RA V-Net: deep learning network for automated liver segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate liver segmentation from CT images is crucial for disease diagnosis and surgical planning.
• Deep learning models show promise for automating medical image segmentation but often struggle with achieving sufficient accuracy.
• Existing methods face challenges in precise feature extraction and handling complex anatomical structures.

Purpose of the Study:

• To develop an advanced deep learning model for enhanced liver segmentation in CT images.
• To improve the accuracy and reliability of automated medical image segmentation.
• To provide a robust tool for pre-surgical planning and diagnosis.

Main Methods:

• Proposed Residual Attention V-Net (RA V-Net), a U-Net based architecture.
• Introduced Composite Original Feature Residual Module for superior feature extraction and gradient stability.
• Incorporated Attention Recovery Module for spatial attention and Channel Attention Module for feature refinement.

Main Results:

• RA V-Net demonstrated significant improvements in segmentation accuracy on Lits2017 and 3Dircadb datasets.
• Achieved higher Dice Similarity Coefficient (DSC) and Jaccard Similarity Coefficient (JSC) compared to standard U-Net.
• Specifically, RA V-Net exceeded U-Net by 0.1107 (DSC) and 0.1214 (JSC) on Lits2017.

Conclusions:

• The proposed RA V-Net model offers superior performance in liver segmentation, minimizing over- and under-segmentation.
• The model precisely delineates organ edges, providing a reliable basis for surgical planning.
• This deep learning approach advances the field of medical image analysis for clinical applications.