TABLE I. EM Training Algorithm.

Algorithm 1. EM Training of UNet-2.

INPUT , , infected region predicted from UNet-1.

INITIALIZATION from UNet-1.

1 WHILE not stopped:

2 Get a minibatch;

E-step:

3 FOR in the minibatch:

4 , consolidation prediction;

5 FOR each pixel j:

6 , posterior probability;

7 IF THEN ELSE ;

M-step:

8 Compute and update .

RETURN .