Author + information
- ↵*Reprint requests and correspondence: Dr. Hirshfeld, Cardiac Catheterization Lab, Hospital of the University of Pennsylvania, 3400 Spruce Street, Philadelphia, Pennsylvania 19104-4283
The current issues (April issue) of the Journal of the American College of Cardiology (JACC) and the European Heart Journal (Eur Heart J) contain three related manuscripts that report the findings of a major international clinical study of the impact of digital data compression on the diagnostic accuracy of coronary angiography. Because the topic is complex and most practitioners have a limited background in this area, this introduction is written to assist readers in understanding the study’s findings, interpreting the results, and applying the information to their clinical practices and administrative decisions. This introductory article represents the views of the authors and should not be construed as official policy of the American College of Cardiology (ACC) or the European Society of Cardiology (ESC).
These studies were undertaken to provide a scientific basis for identifying standards to be incorporated into the Digital Imaging and Communications Standard for Medicine (DICOM). The pioneers of DICOM proceeded on the premise that eventually digital imaging techniques would completely replace analog recording methods such as film or videotape. They correctly recognized that a successful transition to the digital era would require carefully written standards to ensure that patient data and images could be exchanged without quality degradation.
Beginning in 1992, the cardiology community, led by the ACC, initiated the current process of standardizing the digital file format and image exchange medium for cardiac X-ray angiography. A committee jointly directed by ACC and industry representatives has met monthly in Washington, D.C. for the last six years to develop the current angiographic standard. This effort was broadened to include the full participation of the ESC in 1996 and was joined by the Japanese cardiology societies in 1998. In 1995, a basic cardiac X-ray angiographic application profile, defined as 512 × 512 pixel images with 8 bits (256 levels) of gray scale per pixel, was first demonstrated and included in the DICOM standard. This application profile also designated the recordable compact disk (CD-R) as the approved exchange medium for angiographic images. Within a few years, the DICOM cardiac standard became universally accepted throughout the world as the only method appropriate for exchange of digital angiographic images. More recently, the DICOM Standards Committee has adopted a high-resolution standard, defined as 1,024 × 1,024 pixel images with up to 12 bits (4,096 levels) of gray scale per pixel.
The large quantities of digital data required to represent a cardiac angiographic study present a major challenge. In the standard DICOM file format, each cine frame contains 256 kilobytes of data. At a 30 frame per second (fps) acquisition rate, each second of cine requires 7.5 megabytes (MB). For 1,024 × 1,024 × 10 bit images, each second of imaging can require more than 35 MB. A typical diagnostic study 60 s in length and acquired at the lower resolution of 512 × 512 × 8 generates approximately 450 MB of digital data.
The large size of these digital data files poses a number of operational challenges:
1. Data storage: An active cardiac catheterization laboratory performing 3,000 procedures per year would generate approximately 1.4 terabytes (1,400 gigabytes) of data per year.
2. Reading device performance: Real-time image reading requires data transfer rates of about 8 MB/s. Read speeds of CD-ROM readers, although improving, are currently capable of full speed replay of 512, but not 1,024, matrix image data. If display devices were not required to handle such large data files, the demand on hardware capacity would be reduced, with concomitant reduction in cost.
3. Network transfer issues: Transfer of a complete cardiac angiographic sequence over a hospital network (from the Cardiac Catheterization Laboratory to another physician’s office, for example) would require the transfer of large data files, which would take considerable time over typical shared hospital networks and would compete with other network traffic, thus degrading performance of the entire network.
Files acquired according to the DICOM high-resolution standard will magnify these problems five-fold. Consequently, there is considerable incentive to employ data compression to achieve file-size reduction. Data compression can be either “lossless,” meaning that the compressed file can be reconstructed to reproduce the original file exactly, or “lossy,” which means that the file reconstructed after decompression, although a close approximation, is not an exact replica of the original. Lossy compression can cause visible degradation of the original image. The integrity of the patient data is the paramount concern, which is why the original DICOM standard for angiocardiography specifies that only lossless data compression can be used. This permits file-size reductions of approximately 2:1, with perfect reconstruction of the original file upon decompression.
Because of the potential benefits that would accompany substantial file-size reduction, there has been considerable interest in employing lossy compression to angiographic imaging studies. In general, the degree of compression determines the amount of degradation in such a way that images compressed at 10:1 are altered to a greater extent than images compressed at 5:1. There are lossy compression algorithms that can reduce a file size by as much as 40-fold and still permit reconstruction of a degraded, but reasonably recognizable, image. It would be tempting to employ these techniques in order to surmount the problems outlined above. However, the effect of lossy compression on image quality, and particularly its effect on diagnostic accuracy, had previously not been determined. Specifically, there were no data to indicate whether there might be a level of lossy compression that would yield a major reduction in image data file size while still yielding acceptable image quality.
Consequently, the DICOM Working Group I, together with ACC representatives and the ESC Digital Imaging Committee, designed and carried out the study reported in the three articles that follow. The goal of the project was to determine the impact of various levels of lossy compression on the key clinical attributes of digital cardiac angiographic images. The findings of the study will facilitate formulation of a rational policy for standards development with respect to the acceptability of lossy compression.
The task facing the DICOM committee—the study of lossy compression—was formidable and unprecedented. There were no prior large-scale clinical studies on which to design the current trial. After considerable debate and discussion, the following parameters were selected:
• The study would use a compression method known as motion JPEG (Joint Photographic Experts Group). Originally developed for still photography, JPEG compression is widely used outside medicine in products such as digital cameras. The JPEG method was chosen because it represented the only method available at the time with software in the public domain and associated hardware chips capable of rapid compression and decompression.
