other object measures. These measures describe and charac-
can be sharpened or blurred by using high- and low-pass filters, terize the objects in the image. Finally, the object measures
respectively [11]. This will be described later in this paper. are used to classify the objects into specific categories [2].’’ Seg-
mentation operations are used in 3D medical imaging [6,7].
3. Image synthesis: These processing operations ‘‘create im-
Classes of Image Postprocessing ages from other images or nonimage data. These operations are
There are several operations used in digital image processing used when a desired image is either physically impossible or im-
to transform an input image into an output image to suit the practical to acquire, or does not exist in a physical form at all
needs of the observer. Baxes [2] and Gonzalez [3] identify at [3].’’ Examples of these operations are image reconstruction
least five fundamental classes of operations (Figure 5), including techniques, which are the basis for the production of CT and
image enhancement, image restoration, image analysis, image MR images, as well as 3D visualization techniques [6], which
compression and image synthesis. Although it is not within are based on computer graphics technology.
the scope of this paper to describe all of these in any great detail, 4. Image enhancement: The purpose of this class of process-
it is noteworthy to mention the purpose of each class and state ing is to generate an image that is more pleasing to the observer.
their particular operations. As image enhancement and image Certain characteristics such as contours and shapes can be en-
compression are most commonly used by technologists and hanced to improve the overall quality of the image. The opera-
radiologists, these will be described in more detail later in this tions include contrast enhancement, edge enhancement, spatial
paper. For a more complete and thorough description of these and frequency filtering, image combining, and noise reduction.
classes, readers are referred to the work of Baxes [2] and 5. Image compression: The purpose of image compression
Gonzalez [3]. of digital images is to reduce the size of the image to decrease
1. Image restoration: The purpose of image restoration is to transmission time and to reduce storage space. In general, there
improve the quality of images that have distortions or degrada- are two forms of image compression, lossy, and lossless com-
tions. Image restoration is commonplace in spacecraft imagery. pression. In lossless compression there is no loss of any informa-
Images sent to Earth from various camera systems on spacecrafts tion in the image (detail is not compromised) when the image is
are subject to distortions or degradations that must be corrected decompressed. In lossy compression, there is some loss of image
for proper viewing. Blurred images, for example, can be filtered details when the image is decompressed. The latter has specific
to make them sharper. uses, especially in situations when it is not necessary to have
2. Image analysis: This class of digital image processing exact details of the original image.
allows measurements and statistics to be performed as well as This paper will address only details of image enhancement
image segmentation, feature extraction, and classification of and image compression, as these are the most popular image
objects. Baxes states that ‘‘the process of analyzing objects in postprocessing operations used in medical imaging technologies.
an image begins with image segmentation operations, such as
image enhancement or restoration operations. These operations
Image Postprocessing Operations: A Conceptual Overview
Both past and current-day image postprocessing technology
includes a wide range of image processing algorithms for use
in digital radiology [3,7–11]. These include point processing
operations such as gray scale processing (windowing, image
subtraction, and temporal averaging); local processing operations
(such as spatial filtering, edge enhancement, and smoothing);
and global operations such as the Fourier transform (FT). It is
not within the scope of this article to describe the details of these
processing algorithms; however, a conceptual overview of point
and local operations for single images (as opposed to multiple
images) will be presented, since a number of them are used in
digital radiology.
Image processing operations are intended to change the in-
tensity values of the pixels in the input image, and to display
the resulting changes in the output image with the goal of
changing the characteristics of the image to suit the needs of
Figure 5. Classes of Image Processing Operations. The 5 fundamental classes
the observer to enhance diagnosis. This paper will outline the
of image processing operations according to image processing experts Gonzalez elements of those operations that are specifically intended to
[3] and Baxes [2]. See text for further descriptions of each of these. change and optimize image contrast, improve image detail by

28 E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41
sharpening the image, and decrease the noise present in the
image.

