Image Recognition
We trained MILLIE models to distinguish between normal, ALL and AML samples and we evaluated their classification performance both at sample and cell-level on separate hold out test sets. For each sample, our approach extracts patches containing individual cells from high-resolution fields of view of peripheral blood films and bone marrow aspirates (Fig. A histogram-based segmentation28 (see “Methods” section) was employed to generate binary masks corresponding to individual cells from the RGB images. Once trained, individual cells can be passed one-by-one through the MILLIE models which classifies them as indicators for the specific disorders MILLIE learned to predict at a sample level. At testing time, all image patches from each sample were passed through the network.