Is 'Joint Photographic Experts Group (JPEG)'. ISO/IEC JTCl/SC29/WG1 Convener - Dr. Daniel Lee. Hewlett-Packard Company, 11000. Wolfe Road, MS 42U0, ...
Integration of Radiographic Images with an Electronic Medical Record J. Marc Overhage MD PhD, Alex Aisen MD, Michael Barnes MD, Mark Tucker, Clement J. McDonald MD Departments of Medicine, Radiology and Family Practice, Indiana University School of Medicine and the Regenstrief Institute for Health Care, Indianapolis, IN clinical information for many years at Wishard Memorial Hospital, University Hospital and Riley Hospital for Children.'3 In 1997, the Department of Radiology installed a PACS system (Agfa Impax) at all three ofthese hospitals. Inpatient, outpatient and emergency department studies were included in the PACS implementation. As is typical, radiology installed PACS review stations in selected locations such as ICUs as well as in the radiology department. In 1998, as we were converting the RMRS to provide web access, we decided to integrate access to the radiographic images stored in the PACS into the RMRS.
Abstract Radiographic images are important and expensive diagnostic tests. However, the provider caringfor the patient often does not review the images directly due to time constraints. Institutions can use picture archiving and communications systems to make images more available to the provider, but this may not be the best solution. We integrated radiographic image review into the Regenstrief Medical Record System in order to address this problem. To achieve adequate performance, we store JPEG compressed images directly in the RMRS. Currently, physicians review about 5% of all radiographic studies using the RMRS image reviewfunction.
Design objectives Our objectives in integrating radiographic image access into our EMR were: (1) simplify access to radiographic images by removing barriers and increasing speed; (2) integrate image data with other relevant information including radiographic reports; and (3) be cost effective in terms of network bandwidth and review workstation costs.
Introduction Radiographic images are important and expensive diagnostic tests. However, the provider caring for the patient often does not review the images directly due to time constraints; relying instead on the radiologist's dictated report. If the provider does review the image, they must often travel to a hospital area distant from the patient care areas and then locate films in the reading room or file room. If, instead, images can be distributed in a rapid and efficient fashion to points of clinical care, we could improve quality and reduce costs of patient care.1.2-4.S Picture archiving and communications systems (PACS) can be used to distribute images, but this solution is costly,67 requires additional space,8 consume significant network bandwidth9 and increases the end user support burden.'0 An altemative method for distributing images is to integrate the image review function within an EMR or HIS.""2This approach integrates all the relevant data in a single place and overcomes most of the limitations of using a PACS for this purpose.
System description Our original design was to store thumbnail of each image in a radiographic series along with a pointer to the full size image stored in the PACS in the RMRS and then perform a DICOM14 fetch when a user needs to view the image. Unfortunately, the archive was not able to handle the demands imposed by this approach. The time to retrieve a study and display the first image for the user was often over one minute and could be five minutes. This performance level was unacceptable and we modified our strategy to store a JPEG'5 compressed copy of each image in the RMRS database itself. We had previously implemented a bulk or image storage facility within the RMRS database. This facility was designed to store large binary data such as electrocardiograph and radiographic images. We chose quality factors for the JPEG compression that resulted in 10:1 compression for
Background The Regenstrief Medical Record System (RMRS) has served as a primary source of
1067-5027/01/$5.00
2001 AMIA, Inc.
513
v _ ~ ~ -V
study while it displays conventional radiographs
conventional radiographs and 20:1 for CT/MRI/US scans. Modaliy
F.-1 - .1 7.1
Kodak DR
Agfa
Archiw
as individual "films" at one-quarter size. We find that, for some purposes, these thumbnails are sufficient for the care providers' needs.
Imlage Proc)essor
The image processor requires 30 seconds to process each conventional radiograph and 2 seconds to process each CT/MRI/US image. There are a highly variable number of images to store, so this time is quite variable. The ratelimiting step at this point is sending the images to the RMRS, waiting for it to be stored and the RMRS to send an acknowledgment. We have identified a few inefficiencies in this process that we can improve.
\jM K
The RMRS receives textual radiology reports as HL7 messages after the images arrive and stores them in the RMRS TRF or text report facility. Various systems send reports with accession numbers that allow the software to unequivocally match them with the images they describe. Other systems, however, send reports without accession numbers and we must rely on the modality identifier and parsing portions of the text report to match the report with the image. Because the image processor receives the image first and the DICOM study does not include a very good description of the study, we simply label the image as [modality] imaging study (MRI imaging study or US imaging study) and then replace this label with a more specific label (LS Spine MRI with gadolinium or Testicular Ultrasound) once the software matches the image to a report. There were commonly studies of multiple body parts in a single DICOM study (a head and an abdomen for example). Often, in a specialized environment such as ours, different radiologists would interpret these studies and thus separate textual reports describing each of these areas making matching the appropriate reports and image series challenging.
