Improved ROI Algorithm for Compressing Medical Images
Mohamed Nagy Saad
1
and Ahmed Hisham Kandil
2
1
Biomedical Engineering Department, Misr University for Science and Technology, 6
th
of October City, Egypt
2
Systems and Biomedical Engineering Department, Cairo University, Giza, Egypt
Keywords: Medical Image Compression, Huffman Coding, Arithmetic Coding, Discrete Cosine Transform, Wavelet.
Abstract: The digital medical images have become an essential part of the electronic patient record. These images
may be used for screening, diagnosis, treatment and educational purposes. These images have to be stored,
archived, retrieved, and transmitted. Compression techniques are extremely useful when considering large
quantities of these images. In this paper, four compression techniques are applied on three medical image
modalities. The compression techniques are either lossless or lossy techniques. The applied lossless
techniques are Huffman and Arithmetic. The applied lossy techniques are Discrete Cosine Transform (DCT)
and Wavelet. The modalities are Ultrasound (US), Computed Tomography (CT) and Magnetic Resonance
Imaging (MRI). The observed parameters are both the compression ratio (CR) and total compression time
(TCT) (compression time + decompression time). The target is to maximize the CR while preserving
images’ information using the best compression technique. The maximum accepted CR for each image is
chosen by three experts. The last enhancement is done by isolating the region of interest (ROI) in the image
then applying the compression procedure. Applying the ROI technique on the studied cases by the experts
gave promising results.
1 INTRODUCTION
Over the past 30 years, information technology (IT)
has facilitated the development of digital medical
imaging. This development has mainly concerned
US, CT, MRI, angiography, nuclear medicine,
mammography, and computed radiology. All these
modalities are producing ever-increasing quantities
of images. The challenge is the reduction of the
images’ sizes stored in the medical servers with the
decrease of their transmission time without affecting
the information in the images. The risk of losing a
piece of diagnostic information does not sit well
with medical ethics. The evolution of digital
imaging, retrieval systems and Picture Archiving
and Communication Systems (PACS), alongside
compression systems, has resulted in changing
attitudes, and compression is now accepted by
medical experts.
The digital medical images have specific
features. Some of these features are: uniform
background, relatively large homogenous regions,
and higher resolution than general images. The
whole compression process can be described as a
method divided into compression and
decompression processes. At the compression
process, the input file is represented in a more
compact format where the number of bits of the
compressed file is fewer than the number of bits of
the original file. At decompression process, the
original file is reconstructed from the compressed
file either totally-as in lossless techniques; or
approximately-as in lossy techniques (Acharya and
Tsai, 2005).
1.1 Compression Ratio
CR is the ratio of the reduction in number of bits
representing the digital image after compression to
the number of bits representing the original digital
image. CR is the most significant metric of
performance measure of a data compression
algorithm. However, the lossless techniques yield
modest CRs, while the lossy techniques yield higher
CRs. Eq. (1) shows equation used to produce CR
(Nait-Ali and Cavaro-Menard, 2008).
(1)
Where CR is the calculated compression ratio, a is
184
Nagy Saad M. and Hisham Kandil A..
Improved ROI Algorithm for Compressing Medical Images.
DOI: 10.5220/0004325301840189
In Proceedings of the International Conference on Biomedical Electronics and Devices (BIODEVICES-2013), pages 184-189
ISBN: 978-989-8565-34-1
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
the original image size and b is the compressed
image size.
Recent works have studied the use of JPEG 2000
(Joint Photographic Experts Group) compression
standard on a variety of medical images. The
Discrete Wavelet Transform is the basis of the
JPEG2000 image compression standard
(Kesavamurthy et al., 2009). Table 1 shows the
range of acceptable CRs defined after analyzing the
accuracy of the diagnosis based on medical experts
(Nait-Ali and Cavaro-Menard, 2008).
Table 1: Applying JPEG 2000 standard on medical
images.
Image Type
Acceptable compression
ratio
Digital chest radiograph
20:1 (so that lesions can
still be detected)
Mammography 20:1 (detecting lesions)
Lung CT image
10:1 (so that the volume of
nodules can still be
measured)
Ultrasound 12:1
Coronary angiogram
30:1 (after optimizing
JPEG 2000 options)
1.2 Total Compression Time
TCT is the time delay required for compressing and
decompressing the image. TCT is one of the
parameters that measure the performance of the
compression algorithms. Compression algorithm
may be effective at the CR but it needs to be
effective at compression time also. The complex
compression algorithm requires relatively long time
leading to serious problems in interactive
applications. The compression algorithm designer
may be advised to decrease the complexity of the
algorithm so as to decrease the TCT (Acharya and
Tsai, 2005).
2 MATERIALS AND METHODS
The flow chart shown in figure 1 illustrates the
compression process as a whole. The compression
process starts with importing the selected image to
the program. The imported image is compressed.
