Figure 8.

SNN architecture used for patient verification on the ChestX-ray14¹⁶ dataset. The feature extraction blocks (light blue) share the same set of network parameters and produce the feature representations ${\mathit{z}}_1$ and ${\mathit{z}}_2$ (yellow). After merging (orange) and an additional FC and sigmoid layer $\sigma$ (light blue), the network yields the final output score $\widehat{y}$ (green). For our patient re-identification experiments, we used the same architecture but ejected all the layers from the merging layer onwards.