Skip to main content
. Author manuscript; available in PMC: 2020 Mar 30.
Published in final edited form as: Lab Invest. 2019 Sep 30;100(1):98–109. doi: 10.1038/s41374-019-0325-7

Figure 2. Computational detection and classification of cells in bone marrow aspirate smears.

Figure 2.

(A) Cell detection was performed using a Faster R-CNN network built on the resnet101 fully convolutional network. (B) Following cell detection, a separate convolutional network was used to classify the detected cells into 12 cytological classes. (C) Detection and classification accuracy were evaluated through 6-fold cross-validation to measure detection and classification accuracy using human annotations of cytological class and bounding box location. Cross-validation was performed at the case level, so that annotated cells from each case were assigned entirely to either the training or testing set.