Impact of various imaging modalities on PACS archiving and storage

Abstract

The 11 year history of the PACS archive at the Medical University of South Carolina was used to investigate the impact of changes in technology for various modalities on PACS storage requirements.

A Unix shell script was used to perform queries against the PACS archive database to extract statistical information for the various modalities.  During the time period investigated, major changes in imaging technologies were implemented, including shifts from single slice CT to dual/quad slice and on to multi-detector 16 and 64 slice Multi Detector Computed Tomography (MDCT).

The number of images stored on the PACS archive increased by over an order of magnitude (from 131000 per month to 1.7 million images per month), mostly from MR and CT studies.  Total storage requirements increased by two orders of magnitude from 48 GB/month to 1.2 TB/month.  Increases in PACS storage has been driven primarily by the introduction of MDCT and digital mammography, currently accounting for 47% and 17% of the total storage requirements at the expense of CR (down from 70% to 18% of the total).

Traditionally, PACS storage modeling has been driven by CR storage requirements.  Current experience with our PACS has shown that this has shifted away from CR to CT, MR and digital mammography being the primary drivers behind increased PACS storage requirements.  Knowing how different modalities impact PACS storage requirements is important to the archive planning process.

Introduction

The effect of technology changes, equipment upgrades and additions in seven different modalities on a PACS archive was investigated on a well established PACS system in continuous operation for 11 years.  During this period there were significant changes and expansion in the installed imaging equipment and use.  This included changes from single and dual slice CT scanners to 16 and 64 slice CT scanners, additional MRI scanners, fluoroscopy and angiography suites, PET and PET/CT and digital mammography.

PACS was first established at our facility in 1990.  Computed radiographic equipment was installed in an effort to improve the access of portable radiographic examinations by electronic distribution to remote viewing terminals in four ICU areas.  Lack of integration to the Radiology Information System (RIS) resulting in inconsistent demographic data, limited computing power and additional operational deficiencies led to the subsequent replacement of this preliminary PACS.

In 1992, the next phase of the PACS was initiated.  This phase involved the implementation of a dedicated PACS for ultrasound and nuclear medicine.  Upon success of this 'mini-PACS', a decision was made in 1995 to go forward with full PACS implementation.  Installation began in late fall of 1995.  Throughout the first half of 1996, computed radiography, CT and MRI were added to the ultrasound and nuclear medicine modalities.  Additional modalities were added to the PACS as they were introduced to the facility.  These subsequent integrated modalities included bone densitometry, endoscopic ultrasound, digital mammography and positron emission tomography.  The current installed base of imaging equipment at MUSC includes five MRI scanners, seven NM gamma cameras, eight CT scanners, one PET/CT scanner, four digital mammography units, six CR readers, one DR room and six ultrasound units.

The MUSC PACS archive currently archives everything that is sent to it, operating under an 'archive everything-delete nothing' model (Figure StorageModel).  All images acquired at the modality are sent to one of three network gateways which contain rules that determine where

images are distributed.  The network gateways are responsible for distributing the images to a fourth network gateway that interfaces with the archive storage managers, the Web1000 server for general image viewing, and a TeraRecon thin-client server for 3D reconstruction work.  The network gateways also serve query/retrieve requests from remote workstations (PACS workstations, DICOM software on PCs, etc).  Upon image reception and validation, the studies are routed to any number of locations dependent upon the rules defined for that station or modality.  In general, by default all images are routed to the Primary Cache, Web1000, and to two archive managers. The two archive managers each possess 20TB of SATA disks for image storage and a Tape Library for long-term storage of the data acquired.   Each study routed to the archive manager is written to two tapes. For PET, CT, and MRI modalities, an additional copy of the data set is sent to the TeraRecon server.   Therefore, upon arrival of a study, either 9 or 10 copies of the data exist.  Transient, non-archived copies will eventually fall off the initial network gateway, the secondary network gateway, TeraRecon, and Web1000, thereby leaving permanent copies on the two redundant archive managers and on 4 discrete tapes in two separate tape libraries. Further discussion of archive storage workflow.

