Fig. 1

Schematic representation of the computational pathology pipeline for the detection of foamy podocytes. After case selection and subdivision in training/validation and test sets, biopsies were annotated at both glomerular level (A: giving a label of foamy or not foamy, green and red in the figure) and at podocyte level (B: delineating the podocyte with foamy cytoplasm). Based on these annotations, cases were used to build a classification and segmentation model, for A and B annotations respectively. The obtained algorithms were applied on an external test set to evaluate their performances on either the classification of positive/negative glomeruli of the biopsy (glomerular detail) or the outlining of foamy podocytes within glomeruli (podocyte detail). The obtained pipeline (ZEbra Bodies Recognition by Artificial intelligence, ZEBRA) is available as a free QuPath-WSInfer extension (classification system) as and as a segmentation tool freely available as a GitHub repository.