Unit-Based Histopathology Tissue Segmentation via Multi-Level Feature Representation
This work addresses the problem of efficient and accurate tissue segmentation in histopathology for clinical tasks like tumor-stroma quantification, though it is incremental as it builds on existing tile-based and transformer methods.
The paper tackles histopathology tissue segmentation by proposing a unit-based framework that classifies fixed-size tiles instead of pixels, reducing annotation effort and computational cost while maintaining accuracy. It achieves superior performance over U-Net variants and transformer-based baselines on a dataset of 386,371 tiles from 459 H&E-stained breast tissue regions.
We propose UTS, a unit-based tissue segmentation framework for histopathology that classifies each fixed-size 32 * 32 tile, rather than each pixel, as the segmentation unit. This approach reduces annotation effort and improves computational efficiency without compromising accuracy. To implement this approach, we introduce a Multi-Level Vision Transformer (L-ViT), which benefits the multi-level feature representation to capture both fine-grained morphology and global tissue context. Trained to segment breast tissue into three categories (infiltrating tumor, non-neoplastic stroma, and fat), UTS supports clinically relevant tasks such as tumor-stroma quantification and surgical margin assessment. Evaluated on 386,371 tiles from 459 H&E-stained regions, it outperforms U-Net variants and transformer-based baselines. Code and Dataset will be available at GitHub.