The early adoption of PACS offers an opportunity to examine the changes in the content of the data acquired and the subsequent impact of technology transformations.

Methods

Archive storage statistics for seven different modalities were studied: computed radiography (CR), direct radiography (DX), computed tomography (CT), magnetic resonance imaging (MR), nuclear medicine (NM), ultrasound (US), positron emission tomography (PT) and digital mammography (MG).  Other modalities such as digital fluoroscopy or angiography were not examined since their storage impact was found to be fairly minimal.  A Unix shell script was written to perform SQL queries against the PACS database with the results written out to a comma separated values (CSV) file.  The results from the CSV files were then stored in a MySQL database (MySQL AB, Sweden) and analyzed using Microsoft Excel (Microsoft Corp, Redmond, WA).  For each modality, monthly data on the number of studies, total images, average images/study, maximum and minimum images per study, average frames per image and total frames (for multi-frame image data) between January 1996 and February 2007 were obtained from the PACS database.  This resulted in a total of 134 months worth of data for each modality.  Digital mammography and PET were relative late-comers to the PACS archive, with only 29 and 71 months of data respectively.

For CT and MR data, information on the types of CT and MR scanners and installation dates were available from April 1999 to present (prior to this, the history of the CT and MR scanners in the department was unavailable).  This allowed some data to be normalized to the number of scanners and to correlate the installation of various scanners with observed changes in PACS storage requirements.  During the time period examined, the department went from having four single and dual slice CT scanners to the current inventory of two 64 slice scanners, three 16 slice scanners (one of which is a dedicated CT simulator unit used in radiation oncology), a dual-source CT scanner and a 4 slice scanner.  The number of installed MR units increased from 3 to 5 scanners during this time period.

Because of the way the PACS database is structured, images containing dynamic multi-frame image data are counted as a single image, even though there are multiple images in each multi-frame data set.  Counts for the average number of frames and total frames were obtained using a separate query. 

To compute the total storage requirements per month, image file sizes listed in Table 1 were used.  These values provide a reasonably accurate estimate of the total storage required for each modality.  Image compression used by the archive was not taken into account in order to simplify the analysis.  However, since most PACS archives employ lossless compression using compression factors between 2-3, the effect of compression becomes largely a scaling factor.

File sizes were calculated assuming 2 bytes stored per pixel, except for ultrasound images which are screen captures and stored as 1 byte per pixel (is this true when the images are converted to DICOM format?  Yes).  The effect of technology changes in CT imaging was assessed by examining the data from one room in our facility where the CT scanner evolved from single slice helical to a 16 slice scanner and finally a dual-source CT scanner.

Results

Between July 1996 and February 2007, the total number of studies stored to the PACS archive went from 9000 to around 17 000 per month, with CR consistently making up around 60%-70% of the studies and declining to just over 50% with the introduction of digital mammography (Figure StudyFraction).  These numbers were not correlated with the actual number of patient studies from RIS data, but simply represent what the PACS considers to be a 'study'.

The total number of images per month stored during this period went from 131 000 to nearly 1.7 million, an increase of over an order of magnitude (12.9x).  The majority of the total images were split fairly evenly between MR and CT up until late 2003 when CT began to dominate (Figure 1).  This corresponds to the time following the replacement of the CT scanners in the department with 16 and 64 slice models.

The amount of data sent to the PACS archive each month increased from 48 GB to almost 1.2 TB, an increase of a factor of 25 during the time period examined (Figure 2).  For the majority of the time period examined, CR made up the bulk of the storage requirement.  However, this began to decline around mid 1999 with a concomittant increase in CT and MR storage requirements.  The rate of decline of CR storage relative to the other modalities increased further around 2002  when the multi-slice CT scanners were installed, and still further in 2004 with the introduction of digital mammography.  Figure GBPerMonth shows the total monthly storage in gigabytes sent to the archive.  There are three clearly identifiable phases in this graph: An initial ramp-up phase followed by a long linear phase of gradual increase and then the current phase of rapid growth.

