A confidence-based, artificial intelligence pathology model for diagnosis of intrahepatic cholangiocarcinoma - PubMed
5 days ago
- #Cholangiocarcinoma
- #Deep Learning
- #Artificial Intelligence
- Intrahepatic cholangiocarcinoma (ICCA) is a rare but highly lethal adenocarcinoma with diagnostic challenges due to overlapping features with metastatic liver cancers.
- A retrospective analysis of 544 patients across five European centers was conducted, comparing ICCA and metastases from extrahepatic cancers.
- Three deep-learning architectures (Ctranspath/HistoBistro, UNI/CLAM, CONCH/TITAN) were evaluated, with CONCH/TITAN showing the best performance (AUROC: 0.840).
- A confidence estimation system using Generalized-ODIN (G-ODIN) was implemented, improving AUROC to 0.958 with a 0% false positive rate.
- The final model, AI2CCA, was prospectively validated in 161 patients across France, India, and Korea, achieving high AUROCs (1.00 and 0.965).
- The study demonstrates the clinical utility of a confidence-based AI biomarker for ICCA diagnosis, potentially reducing unnecessary investigations and accelerating treatment decisions.