LOSSLESS COMPRESSION OF FLUOROSCOPY MEDICAL IMAGES USING CORRELATION

2012 IEEE EMBS International Conference on Biomedical Engineering and Sciences | Langkawi | 17th - 19th December 2012

Lossless Compression of Fluoroscopy Medical Images using Correlation and the Combination of Run-length and Huffman Coding Arif Sameh Arif1, Sarina Mansor1, Hezrul Abdul Karim1

Rajasvaran Logeswaran1,2

1Faculty of Engineering, Multimedia University, 63100 Cyberjaya, Selangor, Malaysia

2School of Science & Technology, Nilai University College, 71800 Bandar Baru Nilai, Negeri Sembilan, Malaysia

Abstract— Medical centers produce a massive amount of sequential medical images for examinations such as CT, MRI and Fluoroscopy, where each examination of a patient consists of a series of images. This takes up a large amount of storage space, in addition to the cost and time incurred during transmission. For medical data, lossless compression is preferred to the greater gains of lossy compression, in the interest of accuracy. This paper proposes a new method for lossless compression of pharynx and esophagus fluoroscopy images, using correlation and combination of Run Length and Huffman coding on the difference pairs of images classified by correlation. From the experimental results obtained, the proposed method achieved improved performance with a compression ratio of 11.41 for the proposed combination of Run-length and Huffman coding (RLHM-D) on the difference images as compared to 1.31 for the standard images.

The rest of the image is then compressed using lossy techniques. Combining both techniques in this way provides higher CR while preserving the quality of the regions of interest [1]. Fluoroscopy is a special type of X-ray that provides continuous X-ray images of a patient’s organ structures in a real time as individual radiographic. It is used in many types of examinations and procedures, such as Percutaneous Nephrostomy (PCN), Barium Swallow, cardiac catheterization, and others [3]. The longstanding popular methods for lossless compression include Huffman and Run-length coding. The combination of two effective compression methods Run Length and Huffman Coding was proposed to reduce the data volume, pattern delivery time, and save power in scan application [4]. In medical images, the combination of Run Length and Huffman Coding was implemented on magnetic resonance images and x-ray angiograms to achieve maximum compression [5]. The same combination of Run Length and Huffman coding was implemented for the color images after quantization and threshoding the DWT coefficients [6]. In our recent work, we have performed lossless compression of Fluoroscopy medical images using correlation and Huffman coding (HM-D) [7]. The proposed method achieved compression ratio of 7.97. In this paper, we extend the work to adopt the combination of Huffman and Run-length coding. Specifically, we propose a new framework for lossless medical image compression based on classifying the images depending on correlation and coding the difference between sequential images using the combination of Run Length and Huffman coding. To the best of our knowledge, there are no other published works employing this technique in the context of medical image compression.

Keywords-component; Fluoroscopy; ROI; Lossless image compression; Run-length Coding; Huffman Coding; Correlation.

I.

INTRODUCTION

Recent years have seen a tremendous increase in the generating, transmission and storage of medical images. Many researchers have been working towards developing approaches for compressing medical images. Compression techniques can be classified into two main categories: lossless and lossy. Typically, lossless image compression is necessary for medical applications, where perfect accuracy is expected in maintaining the same quality of the original image for medical diagnosis. Nevertheless it comes at the expense of obtaining low compression ratios [1]. In lossy image compression, the reconstructed image is not identical to the original image but is usually reasonably close to it. The compression ratios achieved by lossy compression are higher [2]. An alternative approach to improve the compression ratio (CR) and yet maintain the diagnostic quality of the image is to identify the region of interest (ROI). ROI is the diseased region where doctors should diagnose the abnormality, hence lossless compression would be applied.

978-1-4673-1666-8/12/$31.00 ©2012 IEEE

759

2012 IEEE EMBS International Conference on Biomedical Engineering and Sciences | Langkawi | 17th - 19th December 2012

Correlation theory and contourlet transform were employed to generate a new medical image processing algorithm which provides excellent achievements such as multi-scale analysis, multi directions and time frequencylocalization. This algorithm reconstructs a high quality image with a relative few coefficients [12]. An efficient approach for finding correlation between PSNR (Peak Signal to Noise Ratio) and SSIM (Structural Similarity Index Objective Image quality parameters with subjective MOS (Mean Opinion Score)) for SPIHT (Set Partitioning in Hierarchical Trees) based on six distinct observers identified in [13].

This paper is organized as follows: In section II, we briefly describe the region of interest, correlation, the combination of Run Length and Huffman coding and review related works. In section III, we define our new approach. In section IV, we present the experimental results and discussions, followed by the conclusion section V. II.

