OPTIMUM DCT COMPRESSION OF MEDICAL IMAGES

USING NEURAL NETWORKS

Adnan Khashman and Kamil Dimililer

Intelligent Systems Research Group (ISRG), Near East University, Lefkosa, Northern Cyprus, Turkey

Keywords: X-ray medical images, Optimum image compression, Neural Network, DCT Image Compression.

Abstract: Medical imaging requires storage of large quantities of digitized data Efficient storage and transmission of

medical images in telemedicine is of utmost importance however,. Due to the constrained bandwidth and

storage capacity, a medical image must be compressed before transmission or storage. An ideal image

compression system must yield high quality compressed images with high compression ratio; this can be

achieved using DCT-based image compression, however the contents of the image affects the choice of an

optimum compression ratio. In this paper, a neural network is trained to relate the x-ray image contents to

their optimum compression ratio. Once trained, the optimum DCT compression ratio of the x-ray image can

be chosen upon presenting the image to the network. Experimental results suggest that out proposed system,

can be efficiently used to compress x-rays while maintaining high image quality.

1 INTRODUCTION

X-rays or radiographs are images produced on a

radiosensitive surface, such as a photographic film,

by radiation other than visible light, especially by x-

rays passed through an object or by photographing a

fluoroscopic. These images, commonly referred to

as x-rays, are usually used in medical diagnosis,

particularly to investigate bones, dental structures,

and foreign objects within the body. X-rays are the

second most commonly used medical tests, after

laboratory tests.

Recently, teleradiology, which is one of the most

used clinical aspects of telemedicine, has received

much attention. Teleradiology is the transmission of

radiologic images from a site of image acquisition to

a remote location for interpretation in hospitals such

as computerized tomography (CT) scans, magnetic

imaging (MRI), ultrasonography (US), and x-rays.

These radiological images are needed to be

compressed before transmission to a distant location

or due to the bandwidth or storage limitations (Singh

et al., 2007).

There has been a rapid development in

compression methods to compress large data files

such as images where data compression in various

applications has become more vital (Nadenau et al.,

2003). Efficient methods of compression, to

compress and store or transfer image data files while

retaining high image quality and marginal reduction

in size are needed due to the improvements of

technology (Ratakonda and Ahuja, 2002).

The discrete cosine transform (DCT) is possibly

the most popular transform used in compression of

images in standards like Joint Photographic Experts

Group (JPEG). In DCT-based compression the

image is split into smaller blocks for computational

simplicity. The blocks are classified on the basis of

information content to maximize compression ratio

without sacrificing diagnostic information (Singh et

al., 2007). DCT-based medical image compression

has been investigated by several researchers. For

example, in (Chikouche et al., 2008) DCT-based

compression was applied to IRM type medical

images. In (Prudhvi Raj and Venkateswarlu, 2007)

a medical image compression application based on

3-dimensional DCT was proposed. In (Zukoski et

al., 2006) region based medical image compression

has been applied to choose the clinically relevant

regions as defined by radiologists. In (Shih and Wu,

2005) another region of interest based medical

image compression based on genetic algorithms was

also investigated.

The use of DCT and artificial neural networks

has also been investigated in search for optimum

compression methods. In (Dokur, 2008) MR and CT

medical images were compressed using DCT and

neural networks. In (Ashraf and Akbar, 2006)

Khashman A. and Dimililer K. (2009).

OPTIMUM DCT COMPRESSION OF MEDICAL IMAGES USING NEURAL NETWORKS .

In Proceedings of the 11th International Conference on Enterprise Information Systems - Artiﬁcial Intelligence and Decision Support Systems, pages

91-96

DOI: 10.5220/0001865600910096

 SciTePress

another application of neural networks in medical

image compression was also proposed. In (Meyer-

Base et al., 2005) topology-preserving neural

networks were applied for medical image

compression by a ‘‘neural-gas’’ network. In (Liying

and Khashayar, 2005) different image compression

techniques were combined with neural network

classifier for various applications. In (Soliman and

Omari, 2006) a neural network model called direct

classification was also suggested to compress image

data. In (Ciernak, 2004) periodic vector quantization

algorithm based image compression was suggested

and was based on competitive neural networks

quantizer and neural networks predictor.

More works using neural networks for image

compression applications emerged lately, such as

those in (Ashraf and Akbar, 2005), (Northan and

Dony, 2006), (Vilovic, 2006), and (Veisi and

Jamzad, 2007). Recently, a neural network based

DCT compression system that finds the optimum

compression ratios for a variety of images was also

suggested (Khashman and Dimililer, 2007), where

the evaluation method of the neural network-

obtained optimum compression results was based on

the comparison criteria; which was suggested in

(Khashman and Dimililer, 2005).

