AI can help radiology standardize image exam data labeling

The case for standardization in radiology terminology

Radiology practices are inundated with data that lacks consistency in terminology and labeling. This variability complicates communication between systems, hinders efficient data mining, and poses challenges for integrating AI into workflows. Consistent labels are important for  when AI is being tasked with pulling prior exams for a patient or knowing when to apply an automated algorithm to specific types of exams. Data consistency also aids billing and can help enable AI workflow orchestration.

For example, Rankin said a CT scan might be labeled differently across institutions, or even within the same organization, using terms like "chest CT," "thorax CT," or "lung scan." Such discrepancies make it challenging to locate specific exams, share data across platforms, and implement AI solutions effectively.

How AI streamlines operations

Enlitic addresses this challenge by leveraging AI to standardize data labels and descriptions, ensuring that information is consistent and easily accessible.

"Our system analyzes images and metadata, identifies patterns and translates the data into a standardized format tailored to the user’s needs. Whether you’re dealing with study descriptions, contrast usage or modality names, AI removes the variability, making the data more predictable and usable," Rankin said.

"The data comes through us, all the data is standardized and then it flows through to those other systems. The configuration becomes very simple, very manageable at that point so users can pick out the data for a particular algorithm and it becomes really easy," he explained.

This standardization simplifies processes like pre-fetching exams for AI algorithms. For instance, when a hospital uses AI to detect strokes or lung lesions, the system needs to identify and retrieve the relevant studies. Without consistent labeling, such tasks become burdensome and prone to error.

"We make that problem go away. So whether you're trying to teach your AI inside of a hospital research organization, whether you're trying to figure out how to do case allocation as a radiology business, launching the right report template, or doing hanging protocols, there are many use cases that become so much easier if the data's consistent and predictable," Rankin explained.

Broader implications for data mining, billing and training AI

Beyond streamlining workflows, data standardization supports initiatives like data mining for population health studies, billing accuracy, and AI model training. Rankin notes that inconsistent data can lead to underbilling when critical information, such as contrast usage, is overlooked in reports. AI can watch for specific types of exams to check all the contrast was billed correctly, but if the studies are not labeled the same, this becomes an issue.

For hospitals exploring AI applications, standardized data also enables retrospective analysis and quality control. Researchers can efficiently locate exams matching specific criteria, such as imaging studies for diabetic patients, facilitating AI training and validation.

Healthcare data has become a valuable commodity in recent years as organizations recognize the value of this information for research and AI algorithm training. Some hospitals also are monetizing their data, but that requires anonymization to remove patient identifiers, which can be a daunting task.

Looking ahead, Rankin envisions a future where standardized data serves as the foundation for transformative applications in radiology. "Hospitals are starting to see that retaining imaging data can yield long-term benefits, from training diverse AI models to improving patient outcomes. But to realize this potential, data standardization is crucial."