Fig. 2: Performance results for RCC-subtype classification and KTX disease classification on external validation set.
From: Ecologically sustainable benchmarking of AI models for histopathology
Performance results of TransMIL, CLAM, InceptionV3,ViT, and Prov-GigaPath models for RCC subtype (n = 289) and KTX disease classification (n = 173) tasks. a, e, i, m, q show the AUROC and b, f, j, n, r show the AUPRC for all models for the RCC-subtype classification task, including Prov-GigaPath. c, g, k, o, s show the AUROC and d, h, l, p, t show the AUPRC for all models for the KTX disease classification task, respectively. LF label frequency of the corresponding class, TPR true positive rate, FPR false positive rate, AUROC area under the receiver operating characteristics curve, PR-AUC precision-recall area under the curve, RCC renal cell carcinoma, ccRCC clear cell renal cell carcinoma, papRCC papillary renal cell carcinoma, chRCC chromophobe renal cell carcinoma.
