Multi-module UNet++ for colon cancer histopathological image segmentation.

Area of Science:

• Pathology
• Medical Imaging
• Computer Science

Background:

• Accurate segmentation of glandular and cellular contours is crucial for colorectal cancer diagnosis.
• Challenges include nuclear staining heterogeneity, size variations, boundary overlap, and clustering.
• Deep learning, especially encoder-decoder architectures, offers potential for improved segmentation.

Purpose of the Study:

• To propose the RPAU-Net++ model for enhanced pathological image segmentation in colorectal cancer.
• To integrate ResNet-50, Joint Pyramid Fusion Module (JPFM), and Convolutional Block Attention Module (CBAM) into the UNet++ framework.
• To improve the accuracy and robustness of colorectal cancer cell and gland segmentation.

Main Methods:

• Developed the RPAU-Net++ model by integrating ResNet-50 encoder, JPFM, and CBAM into the UNet++ architecture.
• ResNet-50 was used for residual skip connections to enhance feature representation and model stability.
• JPFM facilitated multi-scale feature fusion, while CBAM improved feature discriminability through adaptive weighting.

Main Results:

• RPAU-Net++ demonstrated superior performance compared to mainstream models on colorectal cancer pathology datasets (GlaS, CoNIC) and the PanNuke dataset.
• The model achieved significant improvements in key segmentation metrics, including Intersection over Union (IoU) and Dice coefficient.
• The proposed architecture effectively addressed challenges like nuclear variations and clustering, leading to more precise segmentation.

Conclusions:

• RPAU-Net++ offers a more accurate and effective solution for pathological image segmentation in colorectal cancer.
• The multi-module collaborative fusion approach significantly enhances segmentation performance.
• This model holds promise for improving the accuracy of clinical diagnosis in colorectal cancer pathology.