The compressed file is saved and this is the end of
the compression phase. The decompression phase
starts by the inverse of the compression technique to
give the reconstructed image. Finally, the
reconstructed image is saved.
Figure 1: Flow chart of the whole compression process.
Any compression process is preceded by
preprocessing steps. These preprocessing steps
depend on the different features of the medical
modalities. Some of these features that are studied in
this paper are listed in table 2.
Table 2: Features of studied medical modalities.
Modality DICOM True color Bits/Pixel
US Yes Yes 8
MRI Yes No 16
CT Yes No 12
If
the original image is in the Digital Imaging and
Communications in Medicine (DICOM) format, the
text will be separated and the image will be dealt
with normally, as shown in figure 1. Then, after the
compression and decompression phases are
completed, the text file is combined again with the
reconstructed image to construct the DICOM
reconstructed file.
In case of US, the preprocessing step is the
separation of the red (R) component, the green (G)
component, and the blue (B) component of the true
color image. Then, each component is dealt with
simultaneously as grayscale image. Each component
is compressed then saved then decompressed to form
ImprovedROIAlgorithmforCompressingMedicalImages
185
the reconstructed components as shown in figure 2.
The post-processing step is the concatenation of the
reconstructed components to build up the true color
reconstructed image.
Figure 2: Block diagram for dealing with true color
images.
The choice of wavelet family, which is used in
wavelet decomposition, is a critical issue that affects
image quality. A selected wavelet family must result
in perfect reconstruction. To achieve the best CR,
the best wavelet family and the best decomposition
level must be chosen. The biorthogonal family of
wavelets is used for medical modalities compression
as shown in table 3 (Tolba, 2002).
Table 3: Selected parameters for wavelet coding of
medical modalities.
Parameter Value or Type
Wavelet Family Biorthogonal 3.9
Decomposition Level Four
Retained Energy
(Entropy-Like Criterion)
99.96%
3 RESULTS
This section of the paper is concerned with the
comparative study results of the medical images. A
sample of ten images is tested for each modality and
average results for each technique are compared to
show the best technique. Secondly, five images from
each medical dataset are studied. These sample
datasets are subjected to three experts. Then, the
experts assign the maximum CR that could be
reached without affecting the diagnosis process.
Then, ROI approach shows it’s improvement on CR
of the expert studied cases.
3.1 Comparative Study Results
At this section of the study, the mean of the ten
medical tested images is calculated for CR (%) and
TCT (Sec.) at each technique. Figures 3, 5, and 7
show that DCT technique has the highest CR. Also,
DCT technique has the shortest TCT as shown in
figures 4, 6, and 8.
Figure 3: A comparison of CR results among the four
compression techniques for US images.
Figure 4: A comparison of TCT for each compression
technique related to each other for US images.
Figure 5: A comparison of CR results among the four
compression techniques for CT images.
Figure 6: A comparison of TCT for each compression
technique related to each other for CT images.
Figure 7: A comparison of CR results among the four
compression techniques for MRI images.
43,613
56,025
59,092
60,022
228
6509
2.2
1.7
Arithmetic
Huffman
Wavelet
DCT
61,642
43,966
69,144
73,762
295
14582
1.4
0.36
Arithmetic
Huffman
Wavelet
DCT
35,713
37,895
59,71
77,762
BIODEVICES2013-InternationalConferenceonBiomedicalElectronicsandDevices
186
Figure 8: A comparison of TCT for each compression
technique related to each other for MRI images.
3.2 Experts Studied Cases Results
This section of the paper is concerned with the
results of medical studied cases. These medical
cases are five of each studied modality. The
technique applied on the studied cases is DCT which
is the best technique throughout the four compared
techniques as shown in the previous section.
The compressed images with different CRs are
reviewed by three expert physicians independently.
Then, experts specify the accepted compressed
images and the rejected compressed images. Figures
9, 10, and 11 show the CR results that are selected
by the experts.
Figure 9: US studied cases results.
3.3 ROI Studied Cases Results
The first step in automatic segmentation of the
clinical portion of the medical image (eliminating
the non-essential background) is converting it from a
grayscale image to a binary image by selecting an
intensity threshold as shown in figure 12. The
second step is discarding the regions other than the
ROI by selecting a size threshold as shown in figure
13. The intensity level and the size level used for
thresholding depend on the medical modality and are
selected manually according to the physician
preference.
The third step is filling the holes in the selected
region as shown in figure 14. The fourth step is
clearing the borders of the selected region as shown
in figure 15. The fifth step is outlining the original
image with the selected region as shown in figure
16. This step is performed for verifying that the
selected region is the required ROI.
Figure 10: CT studied cases results.
Figure 11: MRI studied cases results.
