WASHINGTON, D.C.—Although quantitative imaging is radiology’s new frontier, it is not actively used now because it’s complicated. Informatics can help radiology move forward with quantitative imaging, said Daniel L. Rubin, MD, assistant professor of radiology Stanford University School of Medicine in Stanford, Calif., last week during the annual meeting of the Society for Imaging Informatics in Medicine (SIIM).
“Quantitative imaging is all about processing image metadata,” Rubin explained. Metadata include radiologist observations, machine observations and image markups. In the current workflow, physicians use paper data recording forms, verbal communication and text reports to share data. However, none of these elements are currently machine-accessible.
“We lack a standard for representing and communicating image metadata. This results in the inability to exchange and access image metadata as easily as DICOM images,” said Rubin. “The good news is the technology is here, but our informatics infrastructure doesn’t enable it.”
The cancer Biomedical Informatics Grid (caBIG) has developed the Annotation and Image Markup (AIM) schema, which provides a standard for imaging metadata. Image annotation tools can save quantitative data in the AIM format.
At that point, computers can process the AIM annotations to enable applications for radiologists and clinicians such as clinical decision support. “This provides an objective interpretation and reduces variability,” stated Rubin.
The goals of the project include making all image metadata machine-accessible and linking image measurements to the image source.
The caBIG approach employs an open source framework to standardize quantitative imaging. Current PACS workstations support the AIM format, which makes image content explicit and ultimately yields a structured report with controlled terminology. With controlled terminology, the computer can process and store coordinates, internal pixels, observations, calculations and anatomic entities.
“We need annotation tools to operationalize AIM workflow,” said Rubin. For example, an operationalized workflow would automatically generate flow sheets and response graphs.
Early results are impressive with studies showing drastic improvement in data collection, offered Rubin.
Rubin concluded with a vision for future—a new automated workflow for radiologists and a robust platform for data mining. “By having measurements available we can provide a dashboard showing which lesions were measured on prior studies, automatically generate patient response graphs and provide cohort response [evaluations] based on biomarkers.”