EXPERIMENTAL WORK

Identifying and extracting accurately the ROI is essential step before coding and compressing the image data for efficient transition or storage. By using different spatial regions and identifying the region of interest of the image, it is possible to compress it into different levels of reconstruction quality. Images can be classified into three regions: (1) Primary region of interest (PROI), (2) Secondary region of interest (SROI) and (3) background [8]. This way one could accurately maintain the features needed and transmit for medical diagnosis or for scientific measurement, while achieving high overall compression by allowing degeneration of data in the unimportant regions [9]. The correlation was determined by the variance and covariance, where the variance is measured for a dimension with itself, while the covariance is measured between two matrices or dimensions. The formula of variance (Var) and co-variance (Cov) are given as below [10].
 

PROPOSED METHOD

The proposed method is motivated by classifying the images depending on the CC, i.e. depending on the homogeneity between different images. This reduces the redundancy between images. The diagram in Fig.1 summarizes the main steps involved in the proposed framework. Fig.2. show a sample of a pharynx and an esophagus fluoroscopy image used in this work, including the labels area to be processed.

Input image

   



    

III.

Reference Image Correlation Test Image

    



Pre-processing Steps

   

Remove the white areas

 are means of X and Y respectively, R is the where  and  correlation between X and Y. To calculate the similarity between the orginal image and the reconstructed image, the correlation coefficient (CC) is calculated as follows:-

Remove the black areas

Subtract the difference

Huffman Coding

   

Run Length Coding

+ *  !  "#$% &  ' &  $ () /0 -

Coded Vector

,+ *#  !  )- + *#$% &  ' &  $ ()- .

where x = 0, 1, 2,…, M-1, y = 0, 1, 2,…, N-1,$ ( is the average value of the pixels in w,  ! is the average value of f (intensity function) in the region coincident with the current location of w, and the summations are taken over the coordinates common to both f and w, where M and N is the size of the original image. The CC is scaled in the range -1 to 1, independent of scale changes in the amplitude of f and w [11].

Figure 1. The framework of the proposed approach

760

2012 IEEE EMBS International Conference on Biomedical Engineering and Sciences | Langkawi | 17th - 19th December 2012

IV.

Black area

RESULTS AND DISCUSSION

We performed experiments on 104 fluoroscopy images, each of size 512×512 pixels and file size of 256KB, to evaluate the validity of the proposed approach. The performance evaluation, in terms of CR depending on the CC is tabulated in Table 1. The CC value is calculated as in (4), while the CR is calculated as follows:

White area

ROI 1 

23 5 24

where So = File size of the original image, Sd = File size of the difference image.

Figure 2. Sample of the Fluoroscopy image approach

Table 1 shows the correlation values between the tested and the reference images for a random sample of ten images. The higher correlation is correlated with higher redundancy that will give higher compression ratio.

The proposed approach can be divided into three main steps:A. Correlation The correlation coefficients (CC) were used to classify a collection of medical images into groups. The CCs were obtained between the first image and the second image for the first step, followed by the CC between the second image and the third image, and so on.

TABLE 1 Comparison of compression ratio performance (CR) between the proposed method the combination of Run-length and Huffman coding (RLHM-D) on the difference images as compared to the implementation of Huffman coding on the difference images (HM-D) and to the combination of Run Length and Huffman coding (RLHM-S) on the standard image for a random sample of ten images. [CC is the correlation between the tested and the reference images].

B. Preprocessing step The preprocessing step can be summarized as follows:B-1 Remove the black areas from the fluoroscopy images. Depending on the aspect of the fluoroscopy images, there are four black corners in each image. By locating the center of the image and calculating the distance from the center to the edge of the image, we can draw a circle that contains the actual data from the fluoroscopy device. B-2 Remove the white area from the fluoroscopy images. This does not contain any relevant data. Steps B1 – B2 result in the extraction of ROI. B-3 Compute the difference between images by subtracting the test image from the reference image, as most images taken from the same view are mostly similar. Therefore, we can use the first image as the base pattern (reference image) and save the difference results as a vector. C. Coding step After preprocessing (step B), a one dimensional vector is obtained which will be used for coding. Depending on the structure of the vector data, we can decide the method of coding. The combination of Run Length and Huffman coding are selected because they are based on the repetition and on the estimated probability of occurrence for each possible value of the source symbol.

761

Compression Ratio (CR)

Reference Image

Tested Image

Correlation (CC)

