Our manuscript published in Annals of Oncology

Diagnosing Intrahepatic Cholangiocarcinoma (ICCA) is a major clinical hurdle. Because its histological features closely overlap with those of metastatic liver cancers, patients often endure a battery of expensive and invasive tests—such as endoscopies—to rule out other primary tumor sites. These "exclusionary investigations" lead to critical treatment delays and increased healthcare costs.

Here we performed a multi-center study that involved the development of AI2CCA (Artificial Intelligence for ICCA diagnosis), a deep-learning model designed to solve this exact problem using routine liver biopsy slides. We also prospectively validate it on 4 centers across the globe.

Unlike traditional "black box" AI, AI2CCA uses a confidence-based framework (G-ODIN) that allows the model to abstain from a prediction when uncertainty is too high.