Figure 12: Conversion of grayscale image to binary image
based on threshold selection.
Figure 13: Discarding the regions other than the ROI.
The sixth step is finding the extreme points of the
selected region. The extreme points are minimum x-
193
14419
1.1
0.16
Arithmetic
Huffman
Wavelet
DCT
70
75
80
85
90
95
12345
Compression Ratio
(%)
Dr1
Dr2
Dr3
0
50
100
150
12345
Compression
Ratio (%)
Dr1
Dr2
Dr3
70
80
90
100
12345
Compression
Ratio (%)
Dr1
Dr2
Dr3
ImprovedROIAlgorithmforCompressingMedicalImages
187
value, minimum y-value, maximum x-value, and
maximum y-value of the region of interest. The
importance of the extreme points is that they will be
the indices of the final ROI image.
Figure 14: ROI without holes.
Figure 15: Clearing the borders of the selected region.
The last step is cropping the ROI in the original
image as shown in figure 17. The original image is
cropped at the extreme points of the ROI which
specify the width and height of the cropped image.
Figures 18, 19, 20 show the CR results after
applying ROI on the images.
Figure 16: Outlining the original image with the selected
region.
Figure 17: Cropping the ROI in the original image.
Figure 18: US ROI studied cases results.
Figure 19: CT ROI studied cases results.
Figure 20: MRI ROI studied cases results.
4 CONCLUSIONS
The best compression technique is the one who
83,42
87,01
80,48
79,92
81,26
89,07
95,69
96,36
88,13
94,47
12345
CR(before) CR(after)
73,07
94,22
93,78
78,42
81,46
91,92
97,01
95,73
98,42
92,78
12345
CR(before) CR(after)
82,82
93,33
83,52
96,16
90,13
88,17
97,5
90,73
98,52
97,46
12345
CR(before) CR(after)
BIODEVICES2013-InternationalConferenceonBiomedicalElectronicsandDevices
188
reaches a high CR without any deterioration in
image quality. In the considered comparative study,
The DCT technique has proved high efficiency by
reaching maximum CR within minimum TCT. The
experts studied cases results in average CR for US of
82.4%, for CT of 84.2%, and for MRI of 89.2%. The
most valuable point is the isolation of the ROI for
the experts studied cases which results in a
considerable enhancement of the CR results. The
average improvement in CR for US is 10.3%, for CT
is 11%, and for MRI is 5.3%.
A comparison between JPEG-LS, JPEG2000,
and the applied algorithm is shown in table 4. For
CT modality results on JPEG2000, the study was
applied on the lung organ. For CT modality results
on JPEG-LS and the ROI algorithm, the study was
applied on different organs. For MRI modality
results on JPEG2000, the study was applied on the
brain organ. For MRI modality results on JPEG-LS
and the ROI algorithm, the study was applied on
different organs.
Table 4: Comparison between the ROI Algorithm and
Previous Studies.
JPEG-LS JPEG2000 ROI
US
3.4:1
(Clunie, 2000)
12:1 (Nait-Ali
and Cavaro-
Menard, 2008)
13.78:1
CT
4:1
(Clunie, 2000)
10:1 (Nait-Ali
and Cavaro-
Menard, 2008)
20.71:1
MRI
3.6:1
(Clunie, 2000)
20:1 (Terae et
al., 2000)
18.1:1
From table 4, the applied algorithm gives
promising results for US & CT modalities.
JPEG2000 gave the best results for MRI modality.
REFERENCES
Acharya, T. and Tsai, P., 2005. JPEG2000 Standard for
Image Compression Concepts, Algorithms and VLSI
Architectures. New Jersey: Wiley.
Clunie, D., 2000. Lossless Compression of Grayscale
Medical Images - Effectiveness of Traditional and
State of the Art Approaches. Proceedings of SPIE,
Medical Imaging, 3980, 74-84.
Kesavamurthy, T., Rani, S., Malmurugan, N., 2009.
Volumetric Color Medical Image Compression for
PACS Implementation. IJBSCHS, 14(2), 3-10.
Nait-Ali, A. and Cavaro-Menard, C., 2008. Compression
of Biomedical Images and Signals. London, New
Jersey: ISTE and Wiley.
Terae, S., Miyasaka, K., Kudoh, K., Nambu, T., Shimizu,
T., Kaneko, K., Yoshikawa, H., Kishimoto, R.,
Omatsu, T., Fujita, N., 2000. Wavelet Compression on
Detection of Brain Lesions with Magnetic Resonance
Imaging. Journal of Digital Imaging, 13(4), 178-190.
Tolba, S., 2002. Wavelet Packet Compression of Medical
Images. Elsevier Science, Digital Signal Processing,
12, 441–470.
ImprovedROIAlgorithmforCompressingMedicalImages
189