Weakly Supervised Ternary Stream Data Augmentation Fine-Grained Classification Network for Identifying Acute Lymphoblastic Leukemia

. 2021 Dec 22;12(1):16. doi: 10.3390/diagnostics12010016

Algorithm 1. Coarse-to-Fine Prediction.

Require: Trained WT-DFN model

W

for

i = 1, 2, \dots, b

do (batch size

= b

)
1: Obtain raw probability

p_{i}^{r a w} : p_{i}^{r a w} = W (I_{r a w, i})

and out attention maps

F

by Equation (1);
2: Calculate normalize

F_{k}^{*}

by Equation (2);
3: Obtain crop image

I_{c r o p}

and erase image

I_{e r a s e}

by Equations (3) and (4), respectively;
4: Predict fine-grained probability

p_{i}^{c r o p} : p_{i}^{c r o p} = W (I_{c r o p, i})

p_{i}^{e r a s e} : p_{i}^{e r a s e} = W (I_{e r a s e, i})

;
5: Calculate the final prediction

p_{i}

by Equation (5);
end for
return

{p_{i}, \dots, p_{b}}