Table 1.
Descriptions of BCCD and WBCs dataset.
Fig 1.
Sample images from BCCD (The first row) and the WBCs dataset (The second row).
Among them, (a) and (e) are neutrophils, (b) and (f) are monocytes, (c) and (g) are eosinophils, and (d) and (h) are lymphocytes.
Fig 2.
Flowchart of our method.
Table 2.
Comparison of CNN structure between WBC-AMNet and other models.
Fig 3.
Classification accuracy versus the number of iterations in the training phase.
(epoch = 20 and batch size = 32).
Table 3.
Training results of tri-classification of BCCD images under different epoch and batch size.
Table 4.
Training results when epoch = 20 and batch size = 32.
Fig 4.
ROC curve and confusion matrix.
(a) ROC curve of three subtypes of WBC. (b) Confusion matrix of three subtypes of WBC.
Table 5.
Training results when epoch = 20 and batch size = 32.
Fig 5.
Confusion matrices of other CNN models.
(a)VGG. (b)MobileNetV2. (c)ResNet. (d)SE-ResNeXt.
Table 6.
Training results of different WBC subtypes.
Fig 6.
ROC curve and confusion matrix.
(a) ROC curve of four subtypes of WBC. (b) Confusion matrix of four subtypes of WBC.
Table 7.
Statistical results of nine classic CNN models.
Fig 7.
Tri-classification line chart of WBCs dataset.
Table 8.
Tri-classification results of images from WBCs dataset.
Fig 8.
ROC curve and confusion matrix.
(a) ROC curve of three subtypes of WBC. (b) confusion matrix of three subtypes of WBC.
Fig 9.
(a) MobileNetV2. (b) ResNet. (c) SE-ResNeXt.
Table 9.
Quad-classification results of images from WBCs dataset.
Fig 10.
Quad-classification line chart of WBCs dataset.
Fig 11.
ROC curve and confusion matrix.
(a) ROC curve of four subtypes of WBC. (b) Confusion matrix of four subtypes of WBC.
Fig 12.
(a) MobileNetV2. (b) ResNet. (c) SE-ResNeXt.
Fig 13.
WBC-AMNet visualization analysis of attention to different feature maps.