GRU-Net: Gaussian Attention Aided Dense Skip Connection Based MultiResUNet for Breast Histopathology Image Segmentation

TL;DR

GRU-Net tackles nuclei segmentation in breast histopathology by integrating a MultiResUNet backbone with a text-informed Gaussian attention module (GdAM) and a controlled dense residual block (CDRB) to regulate encoder–decoder information flow. Text priors encoded via DistilBERT guide attention, while the CDRB gates skip connections using a learned scale $λ$, improving boundary accuracy and reducing irrelevant feature transfer. Across MonuSeg and TNBC, GRU-Net achieves state-of-the-art Dice and IoU scores, validated by ablation and cross-dataset experiments, though boundary errors and over-segmentation in noisy patches indicate avenues for refinement. The approach demonstrates the value of combining multi-scale feature extraction, text-guided priors, and controlled information flow for robust histopathology segmentation with practical implications for pathology workflows.

Abstract

Breast cancer is a major global health concern. Pathologists face challenges in analyzing complex features from pathological images, which is a time-consuming and labor-intensive task. Therefore, efficient computer-based diagnostic tools are needed for early detection and treatment planning. This paper presents a modified version of MultiResU-Net for histopathology image segmentation, which is selected as the backbone for its ability to analyze and segment complex features at multiple scales and ensure effective feature flow via skip connections. The modified version also utilizes the Gaussian distribution-based Attention Module (GdAM) to incorporate histopathology-relevant text information in a Gaussian distribution. The sampled features from the Gaussian text feature-guided distribution highlight specific spatial regions based on prior knowledge. Finally, using the Controlled Dense Residual Block (CDRB) on skip connections of MultiResU-Net, the information is transferred from the encoder layers to the decoder layers in a controlled manner using a scaling parameter derived from the extracted spatial features. We validate our approach on two diverse breast cancer histopathology image datasets: TNBC and MonuSeg, demonstrating superior segmentation performance compared to state-of-the-art methods. The code for our proposed model is available on https://github.com/AyushRoy2001/GRU-Net.

Figures (8)

• Figure 1: Example of nuclei segmentation. Sample histopathology image and its pixel-level annotations.
• Figure 2: Qualitative comparison of the proposed model with Attention-UNet and U-Net++.
• Figure 3: Overall workflow of the proposed GRU-Net.
• Figure 4: Block diagram representation of CDRB module.
• Figure 5: Block diagram representation of GdAM module.