RGGC-UNet: Accurate Deep Learning Framework for Signet Ring Cell Semantic Segmentation in Pathological Images

Area of Science:

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

Background:

• Semantic segmentation of Signet Ring Cells (SRCs) is crucial for diagnosing SRC carcinoma.
• Deep learning shows promise in computer-aided diagnosis but often involves computationally intensive models.
• Limited ground truth data for SRCs hinders segmentation technique development.

Purpose of the Study:

• To develop an efficient and accurate deep learning framework for SRC semantic segmentation.
• To address the computational overhead and data limitations in existing methods.
• To improve the diagnostic accuracy of SRC carcinoma through enhanced image analysis.

Main Methods:

• Introduction of RGGC-UNet, a UNet-based framework with a novel encoder.
• Utilized residual ghost blocks with ghost coordinate attention for computational efficiency.
• Enriched the DigestPath 2019 dataset with fully annotated SRC mask labels.

Main Results:

• The proposed RGGC-UNet model demonstrated superior segmentation accuracy compared to leading-edge models.
• The model achieved significant reductions in computational overhead.
• Experimental results validate the model's effectiveness and efficiency for pathological diagnosis.

Conclusions:

• RGGC-UNet offers an efficient and accurate solution for Signet Ring Cell semantic segmentation.
• The framework effectively minimizes computational costs while maximizing segmentation performance.
• This advancement holds potential for improving the accuracy and efficiency of SRC carcinoma diagnosis.