RLHM-S

HM-D

RLHM-D

212-2

212-3

0.84

1.33

2.11

2.39

212-3

212-4

0.99

1.34

6.12

9.56

212-5

212-6

0.78

1.36

1.55

1.93

954-6

954-7

0.99

1.36

6.44

10.12

954-18

954-19

0.97

1.31

5.88

8.01

330-3

330-4

0.98

1.31

6.17

9.47

330-4

330-5

0.99

1.31

6.3

11.2

330-5

330-6

0.99

1.31

6.31

11.41

330-6

330-7

0.99

1.31

6.42

10.79

330-7

330-8

0.99

1.31

6.33

10.55

2012 IEEE EMBS International Conference on Biomedical Engineering and Sciences | Langkawi | 17th - 19th December 2012

REFERENCES

From Table 1, it is also observed that the proposed method (RLHM-D) achieved significantly better performance than implementing Huffman coding on the difference images (HM-D) and it also performed better than the standard lossless combination (RLHM-S) compression of the images. For example, images 5 and 6, the standard combination method (RLHM-S) only produced CR of 1.31. Huffman coding on the difference images (HM-D) gained a CR of 6.31 but the proposed method (RLHM-D) implemented based on the CC indications, a CR of 11.41 was achieved for the difference image (5-6). Assuming that image 5 was stored in the standard method, the improvement in CR for storing the vital information of image 6 losslessly improved from 1.31 to 11.41, which is significant 800% improvement. To further evaluate the performance gain, the benefits in terms of storage and transmission should be studied. Table 2 lists the total size of a random group sample (ten test fluoroscopy images), against the size of the images compressed using (HM-D) and the proposed method (RLHM-D). It is observed that the total transmission size of the proposed method is only 15.5% of that of the original images, indicating significant transmission time and cost reduction.

[1] O. Abdel Alim, N. Hamdy and W. Gamal El-Din, “Determination of the region of interest in the compression of biomedical images,” IEEE National Radio Science. Cairo, pp. 1-6, March 2007. [2] F. Sepehrband, M. Mortazavi, S. Ghorshi and J. Choupan, “Simple lossless and near-lossless medical image compression based on enhanced DPCM transformation,” IEEE Communications, Computers and Signal Processing. Victoria, BC, pp. 66-72, August 2011. [3] C. Mei-Yen, L. Chien-Tsai, S. Ying-Chou, C. Lee Theun, C. Chian-Fa, M. Huo and C. Cheng-Yen. “Design and evaluation of a DICOM compliant video fluoroscopy imaging system,” IEEE 9th International Conference on Health Networking Application and Services, Taipei, pp. 248-251, June 2007. [4] M. Norani and M. H. Tehranipour, “RL-Huffman encoding for test compression and power reduction in scan applications,” ACM. Trans. USA, vol. 10, Issue 1, 2005, pp. 91-115. [5] R. Shyam Sunder, C. Eswaran, and N. Sriraam, “Performance evaluation of 3-D transforms for medical image compression,” IEEE International Conference, Electro Information Technology, Lincoln, NE, 6 pp. -6, May 2005. [6] V. Setia and V. Kumar, “ Coding of DWT coefficients using Run-length coding and Huffman coding for the purpose of color image compression,” International Journal of Computer and Communication Engineering, vol. 6, 2012, pp. 201-204. [7] A. Sameh Arif, S. Mansor, R. Logeswaran and H. Abdul Karim, “Lossless compression of fluoroscopy medical images using correlation,” IEEE International Conference International Conference on Engineering and Built Environment, November 2012.

TABLE 2 Total sizes of ten fluoroscopy images against the size of the images compressed using the proposed method. No. of images

Original

HM-D

RLHM-D

10

2560KB

749.62KB

399KB

[8] P. Akhtar, M. Iqbal Bhatti, T. Javid Ali, and M. Abdul Muqeet, “Significance of region of interest applied on MRI image,” IEEE 1st International Conference on TeleRadiology-Telemedicine, Bioinformatics and Biomedical Engineering, pp. 1331-1334, July 2007.

Image Size

V.

[9] R. Kumar.M.S., S. Koliwad, and D. G.S., “Lossless compression of digital mammography using fixed block segmentation and pixel grouping,” IEEE 6th Indian Conference on Computer Vision, Graphics & Image Processing, Bhubaneswar, pp. 201-206, December 2008.

CONCLUSION

[10] M. Younus Javed, and M. Habib Khan, “Wavelet based medical image compression through prediction,” IEEE International Multi Topic Conference, Karachi, pp. 155-159, December 2008.

In this paper, a new framework for image compression based on the grouping the images based on the correlation has been proposed. The technique concentrates on the region of interest to code the difference between the groups of images using the combination of Run-length and Huffman coding. The method has achieved significant improvement in compression performance, and indirectly storage and transmission benefits. Our proposed framework is not coding specific, we would like to implement other lossless coding techniques such as Golomb-rice coding in order to further improve the compression performance.

[11] P. Bharti, S. Gupta and R. Bhatia, “Comparative analysis of image compression techniques a case study on medical images,” IEEE International Conference on Advances in Recent Technologies in Communication and Computing, Kottayam, Kerala, pp. 820-822, November 2009. [12] J. Wang and Y. Kang, “Study on medical image processing algorithm based on contourlet transform and correlation theory,” IEEE WRI World Congress, Computer Science and Information Engineering, vol. 6 pp. 233238, April 2009. [13] B. Kumar, S.P. Singh, A. Mohan and H. Vikram Singh, “MOS Prediction of SPIHT medical images using objective quality parameters,” IEEE International Conference on Signal Processing Systems, pp. 219-223, May 2009.

Acknowledgement The authors would like to thank Dr. Noraini Abdul. Rahim (Head Radiology Department), Khatijah Ali (Radiographer) and Mr Ang Kim Liong (Clinical Research Centre) at Serdang Hospital.

762