IBM is moving from helping hospitals run smoothly to lending a hand in the examination room.
International Business Machines Corp.’s planned acquisition of Merge Healthcare Inc., which sells systems that help doctors store and access medical images, is a crucial step in its plan to put artificial intelligence to use in medicine. IBM announced the $700 million deal last week.
Merge’s crown jewels are 30 billion images, including X-rays, computerized tomography, and magnetic-resonance-imaging scans, that IBM intends to use to “train” its Watson software to identify ailments such as cancer and heart disease. The resulting services, it hopes, will help doctors diagnose and treat patients more effectively and efficiently.
Google Inc., Yahoo Inc. ’s Flickr unit and others use similar software to identify objects in photos. IBM is betting that the same technology that recognizes cats can identify tumors and other signs of diseases. The nascent effort could help IBM capture a larger slice of the $7.2 trillion spent world-wide annually on health care.
IBM’s deal also could reshape the $3 billion market for archiving medical images and breathe life into companies devoted to computer-driven interpretation of images. It highlights the value of such imagery, which typically is anonymized and shared by hospitals for research purposes, as the software technique known as deep learning becomes more prevalent in medicine.
Deep learning, a technique in which software learns to identify patterns by sifting through large amounts of data, has proved successful at interpreting photographs, improving voice recognition in smartphones and detecting fraud in financial transactions.
Identifying a tumor involves similar pattern-recognition techniques, according to John Kelly, senior vice president of IBM’s solutions portfolio and research.
In the long run, IBM and others in the field hope such systems can become reliable advisers to radiologists, dermatologists and other practitioners who analyze images—especially in parts of the world where health-care providers are scarce. But medical scans of the human body are complex.
“In medical data, there’s lots of ambiguity and lots of fuzziness,” said John Eng, an associate professor of radiology at Johns Hopkins University. “It’s kind of messy data, and I think that’s going to be a limiting factor with what IBM does with Watson.”
The effort is in a very early stage. Johns Hopkins is using computers to help doctors identify tumors in chest scans and mammograms, but the benefits of existing software tools are limited and radiologists still largely work without computerized assistance.
“It’s a way off to have a general diagnostic machine,” said Dr. Eng.
IBM has made health care a strategic target of its Watson platform, an assortment of artificial-intelligence software at the heart of several of the company’s business initiatives.
It aims to pioneer a new category of software products that mine data for medical insights, which it could sell to large health-care organizations where it already has relationships.
To that end, it has been testing Watson with researchers at Cleveland Clinic and New York’s Memorial Sloan Kettering Cancer Center. Acquiring Merge brings not only access to the company’s image archive but also its account list of 7,500 hospitals.
“The access to these clinical relationships and the data is valuable,” said Kevin Hobert, the CEO of Carestream Health Inc., a Merge competitor.
While IBM hopes Watson will learn to interpret Merge’s images, it also expects the combination of imagery, medical records and other data to reveal patterns relevant to diagnosis and treatment that a human physician may miss, ushering in an era of computer-assisted care. Two other recent IBM acquisitions, Phytel Inc. and Explorys Inc., yielded 50 million electronic medical records.
Smaller companies devoted to automated interpretation of medical imagery felt the impact of IBM’s arrival immediately.
Enlitic Inc., a San Francisco-based startup with $5 million in angel and seed funding, claims that its software identified malignant lung tumors in X-rays 50% more accurately than a panel of four radiologists.
The company’s CEO, Jeremy Howard, spent the past year approaching hospitals and clinics for access to anonymized images that would demonstrate his system’s utility.
After IBM’s announcement on Thursday, his email inbox brimmed with messages from hospitals offering images, hospital networks interested in trying his technology, and radiology services looking to improve their own accuracy, he said.
Training such software requires huge numbers of images. Merge’s 30 billion MRIs are only a start for IBM. “The way these machine-learning engines work, the more you feed them the smarter they get,” IBM’s Mr. Kelly said.
But such imagery remains limited by regulations and industry reluctance to share what may be sensitive personal information. Enlitic’s Mr. Howard has been approaching hospitals and clinics one by one to persuade them to share their archives.
“People are starting to realize these archives have value,” Mr. Howard said. “Before, these images were sitting around for 25 years gathering dust.”