Figure 1: Visualization of the contextualized report
embeddings for normal (blue) and abnormal (red) cases captured using an
open source ‘lasso’ tool (Wood et al., 2020b) for use in this study.
Hierarchical attention network
In this work we present two hierarchical attention networks (HAN): one
taking as input
intra-slice regions spanning multiple image slices, and another taking
as input individual
slices spanning multiple imaging sequences.