Superpixel Boundary Correction for Weakly-Supervised Semantic Segmentation on Histopathology Images

TL;DR

The paper tackles boundary accuracy in CAM-based weakly supervised semantic segmentation for histopathology WSIs by introducing a dual strategy: multi-layer pseudo-mask fusion and superpixel floodfill refinement. It combines CAMs from three CNN depths with a fused objective $\mathcal{L}=\lambda_1 l(s,p_1)+\lambda_2 l(s,p_2)+\lambda_3 h(s,p_3)$ to infuse multi-scale information, and refines pseudo-masks with a floodfill rule over SLIC superpixels using $n_{ij}=\left|\{pixel\in c_i:\text{class}_p(pixel)=j\}\right|$, $l_i=\arg\max_j \frac{n_{ij}}{|c_i|}$, and $r_i=\frac{n_{i l_i}}{|c_i|}$ with $r_i>\tau$ guiding reassignment. To address background confounding, the paper adopts a causal-intervention view, approximating $P(Y|do(X))=\sum_i P(Y|X,b_i)P(b_i)$ and employing an EM-like iterative algorithm to update an average segmentation mask. Evaluations on the BCSS dataset show a state-of-the-art mIoU of $0.7108$, with notable improvements in boundary delineation for tumor microenvironments, demonstrating a practical and lightweight enhancement to WSSS in histopathology.

Abstract

With the rapid advancement of deep learning, computational pathology has made significant progress in cancer diagnosis and subtyping. Tissue segmentation is a core challenge, essential for prognosis and treatment decisions. Weakly supervised semantic segmentation (WSSS) reduces the annotation requirement by using image-level labels instead of pixel-level ones. However, Class Activation Map (CAM)-based methods still suffer from low spatial resolution and unclear boundaries. To address these issues, we propose a multi-level superpixel correction algorithm that refines CAM boundaries using superpixel clustering and floodfill. Experimental results show that our method achieves great performance on breast cancer segmentation dataset with mIoU of 71.08%, significantly improving tumor microenvironment boundary delineation.

Figures (4)

• Figure 1: Pipeline of multi-layer superpixel correction for WSSS on WSIs.
• Figure 2: Illustration of superpixels on WSI
• Figure 3: Illustration of patches and masks in BCSS dataset.
• Figure 4: Visualization of superpixel floodfill refinement for BCSS dataset.