Algorithm 1 Explainability Pipeline for DPCSE-Net

Require: Input image x, trained DPCSE-Net model F   1:Compute feature maps $A^k$ and class score $y^c=F\left(x\right)$   2:Grad-CAM: Compute importance weights ${\alpha }_k^c$ and generate the spatial heatmap $L_{\mathrm{GradCAM}}^c$ to visualize class-discriminative regions   3:SE-Attention: Extract and normalize channel-wise attention weights s from the SE module to form $H_{\mathrm{SE}}$, indicating the most informative feature channels   4:Integrated Gradients: Calculate the pixel-level attribution map $\mathrm{IG}\left(x\right)$ by integrating gradients from a baseline $x^{\prime }$ to the input image x   5:Return the visualization set $\{L_{\mathrm{GradCAM}}^c,H_{\mathrm{SE}},\mathrm{IG}\left(x\right)\}$