SOURCES OF NOISE IN DIGITAL X-RAYS
There are three major sources of noise in conventional
digital X-rays. The main source of noise is the so-called
secondary radiation which is nothing more than reflected
X-rays which reach the film from all directions. These
unwanted X-rays tend to reduce the contrast quality of the
image. Radiologists have solved the problem by
developing a specially designed grid plate made of lead.
The grid plate absorbs most of the scattered secondary X-
ray and produces an radiograph within the acceptable level
of noise (figure 1). This technique is the most popular
method used in almost all traditional X-ray machines.
Another source of noise is film development and film
handling, although the procedure is fully automated roughly
10 to 15% of the processed X-rays are poorly developed.
Image with poor contrast quality whether over or under-
exposed, is usually labeled as a rejected image. All rejected
images are thrown away and the X-ray must be retaken.
Rejected pictures make up 20% of the total number of X-
rays taken per year (Jassam, 1992). The rejected pictures
cost hospitals millions of dollars per year and expose the
patient to unnecessary radiation.
With advancing computer technology and increasing
public concern with both the radiation level and insurance
costs, the need to transfer from pictorial to digital X-rays is
growing rapidly. This transformation will introduce
additional sources of noise both geometrically and
radiometrically. The unblocked secondary radiation, poor
film developing and handling, and the digitization process
introduces noise into the digital X-ray. The noise to signal
ratio is relatively high in the case of rejected images and
within an acceptable limit otherwise. Image processing
techniques proved to be effective in minimizing the number
of rejected pictures (Jassam, 1992). To maximize the
amount of information extracted from the X-ray images
and to increase their visual quality, the noise has to be
suppressed. Low-pass filters are often used in image
processing to remove random noise from a digital image.
Unfortunately, as noise is removed from an image, details
in the image are also lost. The goal of removing noise in an
image is therefore to strike a balance between noise and
detail which is appropriate for the application at hand.
To determine this balance between noise and detail, the
user applies a variety of low-pass filters, of various sizes, to
the image and then compares the results. This comparison
can either be quantitative; employing statistical analysis
techniques, or qualitative; examining the physical
appearance of the image. Both the statistical and physical
results of applying low-pass filters to an X-ray image in the
spatial and frequency domains, using Intergraph
Corporation's Imager software (TIGRIS 1988), are
addressed.
patient
y secondary
ve X \ X table radiation
Film
Figure 1. secondary radiation reach the film,
are the main source of noise.
TREATMENT OF NOISE IN DIGITAL
X-RAY IMAGES
In the process of generating a digital image, many Sources
of error are encountered. These include errors in acquiring
the data, errors in digitizing the data, and errors in
transmitting the data. As a result of these errors, digital
images often contain individual pixels which vary abruptly
in intensity from their neighboring pixels and which do not
reflect the scene they represent. These pixels are referred to
as random noise (Richards, 1986). To remove noise, low-
pass (smoothing) filters are applied to the image. It is
important to note that noisy pixels are like edges and lines
in that they stand out from their neighbors, and because of
this similarity, removing noise in an image also results in
removing, or at least blurring, edges and lines.
Filters can be applied in the spatial or the frequency
domain. In the spatial domain, the values of the pixels in
the resulting image depend directly upon their original
values and the values of their original neighboring pixels.
In the frequency domain, the resulting pixel values depend
on the horizontal and vertical frequency components in the
original image and not directly on individual pixel values
(Richards, 1986). However, in both domains, low-pass
filters serve to reduce the contrast amongst neighboring
pixels, and since noisy pixels are pixels whose intensities
vary abruptly from the intensities of neighboring pixels,
applying low-pass filters reduces the amount of contrast
and therefore the amount of noise in the image.
Spatial domain
To filter an image in the spatial domain, a kernel is moved
over the rows and the columns of the image and the value
of the pixel located in the center of the kernel is replaced by
the sum of the products of the kernel elements and the
corresponding image pixel values (Richards, 1986).
Mathematically, this can be expressed as follows:
M N
ax = X LO WM) ars 1
m=1n=1
where :
g(x,y)is the image new brightness values,
M is number of rows in the kernel,
N is number of columns in the kernel,
f(m,n) is the pixel brightness value addressed
according to the kernel position, and
t(m,n) is the kernel entry at that location.
The low-pass filter provides a way of removing noise by
changing the pixel value of a noisy pixel to the average
value of its surrounding pixels. However, not all pixels
whose brightness values differ from surrounding pixels
represent noise. Therefore, low-pass filtering results in less
detail in addition to less noise. The size of the kernel
determines the number of pixel values used to calculate the
replacement pixel value. Therefore, larger kernels use
more pixel values in the averaging process and this results
in smoother images.
Other smoothing filters are the median, and the mode .
The median filter functions similarly to the low-pass filter,
however, the center pixel is replaced by the median value
of the pixels covered by the filter. As a result the median
filter is less likely to smooth edges. The mode filter also
functions similarly to the low-pass filter, however, in this
case the center pixel is replaced by the mode of the pixels
covered by the filter.
