Algorithm 1 DRDCNN model

Input:  two training sets: S1, S2;

test set: T;

models: U-net (U), VGG (V);

transferred weights from the published study: W

dot density maps: DS1;

cell counts label: Cconcatenate(S1, S2).

Procedure:

1.   L1 = f1 (U (Wu, S1), DS1))

train a set of weights W1 which was initialized by Wu to minimize the loss function L1.

2. L2 = f2 (V (W2, U (W1, concatenate (S1, S2)), Cconcatenate(S1, S2)))

W1 was initialized from Step 1 and fixed to untrainable.

W2 was trained to minimize the loss function L2.

Output:  predicted cell counts: cDRDCNN.

cDRDCNN = V (W2, U (W1, T))