Figure 1 -- Diagram of image flows for integrating PACS images with the RMRS EMR. The MRF stores numeric or coded data and the TRF stores textual reports. Figure 1 is a diagram of the data flow. We primarily rely on a DICOM push from the archive to indicate the availability of an image but backstop the push strategy with polling the archive periodically to identify images that are in the archive but are not stored in the RMRS. The software compares the list of studies added to the archive in the interval since the last query with the studies received by the push mechanism and then retrieves any missing studies from the archive. This DICOM push is to an Intel 500 MHz single processor computer with 256 Mbytes of RAM that queues the DICOM studies and processes them to create the thumbnail images, JPEG compress the images and store the links along with this data in the RMRS. This computer is located on a network segment with the image archive to minimize impact on overall network performance. The links are pointers to the image data in the RMRS image file but also contain pointers to the original DICOM study. This pointer will allow us to revert to our original strategy of relying on the PACS for image storage and retrieval once the archive can provide adequate real-time performance. This also allows the PACS to serve as a backup in case the RMRS has trouble retrieving the image from the image file. This might occur during peak processing times when the image processor may fall somewhat behind in compressing the images and storing them in the image file. The software creates thumbnails to emulate the film formats with which users are familiar. The software displays CT/MRI/OS images as "contact sheets" of images from a
1067-5027/01/$5.00 © 2001 AMIA, Inc.
In order to minimize the training and support burden and to improve the opportunities for spontaneous discovery, we chose to make access to radiographic images and reports similar to access to all other data in the RMRS.
Clinicians log in or authenticate themselves to the RMRS and then they can review data either as a flow sheet or as clinical result reports. In
514
anou--BGI
Abdomen MRI
I
I Pulmonary
i
i
I
ii
0
r Abdomen XR
IMPRESSION: WISHARD 1. Abnormal but nonspecic bowel gas pattern.
ER
nonspelmc bowel gas pattemrn
E
r _C__be_
t
cavItation.?f
__
..................................................................atrll r. CT.....
r Cles PA & Let XR
HOSP right "said NOS bIlatera alveolar
In1trate Interstitial mark
heart normal bilateral alveolar
nItrate
WISHARD ER WJL
lltrate? overlniatlon
IMPRESSION: WISHARD Intarval ER decrease In nag lt Iniltrate.
I
lnula Iftrate sme
Figure 2-Example flow sheet display illustrating the summarized radiographic information and links to reports and images CT/MR to determine key aspects of the series either case, they can limit the results they want to and used these to create clinically meaningful review to those of interest or they can browse labels. Conventional radiograph labels were results to find those they are looking for. The RMRS browser interface (Figure 2) displays the usually adequate to identify the series (PA and Lateral for example). impression from the radiology report along with one link to the full report (document icon) and a When the user clicks an image link, the software second link to the images Qong bone icon). displays the thumbnails in less than one second. To display the images, we created a simple threeAt this point, the program has only sent the thumbnail images to the browser. Once a user pane image viewer using HTML and JavaScript selects an image, the software retrieves that (Figure 3). The viewer provides navigation of images within the study, zooming and image and sends it to the browser that displays the image in the top right panel of the viewer at a window/level controls. Users can navigate default size (1/4x for CR and 2x for CT/MRI). among images both by selecting the series and then selecting the image within the series that The user chooses a different size or changes the window/level for the image, the browser requests they want to review and by reviewing the thumbnails and selecting the images of interest the new image that the software constructs on the by clicking. The bottom pane displays the server from the original data. This approach thumbnail images, the top left pane the controls ensures that the software only sends the data the and the top right pane serves as the image users needs to the browser. The viewer displays the new image within 1-5 seconds depending on display space. The viewer also provides a link to the text report, the DICOM header infornation, the image size. and the IU Radiology Departnent's on-line atlas Status Report of normal radiographic anatomy. We used The image review functionality has been elements of the DICOM study header for available to users from any browser capable
515
La.
Figure 3- Sample of the three-pane browser based image viewer. The bottom pane shows the thumbnail images; the top left pane contains the navigation, scaling and window/level controls and the toD rieht nane contains the imaee clinical workstation for about 18 months. During a typical one week sample period in February 2001, there were 1,834 CRs, 328 Computed Tomograms, 62 MRIs performed at Wishard Hospital and 523 CRs, 463 CTs, 143 MRIs, and 114 Ultrasounds performed at University/Riley Hospitals. During that week, physicians accessed 799 images in 98 studies (4.4%) at Wishard and 999 images in 69 studies (5.5%) at
University/Riley using the EMR viewer. Figure
>
4 shows the trend in average daily usage over several months.
