Supervised Information Mining From Weakly Paired Images for Breast IHC Virtual Staining

Area of Science:

• Digital pathology
• Computational imaging
• Biomedical image analysis

Background:

• Immunohistochemistry (IHC) is crucial for breast cancer diagnosis and treatment planning but is complex and costly.
• Hematoxylin and Eosin (H&E) staining is a more accessible alternative, but lacks IHC's specific diagnostic information.
• Virtual staining from H&E to IHC could bridge this gap, but accurate image registration and supervision are challenging.

Purpose of the Study:

• To develop a virtual staining method to generate IHC images from H&E images for breast cancer.
• To address the challenge of inaccurate pixel-level pairing between adjacent H&E and IHC tissue layers.
• To improve the accuracy and applicability of virtual IHC staining in clinical settings.

Main Methods:

• Proposing a generative adversarial network (GAN) incorporating Optimal Transport-based Supervised Information Mining (OT-SIM) for instance-level supervision.
• Implementing Pathological Correlation-based Supervised Information Mining (PC-SIM) for batch-level supervision, leveraging image correlations.
• Utilizing adjacent layer tissue images for training, despite imperfect pixel-level alignment.

Main Results:

• The proposed SIM-GAN method demonstrated superior performance in virtual IHC staining compared to existing state-of-the-art techniques.
• Quantitative and qualitative evaluations on two benchmark breast tissue datasets confirmed the method's effectiveness.
• The approach successfully mined supervised information from H&E and IHC image pairs, even with registration inaccuracies.

Conclusions:

• The developed virtual staining technique offers a promising solution to reduce the cost and complexity associated with traditional IHC staining.
• SIM-GAN effectively addresses the challenge of non-ideal image pairing by employing novel information mining mechanisms.
• This advancement holds potential for improving personalized treatment strategies in breast cancer through more accessible IHC-like information.