Point Processing Operations

These operations are simple, and are most frequently used in
digital diagnostic imaging. The conceptual framework for this
operation is illustrated in Figure 6. The value of one (point) in-
put image pixel is mapped onto the corresponding output image
pixel; that is, the output image pixel value at the same location
as on the input image matrix depends on the value of the input
image pixel. The operation (algorithm) allows the entire input
image matrix to be scanned pixel-by-pixel, using a ‘‘pixel point
process [2,3]’’ until the entire image is transformed.
One common point processing operation is referred to as
Figure 7. The Histogram. Graph of the number of pixels in the entire image
gray-level mapping or gray scale processing. Other terms that
having the same gray levels (density values), plotted as a function of the gray
are used to describe gray-level mapping are contrast stretching, levels, is referred to as a histogram. See text for further explanation. Ó 2004,
contrast enhancement, histogram modification, histogram stretch- the American Society of Radiologic Technologists. All rights reserved. Reprinted
ing, or simply windowing. Windowing is the most commonly with permission of the ASRT.
used image processing operation in digital diagnostic imaging,
including computed radiography, digital radiography using The Look-Up Table
flat-panel detectors, CT, and MRI.
To illustrate the concept of a LUT, two examples will be
Image contrast and brightness transformations require an
highlighted: a numeric example and a graphic example. Sprawls
understanding of two basic concepts, the look-up-table (LUT)
[12] provides us with two excellent examples, and they will be
and the windowing. However, before these can be understood,
described here to illustrate the concept of a LUT. First consider
the concept of a histogram must be described.
Figure 8, which shows a low-contrast numeric image with a con-
trast difference (that is, the object relative to the background) of
The Histogram
10 (40-30), where 40 represents the background contrast and
The histogram is an essential tool in image postprocessing, 30 represents the object contrast. ‘‘The LUT is then used to
because it can be used to alter image brightness and contrast change the low-contrast numeric image to a high-contrast image
dramatically. An image histogram is a graph of the number of by assigning numbers to the input values 40 and 30 that will
pixels in the entire image, or part of the image having the subsequently change them into 90 and 10, respectively. The
same gray levels (density values) plotted as a function of the contrast difference for the new output image (on the right) is
gray levels, as shown in Figure 7. If the histogram is modified, now 80 (90-10), and therefore this image appears as a high con-
the brightness and contrast of the image will change as well. trast image. During digital image processing, the LUT deter-
This operation is called histogram modification or histogram mines the numbers assigned to the input pixel values that
stretching. If the values of the histogram are concentrated in change them into output pixel values that result in a change
the lower end of the range of values, the image appears dark. in contrast and brightness of the image [1].’’
For a bright image, the values are weighted toward the higher On the other hand, consider Figure 9A, which shows a plot
end of the range of values. The effect of this will be shown later of the input image pixel values as a function of the output image
in the article using Photoshop for a photographic image rather pixel values. In this case the values are the same. ‘‘For example,
than a medical image, since the effect is more dramatic. the value 1024 for the input image matches the value 1024 for

Figure 6. Conceptual Framework for a Point Processing Operation. As can

be seen, one (point) input image pixel is mapped onto the corresponding output Figure 8. The Look-Up-TabledNumeric. Numerical example of the concept
image pixel. The output image pixel value is located at the same location as on of a look-up table (LUT); see text for further explanation. (Reprinted with per-
the input image matrix depending on the value of the input image pixel. mission from Perry Sprawls, PhD, Emory University, Atlanta, GA.)

E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41 29
Figure 9. The Look-Up Table–Graphic. Graphic example of the look-up table (LUT) concept. See text for further explanation. (Reprinted with permission from Perry
Sprawls, PhD, Emory University, Atlanta, GA.)

the output image. The resulting graph is a straight line. Are point to note here is the following: since digital radiographic de-
LUT curves possible? The answer is yes, as is clearly apparent tectors have wide exposure latitude and a linear response, the
in Figure 9B. This is the classic characteristic curve (H and D image displayed without processing appears as a low-contrast
curve) for film. Finally, in Figure 9C, 3 LUT curves are shown: image. A processing example for a chest image using the LUT
a latitude curve, a high contrast curve and an invert curve. Recall, is shown in Figure 10. The low-contrast image is seen on the
that the slope of the characteristic curve influences the contrast left side of the figure, whereas the high-contrast image is illus-
of the image. A steep slope results in a high contrast image while trated on the right side.
a less steep slope (<45 ) will result in decreased contrast [1].’’
Both CR and DR systems make use of a wide range of LUTs
The Concept of Windowing
stored in the system for the different types of clinical examina-
tions (eg, chest, spine, pelvis, and extremities). This means that Windowing is one of the most popular image postprocessing
the operator should select the appropriate LUT to match the operations used by both technologists and radiologists alike to
part being imaged when using these systems. An important change the contrast and brightness of an image [7]. A digital image

30 E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41
 (1000 þ 1000), the WL is 0. In addition, the displayed image con-
trast range (bar graph) is also shown.
While the WW controls image contrast, the WL controls the
brightness of the image. In an image, the displayed gray levels
will range from –1/2WWþWL to þ1/2WWþWL. The dis-
played WW and WL values are always shown on the image.
While narrow WW provides improved image contrast, a wide
WW will show an image with less contrast. This effect is shown
in Figure 12. On the other hand, when the WL is increased, the
image becomes darker since more of the lower numbers will be
displayed as is shown in Figure 13.
It is interesting to note that image subtraction and temporal
averaging are also classified as point processing operations. These
are used routinely in digital subtraction angiography. Essentially,
in image subtraction the pixel values from post-contrast images
are subtracted from pixel values from the first pre-contrast image
Figure 10. The Look-Up Table in Postprocessing. Example of use of the (mask) to show contrast-filled blood vessels, with the other struc-
look-up table (LUT) in postprocessing of a chest image. (Reprinted with permis- tures removed to enhance the diagnostic impressions of the
sion from Perry Sprawls, PhD, Emory University, Atlanta, GA.) radiologist. Temporal averaging, on the other hand, refers to
subtraction of images in time. In particular, temporal averaging
is made up of a range of numbers. By definition, the range of the involves averaging a set of images with the ultimate goal of
numbers in the image is the window width (WW), and the center reducing image noise. The greater the number of images averaged,
of the range is defined as the window level (WL) as illustrated in the less the image noise. Interested readers may refer to Pooley et al
Figure 11. While the range of the pixel values (gray levels) is 2000 [8] for further details of these two techniques.

