Owkin's MesoNet predicts mesothelioma survival from whole-slide pathology images with no pathologist-marked regions, validated on a French cohort and on TCGA
What happened: Published in Nature Medicine, it beat current pathology classification at predicting survival. The regions it relied on lay mainly in the stroma and were tied to inflammation, cell diversity and vacuolisation.
Why it matters: An early case of a startup's model pointing to new tissue features. It is retrospective on stored slides, not a prospective test or a clinical decision.