In a study of nearly 40,000 real-world examinations, the system achieved an average area under the curve (AUC)—a metric where 1.0 represents perfect diagnostic accuracy—of 0.913.
The release of this model addresses the need for specialized tools that can support overtaxed healthcare professionals in complex diagnostic tasks.
According to Damo Academy, the model’s accuracy exceeded that of 23 out of 26 human radiologists in a comparative study.
When using the AI as a secondary tool, these radiologists reduced their time spent on scans by more than 30% and improved their ability to prevent missed diagnoses by 10%.
The open-sourcing of RADAR marks a shift toward generalist medical imaging models, as the training methodology could potentially be applied to other types of diagnostic images beyond the abdomen.
This development follows Alibaba’s previous healthcare AI efforts, including tools for spotting colorectal and pancreatic cancers.
While currently focused on 146 specific clinical findings, the research team aims for the technology to serve as an expert-level assistant to help clinicians manage high volumes of medical data more efficiently.