Figure 1.

Overview of research design for lymphocyte deep learning model development and evaluation. We first developed a knowledge-guided deep learning framework to measure lymphocytes on H&E-stained kidney tissue, where the fidelity of training labels was maximized through single-cell resolution image registration of H&E to IHC staining with a cocktail of anti-CD3 and anti-CD20 antibodies to stain T and B lymphocytes, respectively. Following traditional statistical testing of the model based on receiver operating characteristic (ROC) curve analysis, we then characterized pattern-preserving features of the immune microenvironment based on pathomic feature extraction and calculated their error rates relative to the shift-invariant IHC antibody measurements. Finally, we performed a preclinical experiment to confirm that our deep learning model is able to identify lymphocytes in a genetically engineered mouse model, where Rag2 knockout mice without mature lymphocytes are used as a negative control.