Figure 11. Window Width and Window Level. Graphic illustration of the
definitions of the image postprocessing concepts of window width (WW) and Figure 12. Effect of Window Width. Visual effect of the window width (WW)
the window level (WL). on the contrast of an image, with window level (WL) held constant.

E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41 31
 Figure 15. Convolution Technique. The image postprocessing operation of
digital filtering in the spatial location domain, known as the convolution
technique.

well as reduce the noise in the image and extract certain features
from the image. Spatial frequency filtering can be done in the
frequency domain, or it can be done in the spatial location do-
main. Whereas the former uses the FT, the latter makes use of
the pixel values themselves.
Spatial Location Filtering: Convolution
A common example of filtering in the spatial location domain
Figure 13. Effect of Window Level. Visual effect of the window level (WL) on is convolution (Figure 15). The algorithm for convolution is such
the brightness of an image, with window width (WW) held constant. that ‘‘the value of the output pixel depends on a group of pixels in
the input image that surround the input pixel of interest: in this
Local Processing Operations case P5. The new value for P5 in the output image is calculated
by obtaining its weighted average and that of its surrounding
A local processing operation is illustrated in Figure 14. As pixels. The average is computed using a group of pixels called
can be seen, it is an operation in which the output image pixel a convolution kernel, in which each pixel in the kernel is a weighting
value is obtained from a small area of pixels around the corre- factor, or a convolution coefficient. In general, the size of the kernel
sponding input image pixel. A notable example is that of spatial is a 3  3 matrix. Depending on the type of processing, different
frequency filtering, to be described later. types of convolution kernels can be used, in which case the weight-
An image in the spatial location domain can be transformed ing factor is different [7].’’ In the act of processing, the kernel
into an image in the spatial frequency domain. The spatial fre- scans across the entire image, pixel by pixel. Every pixel in the
quency domain image would be made of high spatial frequen- input image, the pixels surrounding it, and the kernel are used
cies that represent detail, and low spatial frequencies, which to calculate the corresponding output pixel value. It can be seen
represent contrast information in the image. The idea of using from Figure 15 that each calculation requires 9 multiplications
spatial frequency filtering (processing) is to use the high and low and 9 summations. This arithmetic can take time, so special hard-
frequencies to change the characteristics of the image to suit the ware (array processors) are used to speed up these calculations.
needs of the observer to enhance diagnosis. For example spatial
frequency processing can sharpen, smooth, or blur images, as Spatial Frequency Filtering: High-Pass Filtering
The high-pass filtering process, also known as edge enhance-
ment or sharpness, is intended to sharpen an input image in the
spatial domain that appears blurred. The algorithm is such that
first the spatial location image is converted into spatial frequen-
cies using the FT, followed by the use of a high-pass filter that
suppresses the low spatial frequencies to produce a sharper out-
put image. This process is shown in Figure 16 for a CT image.
The high-pass filter kernel is also shown.
Spatial Frequency Filtering: Low-Pass Filtering
Figure 14. Local Processing. A local processing operation is one in which the
output image pixel value is obtained from a small area of pixels around the cor- A low-pass filtering process makes use of a low-pass filter to
responding input pixel. See text for further explanation. operate on the input image with the goal of smoothing. The

32 E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41
Figure 16. High-Pass Digital Image Processing Filter. The effect of a high-pass digital image processing filter on an input image. The output image is much sharper
than the input image. This filter suppresses the low spatial frequencies in the image which contains the contrast information.

