This Article Abstract Figures Only Full Text (PDF) Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Strickland, N H Search for Related Content PubMed PubMed Citation Articles by Strickland, N H

Multidetector CT: what do we do with all the images generated?

Hammersmith Hospitals Trust, Du Cane Road, London W12 0HS, UK

	Abstract

Top Abstract Data size Data transfer Storage cost Interpretation Data storage options Some other storage... Conclusion

 
Multidetector spiral CT technology generates much more acquisition data than the previous generation of single detector spiral CT machines, and permits more sophisticated and better quality post-processing options, which in turn generate even more imaging data. Such large quantities of data have repercussions upon the mode of image interpretation (hard copy vs soft copy), the speed of data transmission and the storage of these data.

	Data size

Top Abstract Data size Data transfer Storage cost Interpretation Data storage options Some other storage... Conclusion

 
Introduction of multidetector spiral CT machines has vastly increased the amount of data generated per study from this modality. Previously in body CT imaging using single detector spiral CT, 10 mm thick sections were routinely acquired. Using multidetector machines, 2.5 mm sections are routinely acquired in scanning the abdomen and pelvis, and 1.25 mm sections may be acquired in musculoskeletal or chest scans. These data are frequently reformatted in sagittal and coronal planes in addition to the acquisition axial plane, and 3D vascular and other post-processed images are often generated as well.

A single 512 matrix CT image contains 0.5 megabytes (512 x 512 x 16 bits) of data, and thus an average CT study composed of, say, 300 2.5 mm images contains 150 MB (300 x 0.5 MB) of data. (The thickness of the slice per se has no effect on the amount of data.) The maximum reversible (lossless) compression applicable to data is 3:1, giving a total of 50 MB reversibly compressed data for a 300 slice CT study. Had such a study been acquired on a conventional single detector scanner in 10 mm thick sections, it would have yielded 75 images, which is 37.5 MB data to be stored, or 12.5 MB data reversibly compressed at 3:1. To put such data in perspective, a plain chest radiograph contains approximately 10 MB data.

	Data transfer

Top Abstract Data size Data transfer Storage cost Interpretation Data storage options Some other storage... Conclusion

 
The rate of data transfer is crucial in visualizing digital images in a timely fashion on diagnostic or review monitors. The standard non-proprietary protocol required for practical image transfer is 100 megabits per second (Mbps) Ethernet (also known as "100 baseT"). This is the minimum data transfer protocol used in PACS (picture archiving and communication systems). It is important to realize that even when such a 100 Mbps Ethernet line is on a non-contended drop, i.e. is on a dedicated line not taking any other data traffic, the maximum data transfer occurring is only approximately 40% of the available bandwidth, that is only 40 Mbps (4 MB s^–1) rather than the full 100 Mbps theoretically available.

We measured the time taken to transfer both a 300 slice and a 75 slice CT study from the short-term PACS database, a RAID (redundant array of independent disks) device, and display them on three different workstations. The workstations used were: a diagnostic PACS workstation (a 900 MHz Dell computer using General Electric Pathspeed version 8.0 software); a PACS Web browser (a GHz Compaq computer using Internet Explorer browser version 5.5); and a third party PACS workstation (a Radworks 900 MHz Dell computer), which is DICOM (digital image communication in medicine) standard compliant and uses a DICOM query–retrieve mechanism to fetch the CT study from the PACS database. The results are set out in Table 1.

The proprietary PACS diagnostic workstation obviously retrieves and displays CT studies from the short-term PACS RAID storage device almost instantaneously, and there is only a 0.5 s extra time penalty incurred by quadrupling the number of images in the CT study.

The retrieval and display time is significantly longer for the PACS Web browser and for the third party workstation, and interestingly the latter takes more than twice as long as the Web browser to display the same studies. This highlights the fact that although universal compliance with the DICOM standard means that correctly configured PACS components from different vendors can be made to work together successfully in a "plug and play" fashion, query–retrieve DICOM transfer is relatively slow compared with a single vendor system even when using the same universal non-proprietary Ethernet transfer protocol. Image transfer using DICOM query–retrieve from a third party workstation is usually without compression, which further slows the process.

