One hallmark unites the winning entries in the top five medical-imaging IT projects of 2012, cosponsored by Radiology Business Journal and the Society for Imaging Informatics in Medicine (SIIM): Each project represents a view beyond the traditional acquisition, archiving, and communication of radiological images. All of the winning entries take a global view of medical imaging:
- mining the data in the DICOM headers and dose sheets to produce a relevant number for patients’ exposure to radiation;
- solving the technical and operational problems of including non-DICOM images in PACS;
- creating a nonlinear, flexible workflow layer that can tell the radiologist whether a brain tumor has grown before he or she looks at the image, as well as creating a worklist for a geographically disparate organization;
- solving the interoperability issues inherent in the movement of pathology images to create a digital consultation portal for pathology; and
- scouring the electronic medical record (EMR) for the data required to create a safe protocol for a study.
The entries were judged on their innovation/ingenuity, on whether they met critical/urgent/unmet needs, on whether they improved quality, on the product/tool/idea validation or evaluation, and on the universality of the application. All six judges are members of the SIIM board: Donald K. Dennison is an imaging-vendor executive; J. Raymond Geis, MD, is a radiologist with Advanced Medical Imaging Consultants, PC (Fort Collins, Colorado); David S Hirschorn, MD, is director of radiology informatics at Staten Island University Hospital in New York; Elizabeth A. Krupinski, PhD, FSIIM, is a research professor in the departments of radiology and psychology at the University of Arizona; Wyatt M. Tellis, PhD, is an informaticist in the radiology and biomedical imaging department at the University of California–San Francisco; and James T. Whitfill, MD, is CMIO of Southwest Diagnostic Imaging, Ltd (Scottsdale, Arizona). Winners received a $1,000 scholarship to subsidize the cost of attending the 2012 annual SIIM meeting (an award made possible by an unrestricted grant from Bayer HealthCare, maker of the Certegra™ informatics platform). Radiation Dose Intelligent Analytics for CT Examinations (RADIANCE) In June 2012, Cook had one month remaining in her residency at the Hospital of the University of Pennsylvania in Philadelphia before she would begin a fellowship there in cardiovascular imaging. As a combined engineering and computer-science major in undergraduate school, Cook’s interest in IT actually preceded her interest in radiology. Cook credits mentors William Boone, MD, and Woojin Kim, MD, for their roles in the birth of the open-source CT dose-monitoring solution called RADIANCE. “We were at a conference when we started discussing the dose-monitoring problem,” Cook recalls. “When I got back to Philadelphia, I tested a few open-source optical–character-recognition (OCR) tools and stumbled upon something that worked. From there, I continued building and extending the application, and within a few months, we realized that we had a tool that could be useful not only to us, but to the rest of the community.” From the dose sheet, kilovoltage, milliamperage, reference milliamperage, volumetric CT dose index, dose–length product, and phantom type are gathered by RADIANCE. It also imports the total study milliamperage and dose–length product, if they are reported. RADIANCE pulls information from the DICOM study header about the study type, scanner, imaging facility, and patient. These data are stored in a searchable database, on top of which the RADIANCE dashboard, scorecards, and tool kit are built. “We can also query the RIS to pull information such as patient height/weight and personnel involved with the study,” Cook says. Over the summer, Cook plans to develop the HL7 messaging necessary to send the data to an EMR, RIS, or dictation system—news likely to cheer California providers required to include volumetric CT dose index and dose–length product in the imaging report. Problem/Objective There is growing interest in monitoring imaging-related radiation delivered to patients. With CT dose parameters conventionally stored as pixel data on image-based dose sheets, large-scale archival/analysis of these data has proved challenging. Data recorded on these dose sheets represent dose delivered to an acrylic phantom, not the individual patient. Solution Newer CT scanners include dose-related parameters in the DICOM study header, as part of the Radiation Dose Structured Report (RDSR). Scanners that cannot produce RDSRs, however, will continue to generate image-based dose sheets. RADIANCE was developed to address this vast repository of CT exams with legacy dose sheets. RADIANCE uses OCR to parse image-based dose sheets from all four major CT vendors and stores dose parameters in a relational database. It can import scanner-generated RDSRs, allowing facilities to maintain a centralized repository of all dose data, even with scanners of different ages. RADIANCE can also generate RDSRs from legacy dose sheets, allowing facilities that produce dose sheets to participate in the ACR® Dose Index Registry. Results RADIANCE has been used to process over 250,000 CT exams from three hospitals. This has produced an extensive database of dose parameters that can be analyzed. Using the RADIANCE dashboard, section chiefs and department administrators can review dose estimates for particular study types and compare doses across scanners (to identify opportunities for protocol optimization). They can review dose estimates according to personnel involved; the emergency department can review dose estimates for a particular patient before deciding on a study. Monthly RADIANCE scorecards are distributed to radiologists, technologists, section chiefs, and clinical administrators. The radiologists and technologists receive an overview of their average dose estimates for studies, as compared with their doses for the preceding month (and the rest of the department’s doses for the current month). Section chiefs can review data for individuals, as well as for specific study types. Administrators can examine long-term dose trends for particular studies to analyze the effect of protocol-optimization measures and the consistency of applying protocols across scanners. Using data from the RADIANCE dashboard and scorecards, the thoracic-imaging section of the department successfully implemented protocol modifications for four common studies: enhanced and unenhanced chest CT exams, pulmonary-embolism chest CT exams, and high-resolution parenchymal chest CT exams. By optimizing tube voltage for patient size (for example, using 100 kV for average patients and 120 kV for larger patients), decreasing the reference tube current, always using tube current modulation, and scanning the patient more than once only if ordered by the radiologist or referring physician, dose estimates for the four study types have been decreased by 30%, 38%, 30%, and 65%, respectively.
