RSNA 2018: 4 ways radiology departments are preparing for big data

Artificial intelligence (AI) and big data can help radiologists provide better care while reducing costs, but a majority of institutions lack the infrastructure to optimally consume and utilize these technologies.

“In reality, when it comes to IT stacks and advanced IT infrastructure our [healthcare] future is everyone else’s past,” said Paul Chang, MD, medical director of enterprise imaging at the University of Chicago, during a morning session at RSNA 2018.

For Chang, the “fundamental issue” at play is the fact that radiologists and physicians had it easy with fee-for-service. But as radiology shifts toward value-based care and contracts involve more shared-risk, radiologists will need data-driven strategies to keep up.

There are four general ways institutions are working to extract relevant data from clinical systems, according to Chang. They are:

1. Reliance on existing systems

This is commonplace for many radiology departments, Chang noted. An administrator or quality team member will “by hook or by crook” figure out a way to generate a report, typically from a vendor.

It’s not a bad method, but a limited one, Chang said. Reports are available, affordable and require no cultural IT change. The negatives, however, outweigh the positives. Data is typically retrospective and vendors don’t always make generating reports a top concern.

2. Data Warehouses

Using this approach, clinicians can correlate such things as utilization with outcomes and perform economic analysis. As long as all the information is housed, data can be correlated across multiple sources, Chang noted, and providers aren’t reliant on vendors for information.

However, the change requires buy-in from leaders, cultural change and a governance model. Typically data warehouses can’t provide real-time data. Perhaps the biggest disadvantage, Chang said, is reliance on native methods to extract data—creating further interoperability problems.

3. Edge device approaches

If done correctly, Chang said, real-time feeds are accessible, including real-time lab reports fed directly into a PACS system—an ideal platform for AI.

This approach can be disruptive, requiring “massive” governance changes in the IT department. Additionally, clinicians remain dependent on vendors, “and I don’t like to be dependent on the vendor” Chang said.

4. Service-oriented architecture (SOA)

The University of Chicago adopted this method 10 years ago, according to Chang. It’s used by most businesses, from pizza shops to insurance companies, he noted. It’s foundation is built independent of vendors, products and technologies.

“This is how every other industry does it,” he said.

Advantages include real-time dashboards and AI. The disadvantages, which Chang called “significant,” include dependence on in-house extract, transform, load (ETL) processes. Significant IT support and software development resources are also required, along with cultural buy-in.

In the long run, Chang said his institution was able to save money by adopting SOA. It will require buy-in from many stakeholders, but ultimately help radiology transition into the value-based era.

“We’re [radiologists] no longer going to be valued just for interpreting images, we’re going to be valued for managing the role of imaging in a very complex environment,” Chang said. “We have to be irreplaceable and self-promote.”