The retrieval and display time of CT (or other multi-image) studies on the diagnostic PACS workstation is actually faster than would be calculated theoretically, because the CT images start to be displayed on the monitor as soon as enough have been retrieved to fill the screen in whatever display format has been selected, while further images are being loaded into the display memory. Subsequent screen pages or cines will be displayed much more quickly, in less than 2 s.

The PACS Web browser also retrieves and displays the CT study faster than would theoretically be expected, because PACS Web browsers generally function with variable compression such that sufficient compression is automatically applied to the images so that they will only be displayed at the maximum resolution possible for the size of the Web browser display monitor. Obviously the more compression applied to an image the faster it will be transferred. A 17'' monitor gives a viewport of 800 x 600 pixels, i.e. above the 512 pixels required to display a CT study at full resolution when seen on a 1:1 cine mode format. Most PACS Web browser software starts to display images at low resolution as soon as the data arrive, rapidly decompressing them on-the-fly, so that the viewer immediately has something to look at on the display monitor: a clever psychological "trick" that diminishes frustration at having to wait for images to arrive!

	Storage cost

Top Abstract Data size Data transfer Storage cost Interpretation Data storage options Some other storage... Conclusion

 
It is difficult to make a realistic estimate of the cost of storing digital data on an archive. We made some calculations to give an idea of the order of magnitude of the costs involved. A common long- and medium-term archiving medium used today is magneto-optical disk (MOD), which is WORM (write once read many, i.e. non-rewritable). Each MOD holds 5.2 gigabytes (GB) of data. A 300 image CT study comprising 50 MB of reversibly compressed data would therefore take up approximately 0.009% storage space on such an MOD. The list price of a single 5.2 GB MOD is approximately £80 (at the time of writing), giving a cost of approximately 72 pence to archive the CT study.

This compares with the cost of conventional film for printing CT of approximately £2 per sheet. Assuming a 20:1 tile format of the images, 15 sheets of film, costing £30, would be required to print 300 images, and that does not take into account the need for printing some of the same images under different window width and level settings ("lung", "mediastinal" and "bone" windows), which could more than double the number of sheets of film required.

	Interpretation

Top Abstract Data size Data transfer Storage cost Interpretation Data storage options Some other storage... Conclusion

 
No-one would doubt that data acquired from a multidetector CT scanner must be viewed for diagnostic purposes at the resolution at which they were acquired, or else there is no point in using the sophisticated technology of multidetector scanning and acquiring thin sections over an extensive body area (with the associated penalty of an increased irradiation dose to the patient and the environment).

The majority of radiologists would agree that these data should be viewed in soft copy form on a workstation, rather than on film, since there are numerous advantages to soft copy viewing of these large amounts of data, including: speed of image review, especially in stack (cine) mode; the impression of real-time cross-sectional review scrolling continuously through the body; flexibility of viewing with rapid application of different tissue contrast width and levels; multiplanar reformats and a whole variety of post-processing applications usually available on soft copy 3D diagnostic workstations. Vendors recognize that multidetector CT scans demand soft copy viewing, and they now always include at least one specialized diagnostic workstation with the CT machine as part of the sales package. Printing the full acquired CT data set onto film produces a large number of sheets of film per study, which is cumbersome and costly for film usage and film packet storage.

It is nevertheless surprising and disheartening to find that many radiologists in the UK are in fact choosing to report these multislice CT scans from radiographic film, despite having access to a specialist diagnostic workstation. In many of these cases the radiologists are seeing only a percentage of the acquired CT data set since only a representative sample is printed to film. It could be argued that it is negligent practice not to look at all the data acquired in the study, regardless of what is stored as the final permanent record. In fact there is a precedent set by many other imaging modalities for reviewing all the acquired data but only storing a very limited amount of it, if any. This is the case in all fluoroscopic screening procedures, in angiography and in ultrasound. In the UK it is the radiological report that is the legal document, not the images.

	Data storage options

Top Abstract Data size Data transfer Storage cost Interpretation Data storage options Some other storage... Conclusion

 
There are four major options for a policy of data storage from multidetector CT scanners, and these will each be considered with their respective implications for film printing and for soft copy archiving only.