The Mayo Clinic DEWY team included (from left) Xiojiang Yang, PhD, Todd French, Bill Ryan, Daniel Blezek, PhD, Steve Langer, PhD, and Bradley J. Erickson, MD, PhD.
DICOM-enabled Workflow-engine System (DEWEY) Some years ago, Erickson and a team at the Mayo Clinic (Rochester, Minnesota) partnered with a global IT company on high-performance computing to put into practice some of the algorithms for computer-aided diagnosis they developed. Invariably, they faced challenges implementing them; for each parameter, they had to create a special workflow for the technologist to send the required data for processing—and another to get the results back to the radiologists. A representative from the IT company told the team about a workflow engine that had been used in other industries to help orchestrate different events. Erickson views this as a seminal moment in the development of DEWEY. “What we built there, we don’t use anymore, but it was a great learning experience,” he says. Radiologists are in no danger of losing their jobs to DEWEY, Erickson says, but the application does assist them in some tedious tasks, such as looking for brain aneurysms or tumors on MRI studies. “Computers are good at the things that humans tend not to like,” Erickson explains. “One of the most common applications of computer-aided detection is screening mammography; it is a task that many radiologists don’t particularly enjoy. It’s very repetitive, and you’re just looking for one thing: cancer. Because the task is very focused on one goal, computers can do that fairly well.” Initially, the team intended to commercialize the system as a next-generation image-management solution that would support the complex workflows of large organizations with multiple image-acquisition sites. Instead, the team developed a system that would work in tandem with PACS and RIS, rather than replacing them. The workflow engine was DICOM enabled to initiate a complex computer-aided–detection workflow to find change in a brain tumor; to create a worklist for a multisite enterprise that will determine where a study should be read based on individual radiologists’ schedules; and to perform any other specific task based on the data in the PACS/RIS. “It’s like the change from procedural programming languages to object-oriented languages,” Erickson explains. “With procedural languages, you build a series of steps in order to get the result, whereas with object-oriented languages, the object has properties that make sure it is processed properly.” Problem/Objective Complex organizations necessarily have complex workflows. Current radiology systems have not kept up with that complexity, relying on linear workflow consisting of few steps. Erickson developed DEWEY to support highly complex workflows that are easily maintained by nonprogrammers. Solution Erickson extended a well-established workflow engine to deal with DICOM data. The system was implemented in phased fashion at the practice. Some workflows were automated and fairly linear, while others involved decision points and DICOM queries. The most complex required a specific MRI acquisition and the most recent matching prior exam, sending that pair of exams to the algorithm and waiting for the algorithm to be completed (or to fail, if it ran longer than the specified time). Erickson also developed a workflow that captures DICOM elements reflecting software versions and compares them with the most recent software-version description, allowing automated detection of modifications to imaging devices. Results Detection of brain-tumor change is a complex workflow because it requires recognition that a brain-tumor protocol is being acquired in the current exam and requires recognition that a prior exam of that type exists in the archive. This is challenging for humans and machines to determine. Prior to the implementation of DEWEY, the current and prior results were available in time for interpretation in 29% of brain-tumor cases. After implementation, 95% of cases were successfully identified and delivered (there were occasional failures of the algorithm itself, as well as occasional failure to identify the proper exams). A simpler workflow was identification of MR angiography (MRA) exams of the brain. For these, Erickson developed an algorithm for highlighting regions suspicious for aneurysm. Preimplementation success was about 70%, but after implementation of this workflow, the success rate was 98%. Failures were primarily due to software changes in the MRA parameters, which resulted in failure to detect the MRA sequence. Erickson also implemented software-change detection. This system was developed to detect changes (usually upgrades) to imaging devices that were not communicated to the department. Most detected changes had been properly communicated, however, and no changes resulted in significant outages during the test period. The motivation for this system was that a system outage previously had been created by an unannounced system upgrade that had subtle (but important) effects on the department. DEWEY uses a graphical user interface to define workflows. This provides a good way to develop, document, and maintain workflows in a complex department. These tools are relatively simple to use and do not require programming expertise. Visible Light Imaging With Enterprise PACS and EMR Kennedy’s winning solution for accommodating visible-light images in the enterprise PACS and