Total storage for CR remained fairly steady at around 200 GB/month.  In recent years, the majority of the increase in total storage requirements has come from CT and digital mammography.  Currently CT accounts for approximately 50% (586 GB) of the monthly data stored to the archive (uncompressed), with CR, MR and MG trailing at 17.0% (200 GB), 16.2% (173 GB) and 14.1% (159 GB) respectively (Figure 3).

For CT and MR, the maximum number of images per study has also increased dramatically in recent years with the introduction of multi-detector CT scanners and more complicated MR pulse sequences (Figure 4).  This reflects anecdotal experience of sites reporting CT studies containing hundreds to thousands of images in the archive.  In almost all cases the additional images come from post-acquisition reconstructions rather than actual scans.  For most of the time period studied, the maximum images per study remained relatively steady at around 600-800 images/study for CT.  This began to increase in mid-2003 to a point where 4000-6000 image studies were noted.  In four months between 2005-2006, there were studies sent to the PACS archive containing over 10 000 images.

The introduction of 16 and 64 slice CT scanners resulted in a significant increase in the average number of images per study generated and sent to the PACS archive which can easily be seen in Figure 5 .  A modest increase in the average number of images per study can be seen with the addition of a CT scanner (going from 4 to 5) in 1999-2003.  The change from 5 to 6 CT scanners in mid-2003 coincides with a period where the department's 5 dual and quad slice CT scanners were replaced with 16 slice CT scanners.  Following this period, a sharp increase in the average number of images per study can be seen.

Looking at the total number of CT images per month (figure CT_TotalImages), the graph can be broken down into two very distinct regions.  The early portion is well described by a linear best fit equation y = 1545.8x + 25391 (R2 = 0.946), with the slope of the graph indicating a linear

growth of around 1546 images per month.  This phase of the graph corresponds to the time period when the department's inventory of CT scanners consisted of single, dual and quad slice scanners.  The second phase of the graph is well described by an exponential equation, y = 2306.2e0.049x (R2 = 0.99).  The exponential coefficient of 0.049 represents a doubling time of approximately 14 months (the number of CT images stored each months doubles every 14 months).  Exponential growth in the number of CT images is somewhat unrealistic however, so this trend is not expected to continue.

A look at the number of images per GB (figure ImgGB) shows that the number of images/GB has increased over time, an indication that the size of the images going into the PACS archive is decreasing.  This is a reflection of the increasing contribution of CT and MRI to PACS storage demands.  In addition, the number of studies/GB (figure StudiesGB) has decreased over the same time period indicating that the size of the studies is increasing (more images/study).

In order to examine the effect of changing technology on storage requirements, the change in number of studies, total images and average images per study was examined for a particular CT site.  The room started off with a single slice helical scanner that was replaced with a 16 slice MDCT scanner in September 2003.  In October 2006 the scanner was replaced once again with a new dual-source 64 slice MDCT scanner.  Figures 6-8 show how the number of studies, total images and average images per study changed with the introduction of new scanning technology.

Study volume for the scanner remained relatively constant through the history of the room.  The large dips in the data represent the months where the room was down for scanner removal and installation.

When looking at the total images per month sent to the PACS archive (figure 7 ), the impact of multi-detector scanners is clearly seen.  During the single slice helical scanner period, image volume remained fairly constant and steady at around 20 000 images per month.  However with the installation of the 16 slice scanner, this quickly jumps to around 50 000 images per month, even though the study volume has remained nearly the same.  The introduction of the dual-source scanner results in a very significant increase in the number of images per month, climbing to well over 200 000 images per month.  The impact of MDCT is perhaps better illustrated by figure 8 which shows the average number of images per study during the history of this site.  MDCT makes it much easier to acquire thinner slices over a larger volume much faster, which is clearly shown in the graph.  Early on with the single slice scanner, the number of images per study remained constant at between 60-70 images per study.  With the additional capabilities of the 16 slice scanner, protocols changed with the average number images per study climbing to 100 and then to over 200.  A tremendous jump to over 600 images per study was observed when the dual-source scanner went online.

