PASB: Pathology-aware Schrödinger bridge for virtual immunohistochemical staining.

Area of Science:

• Computational pathology
• Digital pathology
• Medical image analysis

Background:

• Virtual immunohistochemistry (IHC) staining automates IHC analysis by translating Hematoxylin and Eosin (H&E) images into IHC images using deep generative models.
• Weakly supervised methods are mainstream for virtual IHC staining, using adjacent tissue sections for guidance, thus avoiding precise alignment.
• Current methods struggle with extracting clinically meaningful pathological semantics and capturing histopathological data's complexity, leading to mode collapse and loss of critical structures.

Purpose of the Study:

• To address limitations in virtual IHC staining, specifically the failure to extract pathological semantics and the inability to capture data heterogeneity.
• To propose a novel weakly supervised method, Pathology-Aware Schrödinger Bridge (PASB), for improved virtual IHC staining.
• To enhance pathological consistency and generative quality in virtual IHC staining.

Main Methods:

• The Pathology-Aware Schrödinger Bridge (PASB) method utilizes the Schrödinger Bridge as a generative backbone to enhance diversity and reduce mode collapse.
• Incorporates Constraint-Driven Alignment Learning (CDAL) for high-level semantic supervision.
• Employs Similarity-based Dynamic Path Refinement (SDPR) to improve pathological consistency during image generation.

Main Results:

• The proposed PASB method demonstrates superior performance compared to existing virtual IHC staining techniques.
• PASB achieves enhanced generative quality and pathological consistency in the generated IHC images.
• Generated IHC images show clinical potential comparable to real IHC in downstream diagnostic tasks.

Conclusions:

• PASB effectively addresses the challenges of semantic extraction and data heterogeneity in virtual IHC staining.
• The method significantly improves the pathological consistency and generative quality of virtual IHC images.
• PASB offers a promising approach for automating IHC staining with clinically relevant outcomes.