Uncertainty-Aware AI Tool Improves Digital Pathology for Cancer Subtyping

A persistent obstacle to clinical use is uncertainty: models may be overconfident when faced with unfamiliar or low-quality inputs. Pathology teams need decision support that screens noise, flags out‑of‑scope slides, and provides measurable accuracy safeguards. A new study shows an uncertainty‑aware framework that strengthens AI reliability for cancer subtyping.

At Vanderbilt Health (Nashville, TN, USA), researchers developed TRUECAM, a versatile, uncertainty-aware AI wrapper for digital pathology systems. The authors describe a wrapper as an interface layer that customizes, formats, and automates how users interact with the underlying AI. The team demonstrated TRUECAM primarily in non-small cell lung cancer (NSCLC) subtyping using whole-slide images.

TRUECAM is engineered to recognize out‑of‑scope inputs and to filter noninformative regions, such as normal or poorly stained tissue, that could distort slide‑level inference. According to the authors, these complementary functions enable customizable accuracy guarantees for subtype classification. The framework was evaluated as a wrapper for a widely used NSCLC subtyping architecture and for four newer, more generalized digital pathology foundation models.

Testing incorporated NSCLC whole‑slide images from two geographically diverse cancer research consortia, a curated set of clinically meaningful out‑of‑scope images, and a sequence of real‑world cases from Queen Mary Hospital in Hong Kong. Evaluation also extended to cancer tissues from multiple organs, including breast, brain, and kidney. Compared with existing approaches to trustworthy digital pathology AI, the authors describe TRUECAM as accurate, rapid, and efficient without substantial added costs.

Across experiments, TRUECAM reliably detected out‑of‑scope inputs, abstained on challenging cases to enable deferral to pathologists, and delivered error rates that met prespecified accuracy targets. The approach improved fairness across sex and race and generalized beyond lung cancer datasets. The study was published in Nature Biomedical Engineering on June 23, 2026.

“Achieving trustworthy AI in the medical domain is requisite for realizing the potential of this transformative technology. It's not only that a patient's clinical profile can fall out of scope of your model's training data, but other sources of variation, such as your institution's method of collecting and staining specimens, or artifacts and irregularities that tend to arise in tissue preparation, can also prompt your model to arrive confidently at a mistaken conclusion. TRUECAM provides a thoroughgoing, versatile and efficient solution to these potentially unsafe shortcomings,” said Bradley Malin, Ph.D., professor of biomedical informatics, biostatistics and computer science, and holder of the Accenture Chair.

Related Links
Vanderbilt University Medical Center