Option 1: store everything as acquired
Implications for film
The implications for film of this option are that a very large number of sheets of film need to be printed for every CT study to document all the acquired data since there are numerous different ways of viewing these data. These might include soft tissue, lung and bone window contrast settings, acquired axial sections with multiplanar reformats in coronal and sagittal planes, possibly maximum intensity projection angiograms or 3D reconstructions, etc. Printing such large numbers of films is cumbersome for handling and comparison purposes and for transport, is time consuming for the radiographers retrieving the films, and is expensive for the film budget and physical storage space.

Implications for soft copy
The implications for soft copy of this option are that it is demanding of archive space on either a local dedicated CT archive or on the hospital PACS archives, and thus is also relatively costly. In an established hospital-wide PACS it is very likely that the annual requirements for long-term archiving and the expected rate of growth of data acquisition will have been calculated prospectively prior to the institution possessing a multidetector CT scanner and therefore they will be at least an order of magnitude under-sized if large numbers of these CT scans are to be performed.

Allowing clinicians to access all these images using the Web browser means that they are liable to be pulling large quantities of data over the hospital network for review, which will take up a very significant amount of bandwidth. This is an especially important consideration if the network used for PACS Web traffic is not dedicated to image traffic alone, as is often the case, but is shared with other hospital data such as blood test results. Large amounts of imaging traffic on the network will take an irritatingly long time to download and will negatively impact upon speed of access to other hospital data on the same network, and will make the whole network seem slow.

The DICOM standard divides the CT data into numerous different series according to the acquisition and/or reconstruction parameters used. If the entire multidetector data set is presented to clinicians, the multiplicity of different series can be overwhelming! If, in addition, each series is very large because of the very high number of images it contains, then the time penalty incurred in displaying CT images on Web browser workstations will be intolerable in a busy clinical setting.

Option 2: automatically store a predetermined selection of images
All multidetector CT scanners can be set automatically to print a ratio of images, e.g. 1:4 or 1:6.

Implications for film
Clearly this policy reduces the amount of film generated and thus ameliorates the associated problems. However there is a risk that a relevant but physically small pathological feature may happen not to be visible on the automatically pre-selected image archived (the one in four, or one in six) but that was visible on the discarded images. Thus there will be no archived documentation of pathology described in the report.

Implications for soft copy
Surprisingly this option of automatically storing a ratio of images to soft copy archive is not currently possible on most multidetector CT scanners, even though the same machines can print images in this manner. Obviously it could be achieved by manual selection at the CT console but this would be a very time consuming process for the radiographers and therefore is not a viable option. When automatic pre-selection and send is possible, such a solution will obviously reduce the number of CT images needing to be archived, and will ameliorate problems associated with clinicians calling back very large CT data sets over the hospital network via the PACS Web browser. The same risk as with film of not archiving relevant pathology pertains if the archived slice does not happen to show the small area of key pathology.

Option 3: report the acquired thin sections from soft copy on a workstation, and store or print thicker sections produced by summating several thin sections
Routine body scanning with multidetector CT usually acquires sections at 2.5 mm slice thickness. If these are summated for storage or printing in blocks of four to produce 1 cm thick sections, it can be argued that at least there has been no change in the status quo, since it was previously routine to acquire and store 1 cm thick body sections with conventional spiral CT before the existence of multidetector CT scanners. The thin section data are sent to the specialist workstation for soft copy reporting, and remain stored locally there for a transient period of time depending upon the size of the rewritable local storage.

Implications for film
Multidetector CT scanners can generate and print summated images automatically. This policy obviously reduces the amount of film printed, since only a quarter of the number of images reviewed on the workstation are generated once the thinner sections are summated. Subtle pathology is less well seen, or is equivocal, on thicker sections compared with the pertinent thin section. This option makes it essential to report from the soft copy on the workstation, and not from the film, otherwise most of the data acquired would not even be looked at for reporting!

