Fig. 4: ESPer scores.
From: Ecologically sustainable benchmarking of AI models for histopathology
a–e show the ESPer scores calculated for the performance metrics AUROC accuracy, precision, recall, and F1, for all investigated models for RCC (N = 289). f–j show the same for KTX (N = 173). k, l show how changing the weighting factor \(w\) impacts the ranking of iESPer scores for RCC and KTX, respectively. m shows the projection of ESPer scores calculated from the AUROC metric for 5 years, based on the number of RCC cases in the EU for 2019 (n = 90,042). n shows the same, based on the number of kidney transplant cases in the EU for 2019 (n = 28,189). All CO2eq emissions in this figure are based on the energy mix of Germany. AUROC area under the receiver operating characteristics curve, iESPer inference environmentally sustainable performance, fpESPer future projection environmentally sustainable performance, RCC renal cell carcinoma, KTX kidney transplant.
