Fig. 2: Patch-level quantitative analysis on training cohorts or comparable datasets. | npj Precision Oncology

Fig. 2: Patch-level quantitative analysis on training cohorts or comparable datasets.

From: A universal immunohistochemistry analyzer for generalizing AI-driven assessment of immunohistochemistry across immunostains and cancer types

Fig. 2

a List of eight deep learning models trained on different cohort combinations. H-B, HER2 of breast; P-L, PD-L1 22C3 of lung; P-B, 22C3 of breast; P-LUB, 22C3 of lung, urothelium, and breast; PH-B, 22C3 and HER2 of breast; PH-LB, 22C3 of lung and HER2 of breast; PH-LUB, 22C3 and HER2 of lung, urothelium, and breast. The different stain combinations (e.g. 22C3 or HER2 is utilized or not), are visualized by color. Performance of the eight models in training cohorts, where the stain type may be utilized during training – b 22C3 in lung cancer, c 22C3 in urothelial cancer, d 22C3 in breast cancer, e HER2 in breast cancer. f 22C3 in pan-cancer and g PD-L1 SP142 of lung are not used during training. PD-L1, Programmed Death-Ligand 1; HER2, Human Epidermal growth factor Receptor 2; mF1, mean F1 score. (****p < 0.0001; ***p < 0.001; **p < 0.01; *p < 0.05; ns, not significant).

Back to article page