Imaging Precision Tool Enables Method Comparison Without Ground-Truth Data

Quantitative measurements from medical images increasingly guide diagnosis and treatment, yet their reliability can be difficult to verify in routine care. In many cases, the true value of what is being measured, such as tumor size or uptake characteristics, is unknown, making it difficult to determine which imaging method is most precise. This uncertainty can affect both clinical decisions and evaluation of new technologies. Researchers have now developed a technique that ranks the precision of imaging methods without requiring a gold standard.

Washington University in St. Louis has introduced NGSE-Corr, a mathematical framework that allows researchers and clinicians to assess how precisely different quantitative imaging methods perform when the ground truth is unavailable. The approach is positioned to support developers of artificial intelligence (AI) tools as well as regulators who evaluate such technologies for clinical use. It provides a structured way to compare methods using clinical-type data without needing definitive reference measurements.

The method builds on prior quantitative imaging theory by explicitly accounting for correlated noise across tools that measure the same lesion or biologic property. In practice, measurements from different methods applied to the same tumor are not independent, and NGSE-Corr incorporates these correlations into its ranking process. The team performed numerical experiments demonstrating that the technique can objectively order methods by precision under realistic clinical constraints.

The researchers then conducted a virtual imaging trial to rank three quantitative single photon emission computed tomography (SPECT) methods for measuring regional activity uptake in computer-generated patients with bone metastatic castrate-resistant prostate cancer treated with radium-223. The goal was to determine which method would be most appropriate for that clinical task. The virtual trial enabled controlled comparison while reflecting clinically relevant variability.

Across groups of 50 virtual patients, the methodology accurately ranked the imaging methods in 91% of trials and identified the most precise method in 95% of trials, with performance improving as virtual cohort size increased. The work was published in IEEE Transactions on Medical Imaging. Collaborators were from the Mallinckrodt Institute of Radiology (MIR), Siteman Cancer Center based at Barnes-Jewish Hospital and WashU Medicine, and the Department of Surgery at WashU Medicine.

“Medical images are increasingly being used not only for visual interpretation, but also to derive quantitative measurements of clinically relevant properties. We want these measurements to be precise. Otherwise, they can impact the clinical decisions that are being made based on them,” said Abhinav Jha, Ph.D., associate professor of biomedical engineering at the McKelvey School of Engineering and of radiology at WashU Medicine Mallinckrodt Institute of Radiology (MIR).

“This tool has a lot of potential applications for different members of the community. It could be useful for researchers who are developing new quantitative imaging methods, including AI-based methodologies; physicians who are looking at using different tools to make decisions; and regulators who are interested in evaluating technologies,” said Jha.

Related Links
WashU McKelvey School of Engineering 
WashU Medicine Mallinckrodt Institute of Radiology