Quality of chest X-ray images compressed using JPEG AND JPEG2000
Matko Dodig1, Sonja Grgić2 and Slavica Dodig3
Abstract — The aim of this paper is to compare the quality of digitized chest X-ray images with digitally captured chest X-ray images in children with lung tuberculosis (TB), compressed using popular JPEG and JPEG2000 compression algorithms, for further electronic correspondence. All images were compressed using JPEG and JPEG2000 compression algorithms with six compression degrees (3 bpp, 1.5 bpp, 1 bpp, 0.7 bpp, 0.5 bpp, 0.3 bpp). Picture quality was evaluated by means of usual objective measures of picture quality such as MSE, SNR and PSNR. All compressed images were shown to physicians, who were asked to evaluate diagnostic accuracy of each image (by poor, good or excellent). According to physician's subjective evaluation both JPEG and JPEG2000 compression for all X-ray digitized images (all TB-attributes) were accepted as excellent.
Keywords — Chest radiograph, image analysis, image compression, teleradiology.
I.Introduction
I
maging techniques, such as X-ray imaging, ultrasound, computorizedtomography, magnetic resonance,look "through" the body. They play an important role in medical diagnosis. Some of them (ultrasound) have a signal resolution and noise comparable to standard TV. They can be transmitted without significant loss of quality for primary diagnosis, teleradiology or for video conference presentation. X-ray film has much resolution than television [1]. Compressed imaging can lose its high-resolution content. The fundamental difficulty in testing quality of compressed image is how to decide which test images to use for the evaluations [2]. For teleradiology routine X-ray images must be digitized with very high resolution and sent as purely digital files with no quality loss during transmission [2]. For primary diagnosis X-ray images are transmitted usually as digital camera captured images. Recently, digital radiography has demonstrated significant improvements in image quality. Since digital radiography store-and-forward telemedicine in Croatia is available mostly in clinical hospitals, digitized X-ray images still offer a low-cost alternative for physicians in distant medical centers to obtain second opinion from consultant specialists. The aim of this study was to compare the quality of digitized chest X-ray with digitally captured chest X-ray images in children with lung tuberculosis (TB).
II.Test images
Eight upright posteroanterior chest X-ray images were selected for study: four images were photographed using digital camera (Fig. 1), and four images were digitaly captured. All radiographs were selected by pediatrician and in detail described by radiologist. Photographed X-ray images were placed on 12 mm thin negatoscope (lighted X-ray viewer) and photographed in the same darkened room. The digital camera (Canon Power Shot S45) was mounted on a tripod placed approximately 75 cm from the negatoscope (this distance assures that most chest X-rays filled the camera's LCD monitor; the camera's optical zoom set to the midpoint). Digitaly captured images were obtained using Shimadzu equipment. All images were stored in PC, converted to 24-bit BMP and compressed. To produce test chest X-ray images, two different image compression algorithms were used: JPEG 3] and JPEG2000 4]. JPEG image compression is designed to accommodate limitations of the human eye [5]. Each X-ray image was compressed with six different compression degrees (3 bpp, 1.5 bpp, 1 bpp, 0.7 bpp, 0.5 bpp, 0.3 bpp).
(a) Normal finding /
(b)TB: Infiltration, cavities
(c) TB: Infiltration, cavities, limphadenopathy /
(d) TB: Infiltration
Fig. 1. Chest X-rays of one healthy child (a) and three children with tuberculosis (b - d)
III.Picture quality measures
Among many numerical measures of picture quality, we have chosen next measures: mean square error (MSE), optimized quality factor (OQF), signal-to-noise ratio (SNR) and peak-signal-to-noise ratio (PSNR) [2].
MSE and PSNR are the most common measures of picture quality [6]. All compressed images were shown to physicians, who were asked to evaluate diagnostic accuracy of each image, by poor, good or excellent.
Statistical significance was tested by Wilcoxon test and by Student t-test, while correlation between two variables was expressed by coefficient of correlation (r) [7].
IV.Results
Results of objective quality measures of JPEG and JPEG2000 compressed images of each chest X-ray image are summarized in Tables 1. to 4. PSNR and SNR values in digitaly captured images were lower than PSNR and SNR values in photographed/digitized images in both JPEG and JPEG2000compression, respectively.
