Skip to main content
Fig. 1 | BMC Cancer

Fig. 1

From: Deep learning identifies Acute Promyelocytic Leukemia in bone marrow smears

Fig. 1

Workflow of the multi-step deep learning model for APL recognition. We identified patients with APL, non-APL AML and healthy bone marrow donors by retrospective chart review. Representative images of bone marrow smears (BMS) were labeled according to diagnosis. After image preprocessing, transformation and augmentation, initial cell border proposals were given by the Faster Region-based Convolutional Neural Net (FRCNN) that were manually corrected on an online segmentation and annotation platform based on the VGG image annotator tool. The FRCNN was trained iteratively to improve cell border proposals. Segmented cells were manually labeled according to cell type (myeloblasts, promyelocytes) and Auer rods. Convolutional neural nets were then implemented on the automatically segmented cells for binary classification of individual cell types and features. Their output was used to train an ensemble neural net for the binary classification between APL and non-APL AML or APL and healthy bone marrow donor samples

Back to article page