Fig. 4
From: CLWD: a Chinese histopathology dataset for lung adenocarcinoma subtype classification

Baseline benchmarking under another stratified method (8:1:1). Panels show one-vs-rest ROC curves for the evaluated classes for (A,D) CLAM, (B,E) Graph Transformer, and (C,F) TransMIL on (A to C) CLWD dataset and (D to F) DHMC dataset. Shaded regions indicate variability across repeats (mean ± SD). The diagonal dashed line indicates chance-level performance.