Table 1: Quality measures of JPEG compression for each digitized image
X-ray / GIVEN BITRATE: 3.0 bppOptimized Quality Factor / MSE / SNR (dB) / PSNR (dB)
(a) / 99 / 0.2 / 45.5 / 55.5
(b) / 99 / 0.2 / 44.4 / 55.5
(c) / 99 / 0.2 / 46.2 / 55.6
(d) / 99 / 0.2 / 44.6 / 55.6
GIVEN BITRATE: 1.5 bpp
(a) / 97 / 0.2 / 45.9 / 55.9
(b) / 97 / 0.2 / 44.8 / 55.9
(c) / 98 / 0.2 / 46.0 / 55.5
(d) / 97 / 0.2 / 45.0 / 56.0
GIVEN BITRATE: 1.0 bpp
(a) / 93 / 0.3 / 43.9 / 53.9
(b) / 92 / 0.3 / 41.7 / 52.8
(c) / 93 / 0.3 / 44.7 / 54.1
(d) / 93 / 0.3 / 43.1 / 54.1
GIVEN BITRATE: 0.7 bpp
(a) / 85 / 0.67 / 37.8 / 47.8
(b) / 85 / 1.1 / 36.5 / 47.5
(c) / 85 / 1.0 / 38.7 / 48.2
(d) / 85 / 1.1 / 36.8 / 47.8
GIVEN BITRATE: 0.5 bpp
(a) / 81 / 1.3 / 37.0 / 47.1
(b) / 80 / 1.4 / 35.5 / 46.6
(c) / 81 / 1.2 / 38.0 / 47.5
(d) / 82 / 1.2 / 36.2 / 47.2
GIVEN BITRATE: 0.3 bpp
(a) / 68 / 1.9 / 35.3 / 45.3
(b) / 66 / 2.2 / 33.7 / 44.8
(c) / 67 / 1.9 / 36.0 / 45.5
(d) / 68 / 1.9 / 34.4 / 45.4
Data in Table 1. and Table 3., as well as Fig. 2. (JPEG), show that PSNR values are > 50.0 dB in compression degrees 3.0 to 1.0 bpp and < 50.0 dB in compression degrees 0.7 to 0.3 bpp (two tailed probability <0.01). At the same time coefficient of correlation between PSNR and OQF shows close positive correlation (r = 0.9395), while correlation between PSNR and MSE shows negative correlation (r = -0.9453). In compression degrees 3.0 to 1.0 bpp MSE value is significantly lower than in compression degrees 0.7 to 0.3 bpp (two tailed probability <0.01).
Table 2: Quality measures of JPEG compression for each digitally captured image
X-ray / GIVEN BITRATE: 3.0 bppOptimized Quality Factor / MSE / SNR (dB) / PSNR (dB)
(a) / 99 / 1.1 / 36.9 / 47.8
(b) / 98 / 0.2 / 44.1 / 53.9
(c) / 98 / 0.8 / 39.2 / 49.9
(d) / 99 / 0.7 / 38.0 / 49.3
GIVEN BITRATE: 1.5 bpp
(a) / 97 / 3.5 / 32.1 / 42.7
(b) / 97 / 0.5 / 41.4 / 51.2
(c) / 94 / 0.9 / 39.4 / 49.9
(d) / 97 / 2.4 / 32.8 / 44.3
GIVEN BITRATE: 1.0 bpp
(a) / 83 / 4.9 / 30.8 / 41.5
(b) / 93 / 0.8 / 39.4 / 49.2
(c) / 87 / 2.1 / 34.5 / 44.9
(d) / 85 / 3.2 / 31.7 / 43.7
GIVEN BITRATE: 0.7 bpp
(a) / 67 / 6.1 / 30.5 / 41.1
(b) / 57 / 1.1 / 37.6 / 47.4
(c) / 81 / 2.9 / 33.0 / 43.5
(d) / 62 / 4.2 / 30.3 / 41.9
GIVEN BITRATE: 0.5 bpp
(a) / 57 / 8.0 / 28.5 / 39.1
(b) / 81 / 1.2 / 37.4 / 47.2
(c) / 69 / 4.3 / 31.4 / 41.8
(d) / 64 / 5.5 / 29.2 / 40.8
GIVEN BITRATE: 0.3 bpp
(a) / 29 / 10.9 / 29.7 / 39.4
(b) / 31 / 2.2 / 34.9 / 44.7
(c) / 43 / 8.4 / 28.5 / 38.9
(d) / 31 / 9.0 / 27.0 / 38.6
In JPEG2000 (Table 2., Table 4. and Fig. 3.) compression, PSNR values are significantly higher in compression degrees 3.0 bpp to 0.7 bpp than in compression degrees 0.5 to 0.3 bpp (two tailed probability <0.01). Coefficient of correlation between PSNR and decoded bitrate shows close correlation (r = 0.9277).