output image will appear blurred. Smoothing is intended to Although Photoshop’s most common applications are in com-
reduce noise and the displayed brightness levels of pixels; mercial and consumer level graphics design and photography
however, image detail is compromised. This is illustrated in editing [4], Photoshop has also been a popular tool among
Figure 17. The low-pass filter kernel is also shown. radiologists, technologists, and imaging scientists. With the cur-
rently released version of Photoshop CS3 Extended, Adobe has
Spatial Frequency Processing: Unsharp Masking raised the bar even higher by taking on the world of medical
The digital image processing technique of unsharp (ie, imaging with the addition of two new features: Support for
blurred) masking uses the blurred image produced from the DICOM (Digital Imaging and Communications in Medicine)
low-pass filtering process and subtracts it from the original im- files, and Analysis Tools.
age to produce a sharp image, as illustrated in Figure 18. It can
be seen that the output image appears sharper. Photoshop and Medical Imaging
Geometric Operations The predominant obstacle medical imaging personnel have
Another class of image processing operations that are some- faced with regards to the use of Photoshop was the lack of
times used in digital radiology is that of geometric operations. support for the most common medical imaging file format,
These techniques allow the user to change the position or orien- DICOM. As the standard format for receiving medical images,
tation of pixels in the image (Figure 19) rather than the bright- DICOM files most often contain a series of images generated
ness of the pixels. Geometric operations result in the scaling, from MRI, ultrasound, or CT scans. Photoshop CS3 Extended
sizing, rotation, and translation of images, once again, to en- now includes new features that apply directly to medical
hance diagnosis. imaging.
When working with DICOM files in Photoshop, the tech-
nologist would be presented with a dialog box allowing for a use-
Photoshop: A Commercial Photography Image Processing
ful choice of options:
Tool with Applications in Medical Imaging
 Frame Import: With this option, DICOM frames can be
What is Photoshop?
imported onto Photoshop layers, and the choice of an
Photoshop is a digital image processing software suite from ‘‘n-up’’ configuration display is available to display multiple
Adobe Systems Inc, and is widely considered by the graphic de- frames in a user-definable grid format.
sign community to be the industry standard for digital image  DICOM Dataset: The ‘‘Anonymize’’ option allows one to
editing and photography enhancement. Typically used by anonymize patient data, and the ‘‘Show Overlays’’ option
both expert and novice users, Photoshop offers powerful digital allows overlay displays such as image annotations or text.
image editing tools such as image filters, histogram manipula-  Windowing: The user can make adjustments in the contrast
tion, sharpening/blurring tools, and noise reduction algorithms. and brightness of the frame, either by entering custom

Figure 17. Low-Pass Digital Image Processing Filter. The effect of a low-pass digital imaging processing filter. The output image is blurred (smoothing) compared to
the input image.

E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41 33
 Figure 18. Unsharp Masking. The image postprocessing technique of unsharp masking is essentially used to sharpen images. See text for further explanation.

values or by choosing built-in, common radiology presets Photoshop CS3 Extended offers the medical imaging world
such as default, lung, bone, or abdomen. a whole new set of analysis and measurement tools for extracting
quantitative data on 2-dimensional images. One such tool can
A snapshot of Photoshop’s DICOM File Info Dialog is shown be used to define the scale of an image using a known value
below. Once a DICOM file is open, any Photoshop tool is then as a reference point. Once a scale has been defined, the ruler
available to edit, manipulate, or annotate the file. For example, tool (in Photoshop) can be used to measure the pixel length
the Pencil tool could be used to circle and highlight an area on the of an area of interest in the image, upon which a logical value
image; the Notes tool could be used to add a comment to the can be assigned. For example, if the pixel length of a tumour
file and sharpening can be performed; and even the Dust and (in the image) is 50 pixels, a logical value of ‘‘5’’ scale units
Scratches filter can be used to remove small imperfections in can be assigned. With this type of scale defined, you can now
the image. Images can then be resized and saved to any supported use the Ruler and Selection tools to measure distances and even
file format based on their intended output medium. areas, recording these measurements into a log, which can also

34 E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41
Figure 19. Geometric Processing Effect. Effect of geometric processing operation on an input image. The goal of the processing is to change the orientation of the input
image.

contain data from multiple image files. These logs can be cus- industry), and gray scale colour spaces. Technologists and radiol-
tomized, have their data sorted into columns, and can then be ogists would benefit from working on images in the Lab mode
exported to a text file. There is even a Scale Marker tool that colour space, as colour data could be either hidden or discarded,
can automatically add a graphic to indicate the scale of the im- whereas values of lightness (otherwise known as luminance) could
age. Also included in the new Analysis Tools is a Count tool be worked on to adjust the histogram data for the image.
that can be used to number off features in an image simply One of the most powerful tools in the Photoshop arsenal is
by clicking on them. These data can then be sent to the mea- the Curves tool. The Curves tool allows the user to adjust the
surements log for further analysis and reference. tonal ranges of an image by clicking on the image Levels curve,
Another interesting and useful set of tools are the Auto-Align and dragging on it to adjust. As changes are made to the curve,
Layers tool and HDR (High Dynamic Range) tools. These tools they are reflected in the histogram values shown in the histo-
allow a user to import multiple images of the same subject with gram palette. By using the Curves tool, one can increase/
varying exposure values (even highly over/under exposed im- decrease the contrast, brightness, and tonal range of an image.
ages). These images will be placed on separate ‘‘layers’’ where A few examples of this tool in action are illustrated below:
they will be automatically aligned and merged into a High Dy- Although this example was done on a regular photograph,
namic Range image, offering ideal exposure values throughout the principles and applications on medical images are the same.
the entire image.
With Photoshop CS3 Extended, medical imaging personnel
can now work directly in Photoshop on DICOM files with
powerful editing tools at their disposal. Technologists and radi-
ologists alike will now be able to manipulate and optimize
images for output image display, resulting in better communi-
cation not only with other healthcare personnel, but with
patients as well.

