Figure 7.

Experimental confirmation of lymphocytic identification by deep learning in a genetically engineered mouse model. A preclinical experiment was performed using mice that have been genetically engineered to lack mature lymphocytes as a negative control. Spleens and thymuses from Rag2^-/- or littermate control Rag2^+/- mice were formalin-fixed, paraffin-embedded, cut, stained with H&E, and imaged on a whole slide digital pathology scanner. A pre-trained deep learning algorithm, independently developed on human tissue to detect lymphocytes on H&E images, was applied to the mouse tissue to measure differences in lymphocytic infiltrate between Rag2^-/- and Rag2^+/- mice. Homozygous knockout mice (Rag2^-/- ) were used as negative controls because they lack mature lymphocytes. A heterozygous mouse (Rag2^+/- ) with one intact allele was used as a positive control.