Aug- Sep- Oct- Nov- Dec- Jan- Feb00
00
00
00
00
01
Discussion We have succeeded in tightly integrating a PACS system with our large EMR. We achieved our tdree objectives: (1) simplify access to radiographic images by removing barriers and
Figure 2- Recent monthly usage of the RMRS image viewer
increasing speed; (2) integrate image data with other relevant information including radiographic reports; and (3) be cost effective in terms of network bandwidth and review workstation costs.
Only portable and emergency room radiographs are made digitally at IU/Riley
516
01
First, we simplified access by integrating the radiographic images into the RMRS electronic medical record system. This integration eliminated the barriers of remembering another set of user IDs and passwords and learning to navigate another application.
Binkhuysen FH B. Required functionality of PACS from clinical point of view. International Journal of Bio-Medical Computing 1992; 30(3-4)):187-91. 3 Reiner BI. Siegel EL. Hooper F. Protopapas Z. Impact of filmless imaging on the frequency of clinician review of radiology images. Journal of Digital Imaging 1998; 1 1(3 Suppl 1):149-50. 4 Gur D. Straub WH. Lieberman RH. Gennari RC. Clinicians' access to diagnostic imaging information at an academic center: perceived impact on patient management. AJR. American Journal of Roentgenology 1992; 158(4):893-6. 5 Reiner BI. Siegel EL. Hooper F. Pomerantz SM. Protopapas Z. Pickar E. Killewich L. Picture archiving and communication systems and vascular surgery: clinical impressions and suggestions for improvement. Journal of Digital Imaging 1996; 9(4): 167-71. 6 Wrburton RN. Digital imaging at a community hospital: implications for hospital stays and teleradiology. Intemational Joumal of Bio-Medical Computing 1991; 28(3):169-80. 7 Pilling J. Problems facing the radiologist tendering for a hospital wide PACS system. European Journal of Radiology 1999 Nov; 32(2):101-5. 8 Erickson BJ. Ryan WJ. Gehring DG. Beebe C. Clinician usage patterns of a desktop radiology information display application. Journal of Digital Imaging 1998; 11(3 Suppl 1):137-41. 9 Dwyer SJ 3rd. Imaging system architectures for picture archiving and communication systems. [Review] [43 refs] . Radiologic Clinics of North America 1996; 34(3):495-503 '0 Parasyn A. Hanson RM. Peat JK. De Silva M. A comparison between digital images viewed on a picture archiving and communication system diagnostic workstation and on a PC-based remote viewing system by emergency physicians. Journal of Digital Imaging 1998; 11(1):45-9. " Bradley JE, Ryan J, Gehring DG, Beebe C. Clinician usage patterns of a desktop radiology information display application. J Dig Imaging 1998; 11(3) Suppl 1: 137-141. 12 Dayhoff R. Integration of medical imaging into a multi-institutional hospital infonnation system structure. MEDINFO '95 Procceding 1995: 407-410. 13 McDonald CJ, Overhage JM, Tiemey WM, Dexter P. The Regenstrief Medical Record System. A quarter century experience. International Journal of Medical Informatics 54, 225-253. 1999 2
Second, we were able to combine the images with the report describing the image in the context of other clinical data such as the medication list and microbiology results. Third, because of the JPEG data compression, there has not been a deleterious affect on network function performance and we haven't deployed any further PACS workstations. Compressing the images also allows them to display very rapidly (1-5 seconds). While usage has been slowly increasing, the penetration is still not high. We speculate that the major reason for the modest level of usage is that "old habits die hard" - changing physician behavior is always difficult. There are several other reasons that we believe play a role as well including: the limited number of workstations with sufficient resources to run the image viewer, the value that physicians derive from consultation and discussion with the radiologists when they review films in the radiology department and the lack of links to the images from the report describing the study. There are benefits of ready image availability beyond patient care. Some faculty feel that the availability of images has had significant effect on teaching and oversight of house staff and students. First, rounds no longer are portioned into seeing the patients and then seeing the films. Rather the two processes are integrated in a manner that fosters more rapid decision-making and a more integrated teaching model. Second, the faculty who are supervising house staff can now readily review radiographs from their office or from home to ensure that they are properly interpreted and acted on in a timely fashion. In addition to having the house staff describe the film, the faculty can be examining the study themselves.
14 Digital Imaging and Communications
in Medicine (DICOM). NEMA PS3. 1-PS3.12. Rosslyn, VA.: National Electrical Manufacturers Association, 1992, 1993, 1995, 1997, 1998. Is 'Joint Photographic Experts Group (JPEG)' ISO/IEC JTCl/SC29/WG1 Convener - Dr. Daniel Lee. Hewlett-Packard Company, 11000 Wolfe Road, MS 42U0, Cupertino, CA 950114.
References 'Mosser H. Urban M. Hruby W. Filmless digital radiology--feasibility and 20 month experience in clinical routine. Medical Informatics 1994; 19(2):14959.
517