Examples of Processing Operations

One of the most powerful functions in Photoshop is the ability
to manipulate the image histogram. Photoshop CS3 Extended
now has a histogram palette that allows you to make changes to
the histogram of your image ‘‘on the fly’’ while viewing the under-
lying original image histogram (shown as a lighter shade of gray).
There are several reasons to alter an image, affecting the original
histogram, as the histogram data can often reveal at a glance
whether your image is over- or underexposed, flat (that is, charac-
terized by little contrast), and in what tonal range your image
needs adjusting. For digital photographers, the histograms of
the red, green, and blue (RGB) light channels can be viewed
and altered independently. This allows for exposure and contrast
compensation, and even for colour correction to an image with
a colourcast. In relationship to altering the histogram, Photoshop
allows the user to change the colour space of an image from
RGB, to Lab colour, to CMYK (predominantly used in the print

E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41 35
 For examples of some other useful tools applicable to the
field of medical imaging, see images of the Eiffel Tower using
a high-pass filter and blurring tools.
These are just a handful of the tools available in Photoshop,
but they should give an idea of the potential for using Photo-
shop in processing and viewing medical images. Adobe may still
have a long way to go before Photoshop becomes a standard in
the medical imaging arena; but with Photoshop CS3 Extended,
the potential exists, and the possibilities are turning into reality.

Photoshop in the Classroom

While Photoshop’s main applications are in photographic
image processing, it can also be used in the optimization of im-
ages for instruction, presentations, and printed learning mate-
rials. Applications for teachers, technologists, and even
students are presented next.
As we know, DICOM files may contain several images rep-
resenting slices of an imaging examination. An instructor could
use the Animate tool in the creation of an animation of these
slices, for a truly interactive and enjoyable presentation of
a ‘‘scan in action’’. The use of image slices that are placed on
‘‘layers’’ in Photoshop could be used to tell a story, as different
layers of a scan could be revealed or presented separately.
In some cases, the goal would be to provide a clearer picture
of what is being shown in a medical image (either for instruc-
These tools allow for enhancement of particular areas of interest tion or diagnosis). An author, for example, could easily create
in an image, leading to a more clear representation in highlight- a side-by-side composite image of an original image versus the
ing these areas. Photoshop gives technologists a high level of enhanced (processed) one for comparison.
control over adjusting contrast and tonality in medical images, There is also the ability to crop, resize, and zoom in on areas
which could ultimately lead to a more accurate diagnosis. of an image, which is especially helpful in publications, and

High-pass filter, used for sharpening images, and described previously in this article.

36 E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41
 Blurring tools, such as those for surface blur, Gaussian blur, radial blur, and lens blur.

even presentations. Images can be re-sampled and optimized for [8,11]. For example, the number of images generated in a multi-
web viewing (typically smaller files), or enhanced to higher res- slice CT examination can range from 40 to 3000. If the image
olutions to meet the requirements of printed materials and text- size is 512  512  12, then one examination can generate 20
books. The output of these images can be saved in several file MB or more of data. A CR examination consisting of two images
formats; for instance, a DICOM file could be opened in Photo- per examination with an image size of 2048  2048  12 will
shop, resized or zoomed, then saved as a JPEG file for emailing result in 16 MB of data. A digital mammography examination
to a colleague. The file could even be saved as a high-resolution can now generate 160 MB of data [11]. In addition, Huang
TIFF file to be used in a publication. points out that ‘‘the number of digital medical images captured
With the processing power and tools that Adobe offers in per year in the United States (US) alone is over pentabytes that
Photoshop, it’s easy to see why they have become the standard is, 1015, and is increasing rapidly every year [11].’’ In this regard,
in the graphics industry. The new toolsets offered in Photoshop one can safely assume that similar trends apply to Canada.
CS3 Extended have positioned Adobe for basic image analysis One of the major ways of dealing with this challenge is the
in medical imaging, with room for growth and considerable use of image compression. The goal of image compression is
potential to become a major tool for technologists of the future. to solve the above problems by reducing image data storage
space and increasing the speed of image transmission for large
amounts of digital data, thus decreasing transmission time
Digital Image Compression: An Overview [9,11,13–16].
Earlier in this paper, five fundamental classes of digital image
Compression Ratio: A Definition
processing algorithms were identified (Figure 5). In review,
these include: image enhancement, image restoration, image The impact of image compression on the visual appearance
analysis, image compression, and image synthesis. This section of images depends on the compression ratio used in the process.
will highlight elements of image compression that should be The literature offers several definitions of image compression ra-
in the knowledge skill-set of all digital imaging personnel. tios [9,11,14–16]; however, one that stands out in terms of clar-
Digital acquisition modalities, including multislice CT ity is offered by Huang, who states that ‘‘the compression ratio
(MSCT) and digital mammography, can produce large volumes between the original image and the compressed image file is the
of digital images that are subsequently sent to a picture archiv- ratio between computer storage required to save the original
ing and communication systems (PACS). PACS and Teleradiol- image and that of the compressed data. Thus a 4:1 compression
ogy systems have created challenges with respect to storing and on a 512  512  8 ¼ 2,097,152-bit image, requires only
transmitting large volumes of digital images. These technologies 524,288-bit storage, 25% of the original image storage re-
have resulted in an exponential increase in digital image files quired. [11]’’ This is clearly illustrated in Figure 20.