Implications for soft copy
The CT scanner can be set automatically to send summated thick sections to the PACS, and the full acquired data set to the specialist 3D workstation. This policy results in a four-fold reduction in the amount of data archived on PACS, thus economizing on storage space. It also limits the number of images in the study so that excessive numbers are not available to be pulled back over the hospital network by non-radiological clinicians for review on PACS Web browsers.

It is essential that all reporting radiologists are aware of the policy adopted and understand that they have to report CT studies from the specialist 3D workstation to see all the acquired images, and must not report directly from the PACS as they would routinely for other imaging modalities. Any pertinent thin section images showing subtle pathology seen on the 3D workstation (which may risk being less well seen on a summated image on PACS), or any reformatted images generated on the specialist workstation, can be sent separately by the reporting radiologist from the workstation to PACS to join the summated images already automatically sent to the PACS from the CT scanner. These additional images will appear on PACS as a new DICOM series in the same study.

It is generally necessary to make some adjustments to working practices to ensure that this workflow functions properly. At the time of writing, PACS manufacturers have not implemented the DICOM feature "performed procedure step", whereby separate stages of an imaging study can be verified and closed without closing the whole study. This means that the radiographers must not verify the CT study on PACS after the summated images have been sent from the CT console to PACS, or else the entire study will automatically be "closed" and thereby assume a workflow status whereby no further images can be sent to it from the specialist workstation. It becomes the responsibility of the radiologist to verify the study on PACS when it is deemed complete. It is also important to ensure that the specialist workstation is set up such that any images generated on it are saved in a non-proprietary format, so that when sent to the PACS they can be opened and properly displayed. If the post-processed images are saved in a proprietary format they contain private DICOM attributes that often prevent them being displayed on the PACS, even though they can be sent over to the PACS archive.

The thicker sections stored on PACS may show subtle pathology less well, if at all, and so, as with printing film, this option may result in a poor permanent documentation of such subtle pathology on PACS if additional selected thin sections are not sent to PACS from the specialist workstation (see Figures 1–3). Clearly the task of sending additional post-processed images to PACS represents an additional task for the reporting radiologist (or for the radiographer), which is always an unpopular workflow feature, but the specialist workstation can be set up so that it is merely a couple of mouse clicks to achieve the selected send to PACS. In the vast majority of cases it is unnecessary for additional images to be sent to PACS, so the extra work generated is minimal. In selected cases, where it is deemed always to be useful, the radiographers can be directed to send an entire additional reformatted sequence to PACS directly from the console (for example a coronal or sagittal sequence in a case of aortic dissection).

View larger version (78K): [in this window] [in a new window]   Figure 1. Hazy peritoneal fat density surrounding the pancreas (arrowed in (a)) in a case of resolving acute pancreatitis. (a) 2.5 mm axial slices displayed as acquired directly from the multislice CT scanner. (b) 1.0 cm axial slices produced by summating four 2.5 mm axial slice acquisitions, as sent to PACS. The pathology is more clearly seen on the thin section (a), although it is still diagnostic on the summated thicker section (b).

View larger version (67K): [in this window] [in a new window]   Figure 2. Two hepatic metastases. (a) 2.5 mm axial slices displayed as acquired directly from the multislice CT scanner. (b) 1.0 cm axial slices produced by summating four 2.5 mm axial slice acquisitions, as sent to PACS. The smaller metastasis, lying just posterior to the middle hepatic vein (arrow), is much more clearly seen on the thin section (a), although it is still visible on the summated thicker section (b).

View larger version (63K): [in this window] [in a new window]   Figure 3. A subtle metastasis in segment 7 of the right lobe of the liver (arrow). (a) 2.5 mm axial slices displayed as acquired directly from the multislice CT scanner. (b) 1.0 cm axial slices produced by summating four 2.5 mm axial slice acquisitions, as sent to PACS. The metastasis is not seen on the summated thicker section (b).

In my institution, after an initial period where we used option 3, we have changed to option 4, probably more as a precautionary measure for self-reassurance than because we truly believe it to be medically necessary. We have chosen to use rewritable double-sided optical discs. Each side of these discs has a capacity of 2.3 GB, which means they hold approximately 4700 CT images per side. Our intention is to start rewriting over the data stored on these discs in 2 years.