EMR of Kaiser Permanente Northern California (Oakland) was prompted by a direct request from leaders. Both the operating room and the dermatology service needed PACS functionality, but neither worked with existing radiology workflow. Getting the images into the PACS was not the problem: The team was able to purchase a solution from a trusted vendor/collaborator that applied the DICOM wrapper to the visible-light images. The problems arose when Kennedy and his team tried to fit the alien workflow into a radiology-focused PACS. “I have been doing radiology-based workflows for over 20 years now,” Kennedy explains, “so a lot of assumptions are hardwired, with me. The two most difficult things were unlearning many assumptions and then trying to adapt a radiology-focused PACS to the new solution.” An initial hurdle was moving beyond the procedure-based context that defines radiology, primarily for billing purposes. “In radiology, it’s very clear to me: I have a CPT® code, I have a given procedure, and I have a description for that procedure that’s out of the data dictionary,” he notes. “It’s relatively mechanistic, in terms of how the RIS and the PACS deal with it; it is much more free flowing in the visible-light arena.” He adds, “Most PACS work with the procedure codes and the procedure descriptions out of the dictionary, with the idea being that they want the two alike because they are used for billing. You are not paid in dermatology, however, for the photography that you do.” Ultimately, the team chose to tie the procedure identifier and the procedure number directly to the visit itself in the EMR, so the patient visit becomes the encounter and the accession number. “That took some rewiring, in terms of how the RIS and the PACS deal with each other,” he says. A few other disconnects included the need to deal with color content in a system that is monochrome based. “In general, window level (for example) doesn’t work in color, in our radiology environment,” he explains. “People want to be able to manipulate the images, but they don’t have the tools to do that.” The team also had to develop a polychrome overlay because, particularly in gastroenterology cases, “We couldn’t see the overlays in the images because they needed to have a contrast color and shadow to make them visible,” he explains. Kennedy and his team were taken aback by the response to the visible-light imaging solution: What began with two use cases has quickly multiplied into approximately 20, and teledermatology is both underway and supported by PACS in the Kaiser Permanente Northern California region. In fact, Kennedy predicts that, in time, there will be more visible-light images than radiology images in the PACS. Another issue that was not anticipated was the need to train physicians’ office personnel to take good-quality images. “When you put a camera in someone’s hands, you don’t think of it as a modality; you think of it as just a camera,” Kennedy says. “You basically have to grow a culture around image quality.” Kennedy’s team initially spent a lot of time emphasizing the fact that a physician can’t read a fuzzy image, which requires the patient to return. “This was, frankly, a case of if you build it, they will come,” Kennedy says. “We’re very seriously looking at an entirely new PACS solution for visible light. One of my team members calls it the stem-cell PACS: an undifferentiated tool, capable of being much more flexible than the existing repurposed radiology PACS.” Problem/Objective This project’s aim was to integrate visible-light imaging with enterprise PACS and EMR, and to include support for dermatology, gastroenterology, and surgical endoscopy. Solution The team used DICOM-wrapped JPEG images and MPEG video content, as well as HL7 messaging, to integrate new image-content streams into PACS and the EMR. Results Dermatology and gastroenterology are now fully integrated, for both outpatient and inpatient workflows, in the institution. Work is now progressing for surgical-endoscopy support for 220 operating rooms in the Northern California region. The team discovered very specific differences between traditional radiology workflows and what was needed for appropriate visible-light clinical-use cases in this deployment. Significant effort was needed to adapt or modify traditional radiology workflow tools for use with these new cases, and a number of compromises were made. In the future, Kennedy’s team hopes, more workflow-agnostic systems will come to market to support nonradiology/noncardiology workflows, but few are currently available. Adapting traditional radiology workflow tools for use with visible-light cases will generally result in some level of compromise, and additional effort will be required to meet clinical requirements. Current volumes for visible-light content for the enterprise are around 1,600 studies per day, divided between dermatology and gastroenterology. Full implementation of the operating-room solution is projected to raise the total to around 3,000 studies per day. The Digital Pathology Consultation Portal You would have to travel back in time more than a dozen years to witness the genesis of this winning entry from the imaging-informatics team—including William Cable; Andrew Lesniak; Eugene Tseytlin; Jeffrey McHugh; Liron Pantanowitz, MD; and Anil Parwani, MD, PhD—at the University of Pittsburgh Medical Center (UPMC) in Pennsylvania. As Romero Lauro explains, UPMC was