E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41 37
Figure 20. The Compression Ratio. Graphic illustration of compression ratio. If the original image size is 512  512  8 (2,097,152-bit image), a 4:1 compression
ratio will reduce the size to 524,288-bit image, which requires less storage space.

The above definition will serve as the basis for understanding to 100:1, thus reducing the size of that 5MB image to a mere
what is meant by the term image compression as used in the 50kB [5].’’
remainder of this paper. Lossless compression techniques generally examine patterns
that exist in a string of bits and subsequently present the pattern
Image Compression Basics: A Technical Overview in a more shortened form, thus coding the pixel changes rather
It is not within the scope of this directed reading to describe than the pixel values that make up the image. One such coding
image compression in any great detail; however, the more fun- technique is run length encoding (RLE). In RLE, any item that is
damental concepts (excluding all mathematics) will be outlined repeated in the image is replaced by one value, in conjunction
to provide a reasonable understanding of the technical elements with the number of times that the item is repeated.
involved. A second method of lossless compression, known as
Huffman encoding (HE), can operate on text data as well as
Types of Image Compression on images. This method uses the data to be compressed by first
In general, there are two forms of image compression: lossy constructing a table of the relative frequency of the elements of
(irreversible) and lossless (reversible) compression. In lossless the data. The table is then used to compress the data, with
compression, there is no loss of any information in the image shorter codes assigned to those elements that occur more
(ie, detail is not compromised) when the image is decompressed frequently.
(Figure 21). By contrast, in lossy compression, there is some loss A third lossless compression method for medical images is
of image details when the image is decompressed (Figure 22). called redundancy removal (or background removal), which re-
The latter has specific uses, especially in situations when it is not moves nonuseful data such as the background pixels in the dig-
necessary to have exact details of the original image. The effect ital image. thereby decreasing the size of the image file. (For
of both compression methods will be illustrated subsequently. further description of lossless compression techniques, readers
Lossless (reversible) compression is intended to reduce the may refer to a paper by Seeram [17].)
size of the original image to speed up image transmission and
Irreversible Compression
reduce data storage space, the framework of which is based on
two fundamental steps as described in Figure 23. One short- Lossy, or irreversible, compression methods can compress
coming of lossless compression is its limited compression ratio images at much higher compression ratios, resulting in faster
to a maximum of approximately 5:1 [5]. This means that if image transmission speeds and less image storage space require-
the original size of a digital image is 5 MB, the compressed im- ments compared with lossless compression schemes.
age will be 1 MB. As noted by Brennan et al, ‘‘this may seem Irreversible compression techniques involve three fundamen-
like a substantial reduction until one considers that irreversible tal steps, as illustrated in Figure 24. These include ‘‘transforma-
compression techniques can produce compression ratios of up tion, where the image is transformed from gray scale values to

Figure 21. Lossless (Reversible) Image Compression. Lossless, or reversible, image compression reduces the size of the original image. The most conspicuous difference
compared with irreversible compression is that there is no loss of image information content. See text for further explanation.