	Some other storage considerations

Top Abstract Data size Data transfer Storage cost Interpretation Data storage options Some other storage... Conclusion

 
It is important to remember that the legal record of an imaging study in the UK is the report generated, not the images. There is no legal obligation to make a permanent record or to keep the imaging data from any imaging study in the UK (with the exception of minors and educationally subnormal patients). The Royal College of Radiologists recommends retaining the images for a period of 7 years, but this is only a recommendation.

As already mentioned, there are plenty of precedents in radiology (and other branches of medicine and surgery) for retaining only a subset of the acquired image data, i.e. those considered to show the pertinent pathology. It is not common practice, for example, to keep a video record of an entire barium meal study, ultrasound or bronchoscopy. Thus there is no reason why "a special case" should be made for multidetector CT simply because it was customary to retain all acquired images from the previous generation of CT scanners.

Most PACS permit the manual marking ("flagging") of chosen images in a study as significant or "key" images. These can then be displayed subsequently as a separate series, obviating the need to display the entire study. This is a particularly useful feature for:

• radiologists running a clinicoradiological conference when trying rapidly to demonstrate to an audience the significant pathology in a large cross-sectional imaging study containing many hundreds of images, especially if the radiologist has not reviewed that study him/herself previously prior to the meeting;
  
• clinicians wishing to review the salient images in such a large cross-sectional imaging study. The relevant images demonstrating the pathology have already been pre-selected for him/her by the reporting radiologist and can be displayed relatively quickly on the PACS Web browser as a separate series, without the need to download the entire study across the network to the Web browser PC.
  

The only problem with the PACS feature of designating images from a study as "significant" is that it falls to the reporting radiologist to perform this task, which takes extra time in addition to the diagnostic review and report generation, and is without direct benefit to the radiologist marking the significant images. Thus, the motivation for marking significant images is low and it is not very often done in practice even when the appropriate software is available and easy to use. PACS that automatically assign a key image status to any image upon which any soft copy mark has been added (e.g. a measurement made or an arrow added) are advantageous in this respect.

The two major concerns discussed in this article – storage (space and cost) and network bandwidth limitation (with the movement of very large data sets having a negative impact on image transfer speed) – will shortly become inconsequential. The rapid progress and development of information technology will very soon provide huge available bandwidth and virtually limitless storage. A number of new magnetic disk hardware devices are now coming onto the market. These include SAN (storage area network), NAS (network attached storage) and CAS (content addressable storage). Devices such as a CAS have a huge storage capacity (of 27 terabytes at the time of writing) yet are comparable in price with conventional MOD technology. These new magnetic disk hardware devices will markedly reduce the current RAID storage problem and render it negligible. Retrieval of large data sets to PACS Web browsers slowing traffic on the hospital network will remain, but will be overcome when hospital networks can offer sufficiently large bandwidth at an affordable price.

	Conclusion

Top Abstract Data size Data transfer Storage cost Interpretation Data storage options Some other storage... Conclusion

 
Multidetector CT scanner technology, with its acquisition of volume sets of data rather than single slices, generates vast numbers of images, augmented by the post-processing capabilities that this technology facilitates. These data sets must be reviewed as soft copy on sophisticated workstations. It is the current limitations of soft copy data storage and retrieval that are posing a problem, especially when it is desirable to integrate these studies onto a hospital-wide PACS. Fortunately the "work-around" solutions discussed in this article will be short-lived, as rapid developments in storage devices and network bandwidth will soon provide an acceptable solution. This is important not only for the modality of CT, but also to solve the need to archive increasingly vast amounts of digital data generated by other modalities such as functional MR dynamic studies and digital microscopy.

	Acknowledgments

 
The author wishes to thank Christopher Drinkwater (General Electric Medical Systems [GEMS] IT PACS field engineer at Hammersmith Hospital, London) for help with measuring the CT image transmission times to the various workstations, and Robin Barclay (GEMS IT Region Leader Northern Europe) for help calculating the cost of archive media.

This Article Abstract Figures Only Full Text (PDF) Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Strickland, N H Search for Related Content PubMed PubMed Citation Articles by Strickland, N H