Utilizing Hybrid Mask and Upsampling Attention Gate for Multiple Immunohistochemistry Image Cell Recognition

Area of Science:

• Computational pathology
• Biomedical imaging analysis
• Machine learning in oncology

Background:

• Multi-immunohistochemistry (mIHC) enables simultaneous detection of multiple cellular phenotypes in tissue sections, crucial for cancer diagnosis and treatment.
• Automated cell detection and classification in mIHC images are vital but challenged by high cell density, heterogeneity, and extensive annotation requirements.
• Existing methods struggle with complex mIHC image data, necessitating more efficient and accurate computational approaches.

Purpose of the Study:

• To develop a novel automated model for accurate cell detection and classification in mIHC images.
• To address limitations of current methods, including high cell density, heterogeneity, and laborious annotation.
• To enhance the efficiency and reliability of computational pathology tools for cancer research.

Main Methods:

• A simplified point-based annotation strategy to significantly reduce manual labeling effort.
• A hybrid masking approach (Gaussian and circular masks) to capture diverse cell morphologies.
• Introduction of an Upsampling Attention Gate (UAG) for improved feature extraction in complex backgrounds.
• A post-processing module to resolve cell adhesion issues and enhance detection accuracy.

Main Results:

• The proposed model achieved F1 scores of 0.772 for cell detection and 0.747 for cell classification on an mIHC dataset.
• Demonstrated superior performance compared to existing methods across various evaluation metrics.
• Successfully addressed challenges related to cell density, heterogeneity, and morphology in mIHC images.

Conclusions:

• The developed automated model offers a significant advancement for cell detection and classification in mIHC imaging.
• This method provides a more efficient and accurate solution, reducing annotation burden and improving diagnostic potential.
• The findings pave the way for enhanced cancer diagnosis and treatment strategies through improved computational pathology tools.