by Harwood Academic Publishers Printed in Singapore Color Changes in Electronic Endoscopic Images Caused by Image Compression

In recent years, recording of color still images into magneto–optical video disks has been increasingly used as a method for recording electronic endoscopic images. In this case, image compression is often used to reduce the volume and cost of recording media and also to minimize the time required for image recording and playback. With this in mind, we recorded 8 images into a magneto-optical video disk in 4 image compression modes (no compression, weak compression, moderate compression, and strong compression) using the Joint Photographic Image Coding Experts Group (JPEG) system, which is a widely used and representative method for compressing color still images, in order to determine the relationship between the degree of image compression and the color information in electronic endoscopic images. The acquired images were transferred to an image processor using an offline system. A total of 10 regions of interest (ROls) were selected, and red (R), green (G), and blue (B) images were obtained using different compression modes. From histograms generated for these images, mean densities of R, G, and B in each ROI were measured and analyzed. The results revealed that color changes were greater for B, which had the lowest density, than for R or G as the degree of compression was increased.


INTRODUCTION
As electronic endoscopes (Classen and Phillip,  1984) have increased in popularity, an increasing number of medical institutions have begun to employ various new recording media, such as color video printers, video cassette recorders (VCRs), and magneto-optical video disks, to replace conventional photographic film (Katayama   et al., 1995).In practice, image compression is frequently used to increase the efficiency of elec- tronic storage in magneto-optical video disks and other digital media.We previously reported that patterns extracted through image analysis of the same images differed slightly according to the compression ratio when these images were recorded at different compression ratios using the Joint Photographic Image Coding Experts Group (JPEG) system, which is a representative method for color still image compression (Katayama   et al., 1994).In the present study, mean densities of red (R), green (G), and blue (B) were com- pared between images obtained using different compression modes to assess the effects of image compression on color information in electronic endoscopic images.
Each of the 8 images recorded in all 4 image compression modes were transferred to an image processor (TOSPIX-U, Toshiba) using an offline system and converted to monochrome images (Fig. 1).After this, or 2 regions of interest (ROIs) were selected in each image, for a total of 10 ROIs (Fig. 2).Then, R, G, and B images were recorded in the 4 image compression modes and histograms of these images were obtained for each ROI (Fig. 3).Using these histograms, the mean densities of R, G, and B in each ROI were calculated (Fig. 4).

MATERIALS AND METHODS
Eight images were obtained from 6 patients who were randomly selected from among patients undergoing routine endoscopic examination using a scope for the upper gastrointestinal tract with a CCD incorporating approximately 270,000 pixels (TGI-3000D, Toshiba) combined with an elec- tronic endoscope system (TRE-3000, Toshiba) (Yoshie, 1994).These images were recorded in 4 different JPEG image compression modes (no compression, weak compression, moderate com- pression, and strong compression) using a magneto- optical video disk recorder (MV-200, TEAC).
The memory requirements for these 8 images in each compression mode are shown in Table I.

RESULTS
Mean densities of R, G, and B in the 10 ROIs are shown in Table II in the following order: no compression, weak compression, moderate com- pression, and strong compression.The densities (mean + SD) of R, G, and B in the 10 ROIs recorded in weak, moderate, and strong compres- sion modes are shown in Fig. 5. R and B showed the highest and lowest densities, respectively.
Next, the weak compression/no compression, moderate compression/no compression, and strong compression/no compression ratios of the densities in the 10 ROIs were calculated (mean + SD) to compensate for the differences in area between the l0 ROIs (Table III).The density  FIGURE 2 Monochrome images in which a total of 10 ROIs were selected for 8 endoscopic images.
ratios (mean -+-SD) of R, G, and B in R, G, and B images, respectively, of the 10 ROIs recorded in weak, moderate, and strong compression modes relative to those recorded without com- pression are shown in Fig. 6.Differences in density between images recorded without com- pression and images recorded in different com- pression modes were smallest for R, indicating that the effect of image compression on R was minimal.No clear trends were observed between weak, moderate, and strong compression modes for R or G. On the other hand, compared with images recorded without compression, the differ- ence in density increased as the compression ratio increased for B, indicating that image compres- sion had a pronounced effect for B. 1.200 1. 000 0.800 FIGURE 6 Ratios (mean -1-SD) of mean densities of R, G, and B in 10 ROIs recorded in weak, moderate, and strong compression modes relative to those recorded without compression.DISCUSSION employed in the present study, and images recorded without compression.However, there Tremendous advances have been made in image should be some differences between images compression technology.Although image com- acquired using the MV-200 at different mean pression was initially limited to black-and-white compression ratios, such as 1/10 (weak compres- still images, methods for compressing dynamic sion), 1/20 (moderate compression), and 1/30 color images, such as the Moving Pictures Experts (strong compression).In fact, the results of the Group (MPEG) system, have emerged in recent present study indicate that the difference in the years.Image compression systems for the record- density of B, which was lower than that of R or ing of color still images, such as the JPEG system G (Miyahara et al., 1989), relative to that in used in the present study, have been utilized in images without compression, increased as the various fields.Although electronic filing of medi- degree of compression was increased.Thus, cal images has mainly been used in the field of image compression using the JPEG system has radiography, such as CT, MRI, and NM, a rela- only minimal effects on R and G, which have tively large number of institutions are already high mean densities in electronic endoscopic using or plan to use electronic filing for endo- images, but pronounced effects on B, which has scopic images.Because image compression is a relatively low mean density.However, since the frequently employed in such electronic filing to absolute density of B is small, image compression reduce the volume and cost of recording media at 1/30 has only minimal effects on colors in and to minimize the time required for image electronic endoscopic images, and the resulting recording and playback, the effects of compres- color images are thought to be visually indis- sion on image quality must be thoroughly tinguishable from electronic endoscopic images investigated, recorded without compression.In addition, Color information is critical for electronic although Yamada et al. stated that image proces- endoscopic images, unlike radiographs, and it is sing and image analysis of compressed images difficult to visually distinguish between images suffers from no practical limitations because opticompressed at 1/30 to 1/40 using the JPEG sys-mal photographic conditions, such as lighting, tem, which is utilized in the MV-200 and was are not always achieved in endoscopic images (Yamada et al., 1990), we should emphasize that the images selected for image analysis in the pre- sent study had a high signal level because they were obtained from a short distance from a nearly front view (Katayama et al., 1989) using an electronic endoscope equipped with a CCD incorporating 270,000 pixels (Katayama et al.,  1993), and that these images were far superior to those obtained using a conventional endoscope providing only several tens of thousands of effec- tive pixels.

FIGURE
FIGUREEight endoscopic images transferred to a TOSPIX-U.

FIGURE 3 FIGURE 5
FIGURE 3 Uncompressed images of "f" images (esophagogastric junction) shown in Figs. and 2. R, G, and B images (upper row) and their corresponding histograms (lower row) are shown from left to right.

TABLE II -
1 Mean densities of R in R images (according to compression modes)

TABLE II -
2 Mean densities of G in G images (according to compression modes)

TABLE II -
3Mean densities of B in B images (according to compression modes)

TABLE III
Ratios of densities of R, G, and B in R, G, and B images, respectively, recorded in weak (W), moderate (M), and strong (S) compression modes relative to those recorded without compression (N)