Thread by @Caldera60373705

Our work on Artificial Intelligence and HCC immune signatures in now out in @JHepatology !#AI
#pathology @EASLnews @FondationARC @InfoCancerFoie @Stecarus @jnkath @GhaffariNarmin
@Eric_Vibert @INCa_veillePro

journal-of-hepatology.eu/article/S0168-…
👇👇👇

Immune and inflammatory gene signatures may predict HCC sensitivity to immunotherapy, and we thus train AI models to recognize tumors with activation of these signatures directly from HCC histology. We used approx. 350 HCC slides together with RNA seq data (TCGA series)
👇

We investigated different deep learning models, the first being a patch based approach

the other models were multiple instance learning based, where the analysis focus on some areas of the slides

We obtained good performances, in particular with the CLAM model, on the discovery set of 350 cases

we then validated our best models in a validation dataset of 139 cases with slides encoded in a different format, with different stainings and with a different RNA expression technology