Skip to main content
. 2025 Oct 8;19:1682603. doi: 10.3389/fnins.2025.1682603

FIGURE 1.

Diagram showing a neural network architecture for image processing. Inputs include Flair, T1, T1ce, and T2 brain scans. A 5x5x5 convolution is applied, followed by three downsampling stages with SMW Conv modules. The process incorporates dilation convolution and context feature calibration. The upsampling flow includes gated attention mechanisms. A 1x1x1 convolution produces an output segmentation map. Modules are color-coded: SMW Conv (yellow), dilation and context calibration (green). Symbols indicate concatenation and gating steps.

MAUNet structure.