As big data grows, the need for AI comes into focus
No one questions that the era of big data is here, but Dr. Anthony Chang warns that the deluge of medical information is just beginning.
"By 2020, there will be 200 times more data than any physician can absorb," said Chang, a practicing pediatric cardiologist. "And its doubling every two years."
In his keynote address at the National Healthcare Innovation Summit in Chicago Wednesday morning, Chang said he worries that lives are being lost from the unrealized opportunity.
[See also: Watson has big plans for the future]
"There is nothing worse than when a child is lost because we don't have access to the right knowledge," he said.
The longterm solution Chang is working toward is "Intelligence-as-a-Service," a network that could make it possible for doctors to tap into knowledge from specialists anywhere when they encounter a medical situation that is not responding to treatment.
"Why can't we get access to that intelligence?" he asked.
Today, one of the problems is that the data being collected is unstructured; Chang estimates that about 90 percent of healthcare data being collected is not in a structured format. Until entities like IBM Watson came along, most of the bio- data being collected can't be integrated with data in existing analytic systems.
"And in the future there will be tsunamis of data – particularly genomic data and yet another layer of behavioral data produced by wearables," he said. "And with all that data, there's very little intelligence coming our way. "
The future will require a different approach, in which data is used in a BioIntelligence framework similar to deep learning, said Chang. Multiple layers of analytics are used to extract value from data.
He cited the work being done by Excel Medical Electronics with the BedMasterEx data acquisition solution working with IBM' Watson's InfoSphere Streams technology. He described the solution as four-stage architecture in which data is acquired in a SQL format, moves to an adaptation layer, then to an analytical layer and finally a delivery layer using HTTP. InfoSphere Streams is a sensory interface for Watson, making it possible for unstructured bio data to be analyzed.
Chang then introduced Robert Merkel, vice president of client engagement at IBM Watson Health, who described Watson as a "cognitive system" which is taught, not programmed. It can learn, and improve its performance based on its experiences and it can work with sensory and non-traditional data.
The challenge is substantial. Merkel said the amount of data for a single individual is 0.4 terabytes of medical records, 6 terabytes of genomic data and a crushing 1,100 terabytes of exogenous data (ie., behavioral and environmental readings).
Ignoring all of that data has a price.
Merkel said estimates indicate that "20 percent of what determines a person's health can be found in their clinical data, 20 percent can be found in clinical data and 60 percent in the exogenous data."
Clearly, no hospital will be able to store or analyze that much data. Shared solutions will be needed.
Merkel said IBM's Watson Health Cloud will be an open platform to aggregate the advanced analytics that the next generation of medical data will require. It will be a secure, HIPAA-compliant platform, that is massively scalable. The goal is to enable collective intelligence, taking patient data and applying evidence-based insights in an outcomes driving learning system.
"Imagine that instead of getting a simple reminder that its time to see your physician, you received a message based on real intelligence on your own medical data," Merkel said.
"We hear a lot about physicians being demoralized by computers," Chang said, "but the most exciting times are coming. The best 25 years in medicine are coming up."