C2P-GCN: Cell-to-Patch Graph Convolutional Network for Colorectal Cancer Grading.

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Graph-based machine learning

Background:

• Graph-based learning is favored for colorectal cancer histology image analysis due to its ability to encode tissue structure.
• Current methods often fail to capture entire slide structure and require large datasets.

Purpose of the Study:

• To propose a novel cell-to-patch graph convolutional network (C2P-GCN) for improved colorectal cancer histology image grading.
• To develop a method that integrates local and global tissue structure information from whole slide images (WSIs).

Main Methods:

• A two-stage graph formation approach: patch-level graph construction based on cell organization, followed by an image-level graph construction using patch similarity.
• Utilizing a multi-layer GCN-based classification network on the integrated graph representation.

Main Results:

• The C2P-GCN effectively integrates local patch details and global WSI structure.
• The proposed method demonstrates effectiveness with significantly reduced training data requirements compared to existing models.
• Experimental validation on two colorectal cancer datasets confirmed the method's efficacy.

Conclusions:

• C2P-GCN offers a powerful approach for colorectal cancer grading by leveraging dual-phase graph construction.
• The method enhances the capture of WSI structural data, leading to improved performance and reduced data dependency.
• This approach holds promise for advancing computational pathology in cancer diagnosis.