Fig. 3: Cell type predictions using ROSIE.
From: ROSIE: AI generation of multiplex immunofluorescence staining from histopathology images
A F1 scores (N = 817,765 cells) on the primary Stanford-PGC dataset, comparing the performance of ROSIE to two baselines: bulk phenotyping, which randomly assigns cell types based on sample-level cell type proportions, and morphology features, which uses a three-layer neural network to classify cells based on morphology features and the H&E RGB channels. Data are presented as mean values with error bars as the 95% bootstrapped confidence intervals. B Visualization of cell phenotype predictions from twelve median samples by Pearson R.