38 E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41
Figure 22. Lossy (Irreversible) Image Compression. Lossy, or irreversible, image compression reduces the size of the original image. The most conspicuous difference
when compared with reversible compression is that there is loss of image information content. This type of compression reduces transmission time and occupies less
storage space. See text for further explanation.

coefficients; quantization, where there is a loss of data integrity; big images into large tiles for ease of processing. As the wavelet
and encoding, where reduced coefficients are presented in a com- transform takes place, it generates progressively lower resolution
pact format [14].’’ It is not within the scope of this paper to versions of the wave . in addition to compressing the image, it
describe these in any detail; however, interested readers may has all the information necessary to reconstruct the image in
refer to papers by Seeram [17] and Koff and Shulman [14] for finer detail. The result is a much better image quality than
further descriptions. that achieved with a JPEG file [14].’’
JPEG 2000 is the new wavelet compression standard, and it
Techniques for Irreversible Image Compression is now used in digital medical imaging. JPEG 2000 uses the best
wavelet methods to provide improved quality, especially at high
There are two types of techniques used for irreversible com- compression ratios, among other advantages. The image quality
pression: block compression and full-frame compression. Block advantage is clearly illustrated in Figure 25. Ringl et al [15] have
compression uses a process to compress the image in blocks, reported a comparison of the compression ratios for lossy JPEG
whereas full-frame compression operates on the entire image and JPEG 2000. For example, they report that for CR/DR
rather than blocks of the image. For a further description of chest, CT, MRI, and mammography, the JPEG compression ra-
these methods, readers may refer to the textbook by Huang tios for acceptable compression are 20:1, 10:1, 10:1, and 20:1
[11]. respectively. On the other hand, for the same image types listed
There are two popular block compression techniques. The above, the JPEG 2000 acceptable compression ratios are 50:1,
first is the JPEG (an acronym denoting the Joint Photographic 10:1, 10:1, and 20-25:1 respectively.
Experts Group) standard and wavelet compression. The JPEG For a more complete and thorough description of these irre-
compression algorithm can be either lossless or lossy, and it versible compression methods, interested readers may refer to
was developed for both gray scale and colour images. When the works of ratios [2,3,11,14–17]. In addition, readers will
used on diagnostic images, lossy JPEG compression produces also find updates on JPEG 2000 by visiting the website
prominent artifacts at high compression ratios and particularly http://www.jpeg.org/JPEG2000.htm.
at block boundaries [14,17], and also leads to ‘‘objectionable
‘ringing’ around sharp edges, especially text [14].’’
Effect of Irreversible Compression on Visual Image Quality
These problems can be overcome by the wavelet compression
technique, also known as the discrete wavelet transform, whereby At low compression ratios (8:1 or less), the loss of image qual-
the image is compressed as waves. As noted by Koff and Shul- ity is such that the image is still ‘‘visually acceptable [11].’’ The
man, ‘‘it involves the same steps as does JPEG: transformation, obvious concern that now comes to mind is related to what
quantization, and encoding . the transform organizes the im- Erickson refers to as ‘‘compression tolerance,’’ a term that he de-
age information into a continuous wave, typically with many fines as ‘‘the maximum compression in which the decompressed
peaks and dips, and centers it on zero . and may break up image is acceptable for interpretation and aesthetics [18].’’

Figure 23. Steps in Lossless Image Compression. The lossless compression framework consists of 2 steps: image transformation and encoding.

E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41 39
Figure 24. Steps in Lossy Image Compression. The lossy compression framework consists of three steps: namely, image transformation, quantization, and encoding.
See text for further explanation.

Given that lossy compression methods provide high to very be legal consequences facing radiologists. Therefore, CAR ‘‘has
high compression ratios compared with lossless methods, and accepted the principle of irreversible (lossy) compression for
keeping the term ‘‘compression tolerance’’ in mind, Huang use in primary diagnosis and clinical review, using DICOM
points out that ‘‘currently lossy algorithms are not used by radi- JPEG or JPEG 2000 compression algorithms, at specific com-
ologists in primary diagnosis, because physicians and radiolo- pression ratios set by image type [14].’’ At present, CAR does
gists are concerned with the legal consequences of an incorrect not support the use of irreversible compression for digital
diagnosis based on a lossy compressed image [11].’’ mammography.
A recent survey by Seeram of the opinions of expert radiol-
ogists in the US and Canada on the use of irreversible compres-
sion in clinical practice showed that the opinions are wide and Image Postprocessing: An Essential Tool for Digital
varied. This indicates that there is no consensus of opinion on Imaging
the use of irreversible compression in primary diagnosis. Opin- Digital image postprocessing is now a routine activity in dig-
ions are generally positive on the notion of image storage and ital medical imaging and it is also an essential tool in the PACS
image transmission advantages of image compression. Finally, environment [19]. Technologists and radiologists are already ac-
almost all radiologists are concerned with the litigation potential tively involved in using the tools of image processing, such as
of an incorrect diagnosis based on irreversible compressed im- the digital image processing operations and techniques outlined
ages. For a comprehensive review of the literature on irreversible in this article. Training programs for both technologists and ra-
image compression, interested readers may refer to a paper by diologists are also beginning to incorporate digital image pro-
Seeram [16]. cessing as part of their curriculum. Such activities will only
serve to improve communications with radiologists, medical im-
CAR and the Use of Irreversible Compression
aging physicists, and biomedical engineers, and with equipment
The use of high compression ratios while maintaining image vendors as well.
quality in clinical imaging is possible. However, it appears that A course on image postprocessing for medical imaging tech-
because lossy (irreversible) compression does not preserve all nology programs may include topics such as the nature of digital
the information contained in the original image, there may images, image processing operations and their applications in

Figure 25. Compression Algorithms and Visual Clarity. The effect of JPEG and JPEG 2000 compression algorithms on picture clarity. JPEG 2000 is now used rou-
tinely in digital radiology. (Images courtesy of David Seeram.)

