Skip to main content
. 2021 Dec 16;43(5):1640–1656. doi: 10.1002/hbm.25748

FIGURE 1.

FIGURE 1

Schematic network architecture and workflow of bilinear fusion. Two hundred fifty‐six‐dimensional features were extracted from brain images through a ResNet network. All MaxPooling layers were designed as stride 2 × 2 × 2 and kernel size 3 × 3 × 3. Imaging features and blood parameters were fused in the Multimodal Compact Bilinear Fusion module. The entire process can be completed by an end‐to‐end cascade network