Phikon-Extracted features dataset from TCGA-THYM WSIs

We utilized a publicly available WSI dataset from the TCGA-THYM project, consisting of 242 slides across six thymic tumor subtypes [A, AB, B1, B2, B3, CA]. WSIs were split into patches using the Yottixel method (Kalra et al., 2020). Phikon (Filiot et al., 2023), a pathology-SSL feature extractor, was then applied. Extracted features were organized into uniformly sized 1k bags via a chunking strategy. Bags and their labels were split (80:20) into train-test pth files.