IHC-DualNet: a dual-branch graph-based architecture for interpretable and precise immunohistochemistry tissue...

Area of Science:

• Computational pathology
• Medical image analysis
• Artificial intelligence in oncology

Background:

• Immunohistochemistry (IHC) is vital in oncology for diagnosis and prognosis.
• Manual annotation of IHC images is time-consuming, subjective, and lacks reproducibility.
• Existing automated methods fail to balance global context and local details, leading to segmentation inaccuracies and lack of interpretability.

Purpose of the Study:

• To develop an automated, accurate, and interpretable IHC image segmentation framework.
• To emulate pathological reasoning for improved segmentation performance.
• To address limitations of current automated methods in capturing tissue context and local features.

Main Methods:

• Proposed IHC-DualNet, a dual-branch graph-based framework.
• Global Transformer Reasoning (GTR) branch for tissue-wide context using a novel IHC-GTR layer.
• Local Graph Attention branch for fine-grained local features, integrated via Adaptive Feature Fusion (AdaFuse).

Main Results:

• Achieved superior accuracy and robustness across three challenging IHC datasets (breast, esophageal, lung cancer).
• Demonstrated high performance metrics, including 93.88% Dice and 90.26% Intersection over Union on breast cancer data.
• Successfully segmented differentially expressed regions with high precision.

Conclusions:

• IHC-DualNet significantly advances automated IHC analysis.
• The framework provides accurate, robust, and interpretable segmentation, aiding tumor grading and patient stratification.
• This approach offers a reliable alternative to manual annotation in clinical oncology workflows.