Fig 1.
Description of collected data.
(A) Examples of white blood cells in erythroid series (C1-4) and myeloid series (C5-10). (B) Distribution of collected data. (C) Cellular component distribution in bone marrow.
Fig 2.
(A) Oversampling and (B) Augmentation.
Fig 3.
(A) Illustration of the convolutional neural network. (B) Description of the proposed dual-stage convolutional neural network.
Table 1.
Classification performance of the network trained on different datasets.
Fig 4.
(A) Graph of validation accuracy and training loss during training of network. The dotted red box shows the magnified view of the first 50 epochs. (B) Confusion matrix of AG+OS 600.
Fig 5.
Examples of correctly classified cells by the AG+OS 600 network.
(A) WBCs with backgrounds showing background invariance of the network. (B) Oversampled WBCs showing location invariance of the network. (C) Augmented WBCs showing rotation invariance of the network.
Fig 6.
Examples of incorrectly classified cells by the AG+OS 600 network.
(A)-(D) Cells whose ground truth is band neutrophil. (E)-(H) Cells whose ground truth is segmented Neutrophil.
Fig 7.
Comparison of confusion matrices of AG+OS 600 and dual-stage CNN.