40 E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41
digital radiology, including specific postprocessing operations [3] Gonzalez, R. C., & Woods, R. E. (2008). Digital image processing. Harlow:
in CR, DR, CT, MRI, digital subtraction angiography, digital Prentice-Hall.
[4] Seeram, D. Photography bulletin board. 2007. Available at: www.
fluoroscopy, 3D imaging, and image compression fundamentals photographybb.com.
[20]. Finally, programs may use Photoshop for laboratory ex- [5] Brennan, P., McEntee, M., Seeram, E., Stowe. Digital diagnostic imaging.
ercises that students can engage in to strengthen and enhance Oxford: Blackwell Publishers, in press.
their understanding of image postprocessing. [6] Schlegel, T., Bortfeld, T., & Grosu, A. (Eds.). (2006). New technologies in
radiation oncology. Heidelberg: Springer Verlag.
[7] Seeram, E. (2001). Computed tomography: physical principles, clinical
About the Authors applications, and quality control. Philadelphia: WB Saunders.
[8] Pooley, R. A., McKinney, J. M., & Miller, D. A. (2001). Digital fluoros-
Euclid Seeram is program head and teaching faculty in the copy. Radiographics 21, 521–534.
Medical Imaging Degree program at the British Columbia [9] Seeram, E. (2005). Digital mammography: an overview. Can J Med Radiat
Institute of Technology. He has developed and currently teaches Technol 36:15–23.
several digital imaging courses, including a course in image [10] Pisano, E. D., & Yaffe, M. J. (2004). Digital mammography. Philadelphia
postprocessing in digital radiology, all of which are included Lippincott Williams & Wilkins.
[11] Huang, H. K. (2004). PACS and imaging informatics. New York: John
as either required or elective courses for the medical imaging Wiley & Sons.
degree. [12] Sprawls, P. (2007). [Personal communications.]. Atlanta: Emory University.
David Seeram is an expert on image postprocessing in digital [13] Seeram, E. (2006). Irreversible compression in digital radiology: a literature
photography. He studied applied mathematics at Simon Fraser review. Radiography 12, 45–59.
University and received formal training in a variety of digital [14] Koff, D. A., & Shulman, H. (2006). An overview of digital compression in
medical images: can we use lossy image compression in radiology? Can
photography software programs. He has extensive knowledge Assoc Radiol J 57, 211–217.
of digital photography and image postprocessing; his arsenal of [15] Ringl, H., Schernthaner, R. E., & Kulinna-Cosentini, C., et al. (2007).
tools includes Adobe Photoshop CS3, Lightroom, and Photo- Lossy three-dimensional JPEG2000 compression of abdominal CT im-
matrix. His tutorials can be found throughout the World Wide ages: assessment of the visually lossless threshold and effect of compression
Web and on his own web site, Photography Bulletin Board. ratio on image quality. Radiology 245, 467–474.
[16] Seeram, E. Using irreversible compression in digital radiology: a prelimi-
Dave is a regular columnist for PBASE, an online photography nary study of the opinions of radiologists. Progress in Biomedical Optics
magazine based in Maryland. He is also the editor and publisher and ImagingdProceedings of SPIE. San Diego, CA: July 2006.
of Photography BB.com, an online digital photography maga- [17] Seeram, E. (2005). Digital image compression. Radiol Technol 76(6),
zine. You can visit and chat with Dave at photographybb.com. 449–459; quiz 460–462.
[18] Erickson, B. J. (2002). Irreversible compression of medical images. J Digit
Imaging 15, 5–14.
References [19] Seeram, E. Digital radiography: an introduction for technologists. Clifton
Park, NY: Delmar Cengage Learning, in press.
[1] Seeram, E. (2004). Digital image processing. Radiol Technol 75, 435–455. [20] Seeram, E. (2007). MIMG 7014: image postprocessing in digital radiol-
[2] Baxes, G. A. (1994). Digital image processing: principles and applications. ogy. Medical Imaging Degree Elective Course. Burnaby, BC: British
New York: John Wiley & Sons. Columbia Institute of Technology.

E. Seeram and D. Seeram/Journal of Medical Imaging and Radiation Sciences 39 (2008) 23-41 41