GCSBA-Net: Gabor-Based and Cascade Squeeze Bi-Attention Network for Gland Segmentation

Area of Science:

• Medical Imaging
• Computational Pathology
• Oncology

Background:

• Colorectal cancer is a leading cause of cancer death globally.
• Accurate pathological diagnosis, particularly gland segmentation in histopathology images, is critical for effective treatment.
• Existing gland segmentation methods often overlook crucial texture and multi-scale information.

Purpose of the Study:

• To develop an improved gland segmentation technique for colorectal cancer histopathology images.
• To enhance the accuracy and robustness of automated pathological diagnosis.
• To address limitations in current segmentation methods regarding texture and multi-scale feature extraction.

Main Methods:

• Utilized a Gabor-based module for extracting multi-scale and multi-directional texture information from histopathology images.
• Designed a Cascade Squeeze Bi-Attention (CSBA) module, incorporating Atrous Cascade Spatial Pyramid (ACSP), Squeeze Position Attention (SPA), and Squeeze Channel Attention (SCA).
• Proposed a hybrid loss function to effectively handle data imbalance and boundary ambiguity in segmentation tasks.

Main Results:

• The proposed method demonstrated state-of-the-art performance on the GlaS challenge dataset.
• Achieved superior results on the CRAG colorectal adenocarcinoma dataset.
• Successfully integrated texture, multi-scale attention, and improved boundary detection for enhanced gland segmentation.

Conclusions:

• The developed method significantly advances gland segmentation in colorectal cancer histopathology.
• The integration of Gabor filters and the CSBA module provides a powerful tool for computational pathology.
• This approach holds promise for improving the accuracy and efficiency of colorectal cancer diagnosis.