Figure 2
From: Emphysema subtyping on thoracic computed tomography scans using deep neural networks
The reconstruction network on top of the ResNet backbone for producing dense features. The output of the reconstruction network is fed into either classification or the regression output head depending on the training strategy for predicting centrilobular scores (CLE out) and paraseptal scores (PSE out). Note that we do not use both output heads together for multi-tasking.
