Advanced Data Mining Tools Help in Search, Visualization

CHICAGO - Imagine, while you’re reading a study, being able to access records and images on similar findings and diagnoses - in seconds. Today, it’s a hunt that could take months, but with advanced data mining tools, experts said, this kind of information could be as easy as a Google search.

“When you can index two databases, you can do cross database searching, and instead of taking three months to do this, it takes seconds,” said Woojin Kim, MD, associate director of imaging informatics at the Hospital of the University of Pennsylvania, speaking at an RSNA session Thursday.

Data mining tools can help radiologists extract and convert data into knowledge. And it can be as simple as a familiar search engine, said Kim, who developed a system for his department. Right now, however, radiologists live in a world similar to one without Google. Can you imagine? Yet employing data mining in medicine is a challenge, where the format of the data and the speed of retrieval are paramount.

His department’s system, PRESTO, the Pathology-Radiology Enterprise Search Tool, a Web-based search tool that queries both radiology reports and pathology reports, optimizing and indexing the data. Users can query the system, filter the data, and access more than 13 million radiology reports and 270,000 pathology reports, Kim said. And that’s just the beginning of its capabilities, as it can also provide dashboards and analysis for retrospective views of department activity.

Radiologist at Massachusetts General Hospital are developing a similar data mining tool that takes report access to the next level - data visualization - integrating findings with patient records and presenting it in an easy-to-view format. NETRA, which stands for Navigational Enterprise Tool for Representing Radiology reports, allows the radiologists to see how many exams a patient has had for a chest nodule, for example, presented in a timeline organized by body parts with symbols for the various imaging modalities, said Supriya Gupta, MD, of Mass General’s informatics department, speaking during the session.

Rather than searching each radiology report in the database, a radiologist can see it all in one place automatically, she said. Clicking on a report then communicates the findings on a simulated human model in the application.

“This application supplements the report,” she explained,” Instead of browsing through, I have a 3-De model where findings are laid out. Instead of going back and forth in reports, I can get a visual display of the findings,” including how the nodule in this case might have changed over time.

Now, the application is being trained to identify cumulative dose of the patient, and currently includes CT dose, Gupta said. This information is also plotted on a timeline, and can be used to schedule the next exam.

“A picture is worth a thousand words of text reports,” she said, “and advanced data mining tools help us achieve this."