The aim of the work presented within this paper

is to develop a medical image compression system

using Discrete Cosine Transform and a neural

network. Our proposed method suggests that a

trained neural network can learn the non-linear

relationship between the intensity (pixel values) of

an x-ray image and its optimum compression ratio.

Once the highest compression ratio is obtained,

while maintaining good image quality, the result

reduction in x-ray image size, should make the

storage and transmission of x-rays more efficient,

thus providing compressed images with good quality

and satisfactory information for the medics.

The paper is organized as follows: Section 2

describes the x-ray image database which is used for

the implementation of our proposed system. Section

3 presents the x-rays compression system;

describing image pre-processing and the neural

network design and implementation. Section 4

introduces the method used to evaluate the results

and provides an analysis of the system

implementation. Finally, Section 5 concludes the

work that is presented within this paper and suggests

further work.

Original Image 10%

20 % 30%

40 % 50%

60 % 70%

80 % 90%

Figure 1: An original x-ray image and its DCT

compression at nine ratios.

2 X-RAY IMAGE DATABASE

The development and implementation of the

proposed medical x-rays compression system uses

60 x-ray images from our medical image database

which were obtained from the Radiology

Department at the Famagusta General Hospital

(Famagusta, Cyprus), which contains x-ray images

OPTIMUM DCT COMPRESSION OF MEDICAL IMAGES USING NEURAL NETWORKS

of fractured, dislocated, broken, and healthy bones

in different parts of the body. DCT compression has

been applied to 50 radiographs using nine

compression ratios (10%, 20%, ..., 90%) as shown in

an example in Figure 1.

The optimum DCT compression ratios for the 50

x-ray images were determined using the optimum

compression criteria based on visual inspection of

the compressed images as suggested in (Khashman

and Dimililer, 2005), thus providing 50 images with

known optimum compression ratios and the

remaining 10 images with unknown optimum

compression ratios. The image database is then

organized into three sets:

• Training Set: contains 25 images with known

optimum compression ratios which are used for

training the neural network within the radiograph

compression system. Examples of training images

are shown in Figure 2a.

• Testing Set 1: contains 25 images with known

optimum compression ratios which are used to

test and validate the efficiency of the trained neural

network. Examples of these testing images are

shown in Figure 2b.

• Testing Set 2: contains 10 images with unknown

optimum compression ratios which are used to

further test the trained neural network. Examples of

these testing images are shown in Figure 2c.

Examples of original x-ray images and their

compressed versions using their optimum

compression ratios while training the neural network

are shown in Figure 3.

3 X-RAY IMAGE COMPRESSION

SYSTEM

The optimum x-ray image compression system uses

a supervised neural network based on the back

propagation learning algorithm, due to its

implementation simplicity, and the availability of

sufficient “input/target” database for training this

supervised learner. The neural network relates the x-

ray image intensity (pixel values) to the image

optimum compression ratio having been trained

using images with predetermined optimum

compression ratios. The ratios vary according to the

variations in pixel values within the images. Once

trained, the neural network would choose the

optimum compression ratio of an x-ray image upon

presenting it to the neural network by using its

intensity values.

Figure 2: (a) Training Set examples (b) Testing Set 1

examples, (c) Testing Set 2 examples.

X-ray 1 40% Compression

X-ray 2 10% Compression

X-ray 3 30% Compression

X-ray 4 20% Compression

Figure 3: Examples of Training Set images and their

optimum compression ratios.

ICEIS 2009 - International Conference on Enterprise Information Systems

Adobe Photoshop was used to resize the original

images of size (256x256) pixels into (64x64) pixels.

Further reduction to the size of the images was

attempted in order to reduce the number of input

layer neurons and consequently the training time,

however, meaningful neural network training could

not be achieved thus, the use of whole images of the

reduced size of 64x64 pixels.

The size of the input x-ray images affects the

choice of the number of neurons in the neural

network’s input layer, which has three layers; input,

hidden and output layers. Using one-pixel-per-

neuron approach, the neural network’s input layer

has 4096 neurons, its hidden layer has 50 neurons,

which assures meaningful training while keeping the

time cost to a minimum, and its output layer has nine

neurons according to the number of the considered

compression ratios (10% - 90%).

During the learning phase, the learning

coefficient and the momentum rate were adjusted

during various experiments in order to achieve the

required minimum error value of 0.003; which was

considered as sufficient for this application. Figure 4

shows the topology of this neural network, within

the x-ray image compression system.

