Skip to main content
. Author manuscript; available in PMC: 2020 Dec 3.
Published in final edited form as: Med Image Comput Comput Assist Interv. 2020 Sep 29;12265:3–13. doi: 10.1007/978-3-030-59722-1_1

Fig. 2:

Fig. 2:

Our architecture. For an input highly multiplexed image I, we split channels and feed them into ResNet18 [4]. Next, the embedding encoder extracts an interpretable representation r = [r0, …, rN], where N is the number of channels. Both the backbone and embedding encoders share weights across channels. Finally, we adopt three fully-connected layers to estimate class probabilities p.