The more that we use whole slide analysis, the more we are amazed that anyone tolerates the old way of doing things: doing image analysis or manual reads on small “representative” sample areas. Let’s look at a mouse spleen example.
Histology pattern recognition was tested on mouse spleen sections in an effort to quantitate normal splenic tissue components that are commonly altered in amount during physiologic or toxicologic adaptations. The pathologist identified 5-10 regions per slide for each lesion of interest, and then applied the analysis across the entire slide.
In the top figures, the lymphoid tissue across the whole spleen is 30.1%. But in a “representative” sample below, the lymphoid tissue comprises 46.9% – this is a 56% error!
Use the tissue, the whole tissue, and nothing but the whole tissue when running image analysis. Otherwise you will have some ‘expleening’ to do.
Across an entire splenic section, lymphoid tissue comprises 30.1% in this mouse spleen.
In a smaller evaluation area, lymphoid tissue comprises 46.9%.
Mouse spleens have a specific shape, and the percent of lymphoid tissue in each cross-section can vary greatly with the particular plane of section made through each spleen during tissue trimming. Consequently, if the same exact plane of section is not consistently made when trimming a group of mouse spleens, more “expleening” might be necessary after observing some very strange treatment effects for the study.
This can be generalized. Analysing the full digital slide obviously removes the bias of selecting representative fields in the tissue; but the tissue sample must then be comparable between animals. The bias just goes from the image level to the slide level and must still be controlled. And perhaps it will be more difficult to control: it would be useful to test this.