4 RESULTS AND DISCUSSIONS

The evaluation of the training and testing results was

performed using two measurements: the recognition

rate and the accuracy rate. The recognition rate is

defined as follows: where RR

OHC

is the recognition

rate for the neural network within the radiograph

compression system, I

OHC

is the number of optimally

compressed x-ray images, and I

is the total number

of x-ray images in the database set.

100∗

⎟

⎠

⎞

⎜

⎝

⎛

ODC

(1)

The accuracy rate RA

OHC

for the neural network

output results is defined as follows:

(

)

100*

⎟

⎠

⎞

⎜

⎝

⎛

∗−

−=

ODC

(2)

where S

represents the pre-determined (expected)

optimum compression ratio in percentage, S

represents the optimum compression ratio as

determined by the trained neural network in

percentage and S

represents the total number of

compression ratios.

The Optimum Compression Deviation (OCD) is

another term that is used in our evaluation. OCD is

the difference between the pre-determined or

expected optimum compression ratio S

and the

optimum compression ratio S

as determined by the

trained neural network, and is defined as follows:

(

)

10∗−=

SSOCD

(3)

The OCD is used to indicate the accuracy of the

system, and depending on its value the recognition

rates vary. Table 1 shows the three considered

values of OCD and their corresponding accuracy

rates and recognition rates. The evaluation of the

system implementation results uses (OCD = 1) as it

provides a minimum accuracy rate of 89% which is

considered sufficient for this application.

The neural network learnt and converged after

2960 iterations or epochs, and within 774 seconds,

whereas the running time for the generalized neural

networks after training and using one forward pass

Original Image

(256x256) pixels

Input Hidden Output

Layer Layer Layer

4096

Reduced Size

Image (64x64)

pixels

Figure 4: X-ray Image Compression System.

Compressed Image

OPTIMUM DCT COMPRESSION OF MEDICAL IMAGES USING NEURAL NETWORKS

was 0.015 seconds. These results were obtained

using a 2.0 GHz PC with 2 GB of RAM, Windows

XP OS and Matlab 2008b software. Table 2 lists the

final parameters of the successfully trained neural

network, whereas Figure 5 shows the error

minimization curve of the neural network during

learning.

Table 1: Accuracy and recognition rates for OCD.

OCD

Accuracy Rate

(RA

)

Recognition Rate

(RR

)

0 100 % 15/25 (60 %)

1 89 % 24/25 (96 %)

2 78 % 25/25 (100 %)

Table 2: Neural network final training parameters.

Input nodes 4096

Hidden nodes 50

Output nodes 9

Learning rate

0.003

Momentum rate

0.4

Error

0.003

Iterations 2960

Training time (seconds) 774

Run time (seconds)

0.015

Figure 5: Neural network learning curve.

The trained neural network recognized correctly

the optimum compression ratios for all 25 training

images as would be expected, thus yielding 100%

recognition of the training set. Testing the trained

neural network using the 25 images from Test Set 1

that were not presented to the network before

yielded 96% recognition rate, where 24 out of the 25

images with known optimum compression ratios

were assigned the correct ratio.

The trained neural network was also

implemented using the remaining 10 images with

unknown optimum compression ratios from the

testing set. The results of this application are

demonstrated Figure 6 which shows examples of the

optimally compressed x-ray images as determined

by the trained neural network.

X-ray 1 10% Compression

X-ray21 20% Compression

X-ray 3 30% Compression

Figure 6: Examples of Testing Set 2 image compression

using the trained neural network.

5 CONCLUSIONS

A novel method to medical x-ray image compression

using a neural network is proposed in this paper. The

method uses DCT-based compression with nine

compression ratios and a supervised neural network

that learns to associate the grey x-ray image

intensity (pixel values) with a single optimum

compression ratio.

The implementation of the proposed method uses

DCT image compression where the quality of the

compressed images degrades at higher compression

ratios due to the nature of the lossy compression.

The aim of an optimum ratio is to combine high

compression ratio with good quality compressed x-

ray images, thus making the storage and

transmission of images more efficient.

ICEIS 2009 - International Conference on Enterprise Information Systems

The proposed system was developed and

implemented using 60 x-ray images of fractured,

dislocated, broken, and healthy bones in different

parts of the body. The neural network within the x-

ray image compression system learnt to associate the

25 training images with their predetermined

optimum compression ratios within 774 seconds.