already working on the basics of digital pathology consultation in 1999, when he was CIO of the Mediterranean Institute for Transplantation and Advanced Specialized Therapies (www.ismett.edu), Palermo, Italy, a leading transplant hospital. UPMC helped to build the facility and continues to manage it under an agreement involving two hospitals, the regional government of Sicily, and UPMC. There are clear parallels between the early days of teleradiology and the birth of digital pathology. “We started with the basics: a camera on a microscope, taking pictures,” Romero Lauro recalls. “The pathologist would have to send cases via Internet, attaching the images and sending questions via email. We thought we could do better than that, and we started building more and more automation. It has been a continuous evolution.” In time, UPMC transitioned to a more automated process whereby the remote pathologist was able to upload (to various subspecialists in the UPMC system) not just static images of areas of interest, but entire cases. “As long as we were doing consultations within our own walls—our own domain—that was relatively easy to accomplish,” he says. “Even when we were dealing with Italy, we knew what the challenges were, so we knew what to expect. What really made us think about how to take this to the next level was the opportunity that came to work with a commercial laboratory in China.” He knew that the joint-venture partner, KingMed Diagnostics (Guangzhou, China), was not necessarily going to use the specific vendor solutions in use at UPMC, raising interoperability issues prevalent in digital pathology, where many vendors have their own proprietary imaging formats. To create an environment that would be as vendor agnostic as possible—and easy to implement—the solution uses key open-source software and other custom-developed software, including software that addresses the potential interruptions that can occur when transferring huge whole-slide–image datasets over the Internet. Clients can choose from two options for image sharing: a simpler asynchronous transfer or a client–server method that allows viewing of whole-slide images (but requires more advanced network connectivity between institutions). “With this solution, we create a market for pathology that is open to the world,” Romero Lauro says. Problem/Objective UPMC developed a set of Web-based tools to support remote digital pathology consultations and viewing of whole-slide images. The team addressed the challenge and practical implementation of two different user types: the occasional user (professional or patient) uploading digital whole-slide images and requesting a second opinion online, and the external laboratory or hospital looking for an established real-time consulting relationship covering a high volume of cases. Solution The Digital Pathology Consultation Portal provides subspecialty coverage across the 20 hospitals of UPMC and to external organizations, giving immediate access to expert diagnostic anatomic pathologists in different subspecialty areas. The UPMC Digital Pathology Consultation Portal saves overnight-shipping fees, and results can be received much more quickly than they would be using traditional overnight mail. Pathology-consultation tools based on digital slide images need to allow the client to share imaging data with the consulting pathologist. The solution developed at UPMC allows both the asynchronous upload of whole-slide images and real-time streaming of imaging data directly from the source storage location. Results The average size of whole-slide imaging studies in digital pathology is thousands of megabytes (in contrast to other studies in other specialties, where the average is tens or hundreds of megabytes). The asynchronous-transfer method developed at UPMC leverages technology that frees the user and the remote workstation during the transmission of large datasets, that automatically monitors the transmission for interruptions, and that resumes transmission from any point of failure. UPMC’s solution has the option of installing a software service at the client’s remote-storage location linking the appropriate imaging dataset and streaming the information in real time, while the pathologist is providing the consultation. A sophisticated algorithm only transmits the information that fits in the pathologist’s screen, therefore limiting the data to be transmitted and making quick consultation turnaround possible. Other challenges in digital pathology are the lack of established imaging standards and the fact that proprietary imaging formats from the largest vendors are in continuous evolution. The solution used by UPMC leverages an open-source, vendor-neutral viewer that is compatible with the most common pathology-imaging formats available in the market. In September 2011, the KingMed telepathology Web portal went live for KingMed Diagnostics, providing consultation services to KingMed’s customers in China. UPMC pathologists have the ability to view whole-slide images hosted on an imaging server at KingMed’s facility. The Web portal viewer affords the pathologists the ability to manipulate, annotate, and take snapshots of areas of interest on the whole-slide image; these actions can also be included in the final report. The Radiology Protocal Tool and Recorder (RAPTOR) System Medverd, a radiologist on staff at Washington’s VA Puget Sound Health Care System, Seattle