Solutions
By Gary Taubes
Seeking in tongues
Radiology online
Seeking in tongues
How a so-so document translator became the U.N. of search engines.
by Gary Taubes
New worlds of information could soon open up
to Web users, thanks to a search engine that can break through language barriers. Imagine typing a query in English and then retrieving relevant documents translated from French, German and Italian. That's the intent of the "cross-language information retrieval program" being developed by Salim Roukos and his colleagues at the conversational systems group at IBM's Thomas J. Watson Research Center.
The retrieval program is based on an effort started a decade ago to translate documents with statistical algorithms. The IBM team gathered a huge body of bilingual text from the Canadian Parliament -- 100 million words of translated English and French sentences. Then they designed a program that figures out which sentences are equivalent in the two languages and uses probability to determine which words correspond.
After five years of work, says Roukos, the system translated text fairly well -- though not as accurately as some other approaches. What it did have going for it was speed. So the researchers decided to focus on a use where smooth translation was of secondary importance and speed was of the essence: retrieving information. They added German and Italian automatically by using multilingual articles that, while not direct translations, discussed the same event in different languages.
The statistical translation algorithms are combined with a search engine to produce what Roukos calls a translingual search, which extracts relevant documents (currently from a fixed set of texts stored in a database) in any of the four languages and presents them in English. Fast. "We can translate 10 megabytes in an hour or so," says Roukos. Last November, an assessment by the National Institute of Standards and Technology ranked the IBM translingual search engine the best among nine systems tested.
Although the system is still in prototype, Roukos and his collaborators are looking for ways to extend the technology to new applications. "One area we're considering is topic detection," Roukos says. "If a story is reported on a topic you're interested in -- no matter what the language -- it can be retrieved and sent to you immediately. That's the long-term hope."
Gary Taubes, a freelance writer who lives in Venice, California, is a frequent contributor to Think Research.
Radiology online
A Tel Aviv hospital moves toward "filmless" medicine.
By Chani Sacharen
One of the largest hospitals in Israel
is about to complete a technological coup that many medical institutions have dreamed of over the years. Collaborating
with a team of scientists at IBM's Haifa Research Laboratory, the Chaim Sheba Medical Center in Tel Aviv will soon be a
"filmless hospital," in which doctors view all medical images as digital files on workstations instead of reading film
on lightboards. With Picture Archiving and Communication Systems (PACS), as this technology is known, images can be stored,
enhanced, annotated and viewed simultaneously on multiple viewing stations.
has been gradually catching on over the past decade. Uri Shani, who leads the IBM team working with Sheba, points out
that the technology received a strong boost in the mid-1990s, with the widespread adoption of a standard known as Digital
Imaging and Communications in Medicine (DICOM) version 3.0, which allows hospitals to integrate all imaging modalities
with image archives and to share patient information in a uniform fashion.
The Haifa team has installed its Integrated Digital Medical Records (IDMR) technology in Sheba's radiology department
and emergency room. In contrast to other PACS solutions, the IDMR system lets doctors view images directly
from the central repository, rather than having to download the images to their individual workstations. At the same time,
the system's architecture is flexible enough to allow different departments to maintain their own policies for managing
the hardware that stores their data. The hospital will soon complete installation of 40 additional viewing stations,
extending IDMR throughout the hospital, from inpatient wards to outpatient clinics. "The changes have been dramatic,"
says Uri Gabbay, the hospital's deputy director and chief information officer. "With the old system, it was next to
impossible to track down the film of an X-ray taken during the previous 24 hours because it hadn't yet been filed.
Now, using the IDMR system, physicians can instantly bring up any archived image, whether it was taken 10 hours or
10 months ago."
Because of the critical nature of the information, IDMR stores the data on fail-safe disks that are managed by
high-availability RS/6000® multiprocessor servers. The data is then backed up on tapes for long-term
archiving. Files of patients scheduled for an examination are automatically moved from the tape archive to online disk
the night before. A list of the patients to be seen, including all their previous imaging studies done at the hospital,
is automatically presented to the physicians when they log on to their IDMR workstation in the morning. The entire system
of tapes, archives and workstations can be accessed, monitored and managed through an Internet browser interface.
"Less film" is the term Yaacov Itzchak, director of the department of diagnostic imaging, uses to describe the current
situation. "With the growing demand to install workstations all over the hospital, our dream of transforming the department
into a filmless one is coming true faster than we ever imagined," he says.
The success of the system has led Sheba's radiology department to begin purchasing digital mammography machines.
Equally significant, the system is playing a complementary role in entirely new approaches to surgery. By combining
computer-aided tomography, magnetic resonance imaging and angiogram images stored in the IDMR archive, surgeons can
plan complex neurosurgery using a 3-D visualization of the patient's brain. The actual operations are performed inside
special open MRI machines, providing surgeons with real-time visual guidance.
Buoyed by the results at Sheba, Shani expects the PACS technology to find use in broader networks. The group that is
responsible for IDMR has also pioneered technologies for real-time streaming of voice, video and data over intranets
and the Internet. "Combining new methods for collaboration on medical images with technologies for telephony and
real-time streaming would allow us to extend the IDMR architecture to encompass all the hospitals in an urban or
larger geographic area," says Shani. Widely adopted, PACS combined with such "teleradiology" would be a boon to
sparsely populated areas that lack the skilled personnel to interpret radiological images. Doctors at widely separated
locations could view a digital image together, enabling major medical centers to share their diagnostic expertise. Already,
using advanced communication algorithms that they developed, the
Haifa researchers have shown that large files can
be collaboratively manipulated even over ordinary telephone lines, thereby enabling joint consultations in real time.
But the possibilities do not stop there. "With digital images and medical records," says Shani, "we can begin to think of
using pattern recognition to help interpret radiographs, and data mining to look for disease trends in large populations.
The age of computer-aided medicine is really just beginning."
Chani Sacharen is a technical writer who lives in Kfar Sava, Israel.
|