There is no statistically significant difference between PSNR values in JPEG and PSNRvalues in JPEG2000, p>0.05.
Average values of PSNR for all eight X-rayimages also show that PSNR values in digitally captured images were lower than average values in photographed/digitized images, in both JPEGand JPEG2000compression, p<0.05 (Fig. 2. and Fig.3., respectively).
Fig.2. PSNR (mean values) for all eight X-rayimages according to JPEGcompression. (Photo - photographed images; Digit - digitally captured images).
Table 3:Quality measures of JPEG2000 compression for each digitized image
X-ray / GIVEN BITRATE: 3.0 bppDecoded Bitrate (bpp) / SNR (dB) / PSNR (dB)
(a) / 0.71 / 41.0 / 50.5
(b) / 0.80 / 39.0 / 50.0
(c) / 0.71 / 41.0 / 50.5
(d) / 0.70 / 39.1 / 50.2
GIVEN BITRATE: 1.5 bpp
(a) / 0.70 / 41.0 / 51.5
(b) / 0.79 / 39.0 / 50.0
(c) / 0.70 / 41.0 / 50.5
(d) / 0.70 / 39.1 / 50.2
GIVEN BITRATE: 1.0 bpp
(a) / 0.74 / 40.1 / 50.1
(b) / 0.79 / 39.0 / 50.0
(c) / 0.69 / 41.0 / 50.5
(d) / 0.70 / 39.1 / 50.2
GIVEN BITRATE: 0.7 bpp
(a) / 0.73 / 40.1 / 50.1
(b) / 0.75 / 38.8 / 49.9
(c) / 0.69 / 41.0 / 50.5
(d) / 0.70 / 39.1 / 50.2
GIVEN BITRATE: 0.5 bpp
(a) / 0.50 / 38.7 / 48.7
(b) / 0.50 / 37.2 / 48.2
(c) / 0.50 / 39.8 / 49.2
(d) / 0.50 / 37.9 / 48.9
GIVEN BITRATE: 0.3 bpp
(a) / 0.30 / 37.1 / 47.2
(b) / 0.30 / 35.6 / 46.7
(c) / 0.30 / 38.1 / 47.6
(d) / 0.30 / 36.3 / 47.3
Fig.3. PSNR (mean values) for all eight X-rayimages according to JPEG2000compression. (Photo - photographed images; Digit - digitally captured images).
Table 4: Quality measures of JPEG2000 compression for each digitally captured image
X-ray / GIVEN BITRATE: 3.0 bppDecoded Bitrate (bpp) / SNR (dB) / PSNR (dB)
(a) / 0.75 / 37.6 / 48.3
(b) / 0.82 / 40.7 / 50.5
(c) / 0.69 / 39.8 / 51.4
(d) / 0.79 / 37.1 / 48.6
GIVEN BITRATE: 1.5 bpp
(a) / 0.71 / 33.4 / 44.1
(b) / 0.69 / 40.7 / 50.5
(c) / 0.71 / 40.2 / 50.4
(d) / 0.72 / 35.3 / 46.9
GIVEN BITRATE: 1.0 bpp
(a) / 0.74 / 32.3 / 42.9
(b) / 0.79 / 39.7 / 49.5
(c) / 0.69 / 36.6 / 47.0
(d) / 0.68 / 33.0 / 44.5
GIVEN BITRATE: 0.7 bpp
(a) / 0.73 / 30.9 / 41.5
(b) / 0.75 / 40.7 / 50.5
(c) / 0.66 / 35.1 / 45.6
(d) / 0.65 / 32.0 / 43.5
GIVEN BITRATE: 0.5 bpp
(a) / 0.50 / 29.7 / 40.3
(b) / 0.50 / 38.5 / 48.3
(c) / 0.50 / 33.6 / 44.1
(d) / 0.50 / 31.2 / 41.8
GIVEN BITRATE: 0.3 bpp
(a) / 0.30 / 28.5 / 39.2
(b) / 0.30 / 36.7 / 47.8
(c) / 0.30 / 31.8 / 42.2
(d) / 0.30 / 29.2 / 40.1
There is no statistically significant difference between PSNR values in JPEG and PSNR values in JPEG2000, for both, photographed and digitally captured images, respectively, p>0.05 (Fig.4. and Fig.5.). However, PSNR values in Figures 4 and 5 show that JPEG performed better at lower compression ratios (bitrate >1 bpp) in images that were aquired using digital camera.
Fig.4. PSNR (mean values) for all eight photographed images according to JPEG and JPEG200-compression.