Division, also holds a faculty appointment at the University of Washington. He has long held the belief that making protocols for advanced imaging exams is undervalued in private, public, and university settings. “The process is not optimized,” he says. “You get a piece of paper with one or two lines on it providing the clinical provider’s problem and questions to be answered; then, when one wants more information, it’s often time consuming and cumbersome. If you talk to any radiologist who has protocol responsibility for cross-sectional imaging, he or she will tell you there’s a constant battle between efficiency and effectiveness for that task.” Using funding from the VA Innovations Initiative and leveraging VA IT resources, Medverd designed a prototype for filling those gaps with information from the EMR, with an extensible design that could be rolled out nationally. Because the VA has a legacy health IT architecture with a vast repository of health information, Medverd approached the project with the intention of designing the application in layers. He planned to use Web services, for example, to virtualize the electronic health record, so that a Web application (as opposed to software that needs to be installed on every user’s computer) could be used. “With Web services, all we need to do is build a sort of data-adapter layer into the content-management system for the presentation of the data,” he explains. A happy discovery was the availability of the VA’s Medical Domain Web Services (MDWS), which Medverd and his team used to virtualize the health records. “Frankly, I was not aware of it when I first submitted the idea, and I thought we’d have to build it ourselves,” he says. “The discovery of MDWS was great because somebody else had already done the work, and that’s the advantage of working in layers. That MDWS layer provides the interactivity with the legacy archives that we would have had to build, if it weren’t there.” For a Veterans Integrated Service Network (VISN) to implement RAPTOR, the VISN’s protocol library would be uploaded to the RAPTOR server. Through the uploading process, the VISN would also have the opportunity to cross-link the protocol library with commonly accepted naming conventions in the RSNA’s RadLex. Medverd’s goal of improving efficiency and patient safety throughout the VA system appears within reach. “Given the amount of enthusiasm folks have had, I’m very optimistic that we’re going to move forward,” he says. Problem/Objective The paper-based workflow predominantly used to create protocols for advanced medical imaging at Veterans Health Administration (VHA) facilities is subject to numerous process errors. The RAPTOR system leverages the VHA’s EMR and open-source content-management frameworks to provide an efficient Web environment, with decision support for contrast risk assessment and protocol assignment. Solution The RAPTOR system extracts relevant information for each patient from the EMR and displays it next to the imaging requisition. The Web interface provides access from a variety of systems and includes features to sort the worklist, flag relevant allergy history and renal-function tests, suggest relevant department-approved imaging protocols, suggest standardized pre- and postexam hydration, and suggest premedication for those with a history of contrast reactions. This offers a significant advantage over the prior system by ensuring legibility, standardization, prioritization, multiuser access, and improved patient safety. Additional features of RAPTOR will include secure messaging, restricted ordering access for specialized studies, recognition of order duplication, and logging of physicians’ and staff members’ input into the protocol decision-making process. While this solution will initially be deployed as a pilot at selected VHA facilities, the goal will be deployment across the entire VHA enterprise. Results A review of the current paper-based protocol workflow at one VHA facility evaluated 341 MRI orders over the course of a month, of which 61% were for neuroradiology, 12% were for musculoskeletal imaging, and 6% were for body imaging. The average paper protocol required an elapsed time of 11 days from the time that the study was ordered to the day that the patient was successfully contacted to schedule the exam. It was found that approximately 15% of exams for which protocols had been completed were never performed; for 1%, orders were duplicated but both had protocols prepared, and for 2.5%, protocols were unsigned. Rare (but observed) clerical errors, such as mismatched patient information, further corrupted this system. RAPTOR prototype testing suggests significant process improvement due to real-time data-query capabilities. Unproductive and redundant protocol-making efforts are minimized, the speed of the protocol process is increased due to prioritization and distribution of work within a multiuser-accessible electronic work list, fulfillment of enterprise quality and safety goals is improved due to automated identification and flagging of patients at risk for harm from the performance of advanced medical imaging, and ambiguity in medical-decision responsibility is eliminated through the capture of documentation logs and electronic signatures. Cheryl Proval is editor, Radiology Business Journal. Kris Kyes, technical editor, and Thanh Le, editorial coordinator, Radiology Business Journal, contributed to this article.