Last updated: February 22, 2010

G. R. Thoma
August 18-19, 1994.

Application goals in medical information

The NLM and several thousand medical libraries, covering a broad spectrum ranging from major academic health science center libraries to small rural hospital libraries, constitute a large community entrusted with the mission of providing timely and accurate information to practitioners and researchers in biomedicine. Such information can take the form of bibliographic and factual databases, digital images of documents and medical x-rays, video sequences of medical procedures, digitized color images of dermatologic conditions, procedural rules found in expert systems, output of medical devices such as ultrasound, EKG, EMG and many others. These information types reside in many types of storage systems and some of them, mainly bibliographic databases, are being routinely accessed via telephone modem and value added public networks (Telenet, Tymnet).

As a consequence of several trends, professionals (end users) are looking toward the easy and rapid access to these diverse resources. These trends include: (a) the rapidly increasing global connectivity of the NREN/Internet; (b) the increasing availability of public domain software, both server (e.g., WWW, Gopher) and client (e.g., Mosaic); and (c) increasing power on the desktop, making true multimedia workstations possible.

Projects that address some of these application goals are being done at the NLM's R&D center, the Lister Hill National Center for Biomedical Communications, as at many other universities, libraries, medical centers and other institutions. The projects at NLM deal with document imaging, xray imaging, 3D visualization of anatomic data and others. Document imaging projects include SAIL, DocView, WILL among others. SAIL (System for Automated Interlibrary Loan) is a multi-workstation system connected via a token ring LAN. It accesses the NLM's mainframe computer for interlibrary loan (ILL) requests, parses the requests into document information and user information, passes the document descriptor to a store of about 200,000 already scanned journal page images on optical media, extracts the requested images, and uses an internal fax board along with the user's fax number (previously extracted from the ILL request) to fax the images out to the user's fax machine. The SAIL pilot system is designed to reduce or eliminate the manual activities that are routinely required to fill interlibrary loan requests.

In contrast to SAIL which relies on indirect access to images through the nationwide medical interlibrary loan service, DocView is a direct access system, anticipating the eventual availability of document image databases. It consists of client software running under MS Windows 3.1 that accesses images stored in servers, both inhouse-developed as well as public domain (such as WWW, Gopher, FTP), and retrieves the images over the Internet. DocView enables an end user to preview, display, manipulate, electronically bookmark, segment, cut and paste, and print the document received. DocView has the full capability of communications, document display and manipulation, but it can also use the WWW-Mosaic communications protocol to transfer the images over the Internet, and serve as the necessary viewer or browser. In addition, it can receive documents sent over the Internet by Ariel workstations, a product of the Research Libraries Group, now used in hundreds of libraries worldwide. Ariel serves a need in the document delivery process arising from the simple fact that the vast majority of printed documents (journals, books, reports) are not on servers now and probably are unlikely ever to be. So, paper will have to be scanned and transmitted, as in conventional fax. Unlike conventional fax, however, Ariel uses the Internet as the transmission link, and is therefore limited neither by the narrow bandwidth of the public telephone system nor by the 200 dpi resolution of conventional fax machines.

Since most documents requested by users of their libraries will be in paper form for the foreseeable future, we are also designing a multimodal transmission system that limits the manual activity of an operator to simply scanning a document. This system, called WILL, automatically retrieves ILL requests from NLM's mainframe computer, parses the requests, recognizes the recipient's fax number, mailing address or Internet IP address, and automatically sends the scanned images to the right place by the method requested by the user. Beyond scanning, no other burden is placed on the operator.

An x-ray imaging project, DXPNET, is motivated by the 17,000 x-rays taken as part of the nationwide NHANES II survey and 10,000 x-rays taken in the NHANES III survey (NHANES: National Health and Nutrition Examinations Survey). These xray films (cervical and lumbar spine; hands and knees) need to be preserved, and furthermore need to be read by radiologists. We are scanning and archiving these in an optical disk jukebox, and building the server and client workstations to allow access to the images over the Internet. The user (radiologist) will be able to access and retrieve the images, and enter numeric readings and free text comments in a form on the screen. These readings and comments will then go to the jukebox where they will be stored in a database for access by other interested parties, e.g., epidemiologists, demographers, insurance companies, and government agencies.