Fig.5. PSNR (mean values) for all eight digitaly captured images according to JPEG and JPEG200-compression
According to physician's subjective evaluation both JPEG and JPEG2000 compression for all X-ray digitized images (all TB-attributes) were accepted as excellent.
V.Conclusion
Tuberculosis as a public health may justify the use of digital cameras for consultations regarding radiography. Chest X-ray images can be transferred on-line in real time (or delayed) between local health personnel and consultant physicians in central health institutions.
The first step is to convert the X-ray image into an electronic signal suitable for digital processing and storage. The transfer process, as well as compression, can generate noise that can influence on the quality of transmitted digitized and postprocessed chest X-ray images.
To produce test chest X-ray images two different image system compression algorithms were used: JPEGand JPEG2000. Our results indicate that both JPEG and JPEG2000 compressed chest X-rayimages from the Canon Power Shot S45 camera allow readings of sufficient (excellent) quality for the diagnosis of tuberculosis (as good as digitally captured images).
Mrak M. et al. consider that PSNR values depend on image content [8]. In present study PSNR values in JPEG and JPEG2000 compressions did not depend on X-ray image content because differences between test images are small. Differences in PSNR values occurred between compression degrees 1.0 bpp and ≤ 0.7 bpp in JPEG as well as between compression degrees > 0.7 bpp and ≤ 0.5 bpp in JPEG2000 (two tailed probability <0.01). PSNR values in JPEG compression were directly proportional to OQF, and inversely proportional to MSE. In JPEG2000, PSNR correlated to decoded bitrate. Since there were no statistically significant difference between PSNRvalues in JPEG and JPEG2000 (p>0.05) both compression algorithms can be used for store-and-forward chest X-ray teleradiology. The National Electrical Manufacturers Association (NEMA) has added the JPEG2000 standard to the Digital Imaging and Communications in Medicine (DICOM) standard for compressing medical images [9]. The DICOM Standards Committee creates international standards for the communication of diagnostic and therapeutic information in medical disciplines that use digital images. According to the results of M. M. Sung et al. JPEG2000 can achieve higher compression ratios with less distortion than JPEG algorithm[10]. Since we tend to standardization in teleradiology we consider that JPEG2000 algorithm should be implemented in Croatia as well.
Acknowledgment
We would like to express our gratitude to Boro Nogalo, MD, Đurđica Milković, PhD, MD, Miljenko Raos, MD, MS, and Drago Ćaleta, MD, from Children's Lung Hospital Srebrnjak, Zagreb, Croatia, for submitted X-ray images and for invaluable advice and participation in this study.
References
[1](Mira - teleradiology and digital radiology) T. Sund (2005, July, 1th,
[2]S. Grgić, M. Grgić, M. Mrak, "Reliability of objective picture quality measures", J. Electrical Engineering, vol 55, pp. 3-10, 2004.
[3]ISO/IEC IS 10918, "Digital compression and coding of continuous tone still images", 1991.
[4]ISO(IRC FDIS 15444-1, "JPEG2000 Part I Final draft international standard", 2000.
[5]A. Szot, F. L. Jacobson, S. Munn, D. Jazayert, E. Nardell, D. Harrison, R. Drosten, L. Ohno-Machado, L. M. Smeaton and H. M. Fraser, "Diagnostic accuracy of chest X-ray acquired using a digital camera for low cost teleradiology", Int. J. Med. Informatics, vol. 73, pp. 65-73, 2004.
[6]S. Grgić, M. Grgić, B. Zovko-Cihlar, "Picture quality measurements in wavelet compresion system", Proceedings of the International Broodcasting Convention, IBC99, Amsterdam, The Netherlands, pp. 554-559, Sep. 1999.
[7]Mathworld, (2005 July, 17th. Available:
[8]M. Mrak, S. Grgić, and M. Grgić, "Picture quality measures in image compression system", EUROCON2003, Ljubljana, Slovenia, pp. 233-237, 2003.
[9]Digital Imaging adn Communication in Medicine (DICOM) PS.3.1-2001. (2005, July. 17th. Available:
[10]M. M. Sung, H. J. Kim, S. K. Yoo, B. W. Choi , J. E. Nam, H. S. Kim, J. H. Lee and H. S. Yoo, "Clinical evaluation of compression ratios using JPEG2000 on computed radiography chest images", J Digit Imaging. vol 15(2), pp. 78-83, 2002.
1Senso IS, Zagreb, Croatia, E-mail: , 2 Faculty of Electrical Engineering and Computing, University of Zagreb, Zagreb, Croatia, 3Children's Lung Hospital Srebnjak, Zagreb, Croatia