Shown in Figure OnlineRetiontion are plots of the number of months of online storage 2, 5, 10, 20, 30 and 50 TB disk arrays would be able to provide.  The curves were obtained by taking the cumulative GB stored data at each month, subtracting the disk array size and getting difference in months between the two.  Because of the increase in the amount of data stored in the archive each month, the capacity (in months) of a disk array decreases steadily.  In a similar fashion, the cumulative GB stored data was used to generate a graph showing the size of the disk array required to keep 24 months of storage online (figure 24Month).  At each month, the difference in the number of GB between the current month and 24 months ago was computed.

Discussion

Archive Planning

A downside to the approach taken is that the archive database is unable to accurately report the number of images contained in multi-frame images, such as those coming from fluoroscopy or ultrasound.  Each multi-frame image is counted by the PACS database as a single image, even though it may contain multiple frames/images and use up a proportionately higher

amount of space.

Understanding the use and storage patterns of each modality is an important part of planning for new PACS archive installations or upgrades of existing installations.  

When it comes to CT studies and MDCT, there are two basic approaches to handling the massive influx of images resulting from the acquisition of thin slices.  One approach is to archive only thick slices for the radiologists to review while sending the thin slices to separate short term storage for later 3D post-processing.  The other approach is to archive everything that is acquired and have the PACS archive route images to a 3D server.  The dramatic increase in the contribution of CT to archive storage requirements at MUSC is due to the 'archive everything' approach our institution has taken.

 

 

JAY WILL DO THIS SECTION.   INCLUDING PROJECTIONS. 

 

Conclusion

Three major factors have changed the profile of the information stored on the PACS archive.  First, there is a nearly linear growth pattern in storage needs.  This can be attributed to workflow efficiencies, increased patient capacity and reduced acquisition times.  A second significant factor, multi-detector CT, has not only resulted in an exponential growth in number of images per study, but a commensurate increase in the relative contribution of this modality to the storage profile.  Thirdly, the introduction of digital mammography has played a significant role in quantity of data stored.  This information may be helpful in planning storage requirements in preparation for the introduction of new modalities and technologies.

Of the seven modalities examined, four  (CT, MR, MG and CR) account for nearly 98% of the total storage demand on the PACS system.  The introduction of multi-slice/multi-detector CT systems  makes CT the dominant factor when trying to plan future PACS storage requirements or expansion of and existing PACS storage infrastructure.

Tables

Table 1: Image file sizes for modalities

Modality Image matrix size (pixels)

File size (MB)

CT 5122 0.5

MR 5122 0.5

CR 2954x3598 20

DX 2022x2022 8

NM 2562 0.128

MG 4096x3328 26.6

PT 1282 0.032

US 640x480 0.3

 

Figures

Figure StorageModel: The storage model used at MUSC.  All images are sent from the acquisition station to one of four PACS workflow managers (network gateways).  The gateways then route images to various locations including the PACS archive, 3D reconstruction servers

and web viewing server.

Figure:  Fractional studies per month by modality

Figure 1: Number of images stored each month for each modality as a fraction of the total

Figure 2: Modality storage requirement as a fraction of the total

Figure 3: Monthly storage requirements for each modality

Figure 4: Maximum images per study for CT and MR

Figure 5: Average number of CT images sent to PACS correlated with the number of CT scanners installed.  The sawtooth period in mid-2003 corresponds to a period where CT scanners were being replaced with 16 slice scanners.

Figure CT_TotalImages:

Figure 6: Number of studies per month for a single CT site

Figure 7: Total images per month for a single CT site

Figure 8: Average images per study for a single CT site

Figure OnlineRetiontion: Data retention times in months for various sizes of disk arrays. 

Figure 24Month: Amount of storage required in TB to keep 24 months of data online.

Figure GBPerMonth:

Figure ImgGB: The number of images per GB being stored each month

Figure StudiesGB: Number of studies per GB being stored each month