A deep residual attention-based U-Net with a biplane joint method for liver segmentation from CT scans

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate liver segmentation in CT scans is vital for diagnosing liver tumors, a condition with high morbidity.
• 2D segmentation methods often neglect crucial 3D spatial information, limiting diagnostic accuracy.
• Existing methods struggle to fully leverage the 3D context within CT data.

Purpose of the Study:

• To develop an advanced deep learning model for precise liver segmentation from CT scans.
• To enhance the incorporation of 3D spatial information in liver segmentation tasks.
• To improve the accuracy and efficiency of liver tumor diagnosis through better segmentation.

Main Methods:

• A deep residual attention-based U-shaped network (DRAUNet) was developed, incorporating a biplane joint method.
• The biplane method integrates coronal CT slices with transverse slices to enrich 3D spatial context.
• Novel components include a deep residual (DR) block with deeper layers and dual-effect attention module (DAM) for feature correlation and spatial awareness.

Main Results:

• DRAUNet achieved high Dice Similarity Coefficients (DSCs): 97.3% on LiTS, 97.4% on 3DIRCADb, and 96.9% on Sliver07.
• The proposed method significantly outperformed existing state-of-the-art segmentation networks.
• The biplane joint method effectively utilized 3D spatial information for improved segmentation.

Conclusions:

• DRAUNet demonstrates superior performance in liver segmentation, outperforming current methods.
• The biplane joint method is effective in capturing and utilizing 3D spatial information from CT scans.
• This approach offers a promising advancement for accurate and efficient liver tumor diagnosis.