Once trained, the neural network could recognize the

optimum compression ratio of an x-ray image within

0.015 seconds

In this work, a minimum accuracy level of 89%

was considered as acceptable. Using this accuracy

level, the neural network yielded 96% correct

recognition rate of optimum compression ratios. The

successful implementation of our proposed method

using neural networks was shown throughout the

high recognition rates and the minimal time costs

when running the trained neural network.

Future work will include the implementation of

this method using wavelet transform compression

and comparing its performance with DCT-based x-

ray image compression using larger database.

REFERENCES

Ashraf, R., Akbar, M., 2005. Absolutely lossless

compression of medical images. In EMBS 2005,

Proceedings of IEEE 27th Conference on Engineering

in Medicine and Biology, 4006-4009.

Ashraf, R., Akbar, M., 2006. Adaptive Architecture

Neural Nets for Medical Image Compression. In

ICCIMA 2007, IEEE International Conference on

Engineering of Intelligent Systems, 1-4.

Ciernak, R., 2004. Image Compression algorithm based on

neural networks. In Lecture Notes in Artificial

Intelligence. Vol. 3070, Springer-Verlag, Berlin

Heidelberg, 706-711.

Chikouche, D., Benzid, R., Bentoumi M., 2008.

Application of the DCT and Arithmetic Coding to

Medical Image Compression. In ICTTA 2008, 3

International Conference on Information and

Communication Technologies: From Theory to

Applications, 1-5.

Dokur, Z., 2008. A unified framework for image

compression and segmentation by using an

incremental neural network. In An International

Journal on Expert Systems with Applications, 34, 611-

619.

Khashman, A., Dimililer, K., 2005. Comparison Criteria

for Optimum Image Compression, In EUROCON’05,

Proceeding of IEEE International Conference, 935-

938.

Khashman, A., Dimililer, K., 2007. Neural Networks

Arbitration for Optimum DCT Image Compression. In

EUROCON’07, Proceeding of IEEE International

Conference, 151-156.

Liying, M., Khashayar, K., 2005. Adaptive Constructive

Neural Networks using Hermite Polynomials for

Image Compression. In Lecture Notes in Computer

Science, Vol. 3497, Springer-Verlag, Berlin

Heidelberg, 713-722.

Meyer-Base, A., Jancke, K., Wismuller, A., Foo, S.,

Martientz, T., 2005. Medical image compression using

topology-preserving neural networks. In Engineering

Applications of Artificial Intelligence, 18(4), 383-392.

Nadenau, M.J., Reichel, J., Kunt, M., 2003. Wavelet

Based Color Image Compression: Exploiting the

Contrast Sensitivity Function. In IEEE Trans. Image

Processing, 12(1), 58-70.

Northan, B., Dony, R.D., 2006. Image Compression with a

multiresolution neural network. In Canadian Journal

of Electrical and Computer Engineering. 31(1), 49-58.

Raj, N.P., Venkateswarlu, T., 2007. A Novel Approach to

Medical Image Compression using Sequential 3D

DCT. In ICCIMA 2007, International Conference on

Computational Intelligence and Multimedia

Applications, 3, 146-152.

Ratakonda, K., Ahuja, N., 2002. Lossless Image

Compression with Multiscale Segmentation. In IEEE

Trans. Image Processing, 11(11), 1228-1237.

Shih, F.Y., Wu, Y., 2005. Robust watermarking and

compression for medical images based on genetic

algorithms. In An International Journal on

Information Sciences, 175, 200-216.

Singh, S., Kumar, V., Verna, H.K., 2007. Adaptive

Threshold-based block classification in image

compression for teleradiology. In Computers in

Biology and Medicine, 37, 811-819.

Soliman, H.S., Omari, M., 2006. A neural networks

approach to image compression. In Journal of Applied

Soft Computing, 6(3), 258-271.

Veisi, S., Jamzad, M., 2007. Image Compression with

Neural Networks Using Complexity Level of Images.

In ISPA 2007, Proceedings of the 5th IEEE

International Symposium on image and Signal

Processing and Analysis, 282-287.

Vilovic, I., 2006. An Experience in Image Compression

Using Neural Networks. In ELMAR-2006,

Proceedings of IEEE 48th International Symposium

on Multimedia Signal Processing and

Communications, 95-98.

Zukoski, M.J., Bould, T., Iyriboz, T., 2006. A Novel

Approach to Medical Image Compression. In

International Journal on Bioinformatics Research and

Applications, 2(1), 89-103.

OPTIMUM DCT COMPRESSION OF MEDICAL IMAGES USING NEURAL NETWORKS