Nashville/Hong Kong (LabNews Media LLC) – A new AI framework aims to make artificial intelligence in digital pathology more reliable. The system, named TRUECAM, recognizes uncertainties in cancer subtyping, filters out uninformative image areas, and can refuse to make a decision in unclear cases.
TRUECAM was developed by researchers from the Vanderbilt University Medical Center and the Hong Kong Polytechnic University. It was initially tested for the subtyping of non-small cell lung cancer (NSCLC) on whole-slide images and subsequently expanded to other cancer types and organs.
The framework acts as an interface to existing AI models. It quantifies uncertainties, recognizes inputs outside the training range, and eliminates interfering factors such as normal tissue areas or poorly stained specimens. This is intended to make AI not only more accurate but also fairer and more efficient.
In tests, TRUECAM surpassed existing solutions in accuracy and speed. It also enables customizable accuracy guarantees and improves fairness across gender and ethnic groups. Particularly noteworthy is that the system focuses on the same diagnostically relevant areas as pathologists.
"Trustworthy AI in the medical field is a prerequisite for realizing the potential of this transformative technology," explained Bradley Malin of Vanderbilt. The method addresses uncertainties that could arise from institution-specific preparation methods, artifacts, or unusual tissue variations.
The study was published in the journal Nature Biomedical Engineering (DOI: 10.1038/s41551-026-01694-8). The researchers see TRUECAM as an important step towards safer AI applications in cancer diagnostics.