• The target compression ratios (CRs) would be 6:1, 10:1 and 16:1, hereafter designated as CR6:1, CR10:1 and CR16:1. These encompass the range of potentially useful CRs, including the range of CRs offered by the commercial products that had already been introduced.
• Digital images would be obtained from a variety of high-quality, recent-vintage catheterization laboratories worldwide. Images would be selected to include the subtle diagnostic features presenting the diagnostic challenges that require optimal image quality. These included features such as intracoronary thrombi and lesion calcification.
The committee divided the study into three phases. Phase I (Kerensky et al. ), performed primarily in the U.S., evaluates the ability of observers to detect subtle diagnostic features in angiograms. Phase II (Tuinenburg et al. ) evaluates the lossy images using quantitative coronary angiography to determine whether lossy compression affects the accuracy of quantitative measurements. Phase III (Brennecke et al. ) compares compressed images with the original images side by side to determine the threshold CR that produces a “just noticeable difference” between compressed and original images.
The tests applied to evaluate JPEG compression were extremely rigorous. The committee chose this approach because of the critical nature of coronary angiography, which is inherently limited in the quality of its images and contains barely adequate detail to detect many diagnostic features. In this respect, coronary angiography differs from many other imaging modalities in which a loss of image quality has a less important impact on diagnostic accuracy.
The results were remarkably consistent. Although these investigations were quite different in methodology, all three produced similar findings. At CR16:1, unequivocal degradation of images was evident. At CR10:1, lesser but still measurable degradation was present. At CR6:1, none of the three phases revealed any relevant differences between compressed and the original images. Thus, the three independent phases, using different methods, demonstrated a progressive loss of image quality with increasing JPEG compression.
Based on these findings, the DICOM Standards Committee has decided not to permit lossy JPEG compression for original image recording and archiving. Although the Committee has not yet precluded data compression for network transfer protocols, it remains unlikely that lossy compression will be permitted, because the DICOM standard generally does not distinguish between physical media and network transmission of image data.
There are several reasons to adopt a conservative approach with respect to lossy compression. Ultimately, the Committee concluded that it is important to set a very high benchmark for the DICOM standard. Compression at 6:1 performed acceptably in a closely controlled study employing state-of-the-art image quality and highly skilled readers. It is not clear that these findings would be replicated with lower-quality original images interpreted by less-skilled angiographers. Essentially, CR6:1 allows no margin of error.
Up to CR6:1, the advantage when compared with the currently available 2:1 lossless compression would appear limited when applied to data recording and archiving today, and its potential value will diminish further over time. The CD-ROM read speeds continue to increase. Readers now being introduced can achieve read speeds that permit 30 fps angiographic replay speeds for 512 × 512 images. Future recording media such as the Digital Versatile Disk will provide a data storage capacity equivalent to 15 CD-ROM disks.
If lossy compression is permitted, sequential compression might occur. In this circumstance, an image that had been compressed and decompressed once might be inadvertently subjected to a second or third cycle of compression. The effects of such a process on image quality are unknown but would almost certainly degrade the image further.
Thus, there appears to be little rationale to permit lossy compression for the recording and archiving of digital cardiac angiographic images. On the other hand, network transfer of image files remains an important potential application of lossy compression. For this purpose, lossy compression may offer important performance gains that would justify a modest degree of image degradation. If this is permitted, however, it is essential that compressed images be identified as altered, to warn the user of the potential for further degradation if subsequent compression is applied.
It is important that the reader understand that the study results reported in this issue of JACC and Eur Heart J apply only to one form of lossy compression, motion JPEG. Other compression approaches are under development that will likely yield better performance (improved image quality at various CRs). Therefore, the current studies do not preclude the eventual incorporation of lossy compression into the DICOM angiographic standard. Furthermore, the results of these studies apply only to 512 × 512 pixel × 8 bit images. The impact of compression on 1,024 × 1,024 angiographic images is unknown.
Because questions concerning the application of lossy compression are likely to recur, the DICOM Standards Committee will facilitate further research in this area by placing the image data set from these studies in the public domain. Thus, The International Compression Study provides a valuable resource for designing and testing new compression algorithms, and any investigator who wishes to evaluate a new compression scheme may test it using the original Compression Study data set.
Finally, it should be noted that the trials reported in JACC and Eur Heart J represent an inter-society collaboration between the ACC and ESC. We hope this represents a new strategy for combined ESC and ACC research. The authors wish to thank all parties for the outstanding collaboration that made this inter-society project successful.
- American College of Cardiology
- ↵Brennecke R, Bürgel U, Simon R, et al. American College of Cardiology/European Society of Cardiology international study of angiographic data compression phase III: measurement of image quality differences at varying levels of data compression. J Am Coll Cardiol 2000;35:1388–97.
- ↵Tuinenburg JC, Koning G, Hekking E, et al. American College of Cardiology/European Society of Cardiology international study of angiographic data compression phase II: the effects of varying JPEG data compression levels on the quantitative assessment of the degree of stenosis in digital coronary angiography. J Am Coll Cardiol 2000;35:1380–7.
- ↵Kerensky RA, Cusma JT, Kubilis P, et al. American College of Cardiology/European Society of Cardiology international study of angiographic data compression phase I: the effects of lossy data compression on recognition of diagnostic features in digital coronary angiography. J Am Coll Cardiol 2000;35:1370–9.