PathAI is thrilled to announce the publication of its latest manuscript, in partnership with Flare Therapeutics, in AI in Precision Oncology!

Molecular subtyping of urothelial carcinoma (UC) has the potential to inform prognosis and treatment decisions. However, this process has traditionally required RNA sequencing, a costly technology with limited availability to patients.

In this manuscript, we developed an additive multiple instance learning approach to predict luminal status from routine, H&E-stained UC whole slide images, demonstrating robust performance across three datasets:

• > 0.95 AUROC
• > 89% accuracy
• > 86% sensitivity
• > 82% specificity

A key feature of this algorithm is its ability to generate molecular subtype predictions from a universally available specimen, the H&E-stained slide. This functionality not only potentially eliminates the need to use additional tissue for specialized testing, but also provides the ability to generate standardized prediction scores at scale across digitally-enabled laboratories, including clinical trial sites.

These results support the potential of AI-powered digital pathology approaches for molecular subtyping in UC and similar clinical tasks across other indications.

Read the full paper to learn more about the original algorithm.