In this work, we have developed methods that detect and characterize millions of WBC and RBC in peripheral blood smears across different conditions; with this vast collection of cells, we were able to developed methods for weakly supervised learning which allowed us to 1) infer computational morphotypes (visually cohesive groups of cells) and 2) use these to predict different diseases.
Here, cells are colloured according to their computational morphotype and divided according to their cell type. Since our disease predictions are heavily tied with the stability of computational morphotypes (i.e. their reproducibility across multiple cross-validation folds), we also provide the option to inspect only the stable computational morphotypes.
Clicking on a point will highlight all cells belonging to the same computational morphotype.