Motion video is also of interest to the biomedical user. The NLM and other medical libraries have repositories of videotapes of surgical procedures, grand rounds, ultrasound examinations, educational seminars and others. The current procedure is for a user to search index databases that provide information about the tapes, such as title, brief textual abstract, general topic, creator, institution, etc. The artifact itself, when requested, has to be packaged and mailed out to the requester. A project is under way in which the tapes are being digitized and compressed via MPEG, stored on optical disks, and accessed automatically by a link from a database search.

Signal processing and other tools

Among the diverse technologies required to make the medical information resources mentioned earlier accessible are those having to do with efficient storage, improved image quality, application- specific man-machine interfaces, pattern recognition for feature extraction, and efficient transmission.

In the research agenda for document imaging, the following areas are of interest: compression of two-tone images (JBIG); OCR and image enhancement preprocessing to make it more reliable such as automated border removal, skew detection and page orientation detection; user interfaces allowing easy manipulation of the images; efficient page description and markup languages; scaling and scale-to-gray algorithms for image quality improvement.

In the document imaging projects, a key element is the scanning that is needed to create a suitable image store. In the process of scanning, signal processing stages are: automated border (page edge effects) detection and removal, automated page orientation detection, automated skew detection. All of these not only aid the capture of the document images with high visual quality level, they also provide the quality needed for subsequent processing such as accurate optical character recognition. Also, in the display of the document images, other useful processing includes: scaling (from the scanned 300 dpi image to VGA or SVGA resolution), and scale-to- gray (to shade the image in suitable gray levels rather than black or white to increase readability). Ordinary scaling replaces groups of pixels by a single pixel, either black or white, while scale-to- gray replaces them with pixels of intermediate gray values. Ordinary scaling allows fast compression via runlength coding but the characters could appear broken and readability could be impaired. Scale-to-gray increases readability by shading the character edges, but image compression is slow. Document compression via JBIG has been shown to be more efficient than CCITT G4 particularly for dithered images, and appears to warrant further study (ref: Bakalidis, et al; W. Equitz).

In the case of digitized x-rays and color originals, the following areas are of interest: compression of multiple gray level images; design and optimum selection of quantization tables for the DCT process within JPEG; automated and user-directed segmentation based on pattern recognition for feature extraction by mathematical morphology techniques; noise removal by morphologic operators; contrast enhancement by histogram equalization; artificial neural networks for image classification.

In the case of motion video sequences: compression by MPEG or motion JPEG; link up of a citation received from a bibliographic database to a video sequence or some part of a sequence. Of particular interest to the ongoing development of the NII is that transmission over the Internet of digitized images is relatively slow, especially for large files such as those resulting from the scanning of x-rays. In our project, cervical and lumbar x-rays when scanned result in files approximately 5 and 10 MB respectively. Since lossy compression requires further study as to whether or not important information is lost or if unwanted artifacts are acquired as a result of the compression technique itself, lossless compression methods are the only acceptable ones though these seldom deliver a compression ratio of better than 2. To overcome this limitation, multisocket transmission methods are being explored. These methods are based on the observation that during a conventional single socket linkup the transaction overhead for TCP/IP limits the effective rate at which data can be transferred. However, if the image to be transmitted were to be sectioned, each section assigned a pair of sockets (at the transmission and receiving ends), and the sections transmitted and reconstructed at the other end, the effective transfer rate could be higher. Preliminary tests have shown a three-fold increase in transmission speed for about 20 socket pairs. The optimum number of socket pairs need to be determined, however, since increasing this number indefinitely is not fruitful because the creation and maintenance of the sockets requires computing resources.

Another element in access is the increasingly distributed nature of medical information at many points on the Internet. Also, it is the case that there is a trend toward multimedia workstations that enable a user to view, manipulate and seamlessly move among different data types on the desktop. Both these trends suggest intelligent gateways to the appropriate information sources, the extraction of the query-driven data, and reconstruction of the data at the client workstation for presentation to the user for viewing and to possibly do further processing and analysis.

Avenues for collaboration

There are many ways for independent researchers at universities and research centers to have their work supported. Several governmental agencies (NSF, ARPA, NIH, NLM, NIST, NTIA, etc.) issue extramural research grants and contracts through various mechanisms: the extramural grants activities and broad agency announcements. In addition, individual investigators can be supported through consultancies by purchase order, intergovernmental personnel act (IPA) appointments, expert consultants, visiting scientists, etc. A more recent approach is the Cooperative Research and Development Agreement (CRADA) which is a means to transfer technology from federal labs to the private sector, which allows a private group to take prototype designs from the labs and develop them into commercial products, and share any ensuing royalties with the government.