Edge-guided multi-scale adaptive feature fusion network for liver tumor segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Automated segmentation of liver tumors on CT scans is critical for diagnosis and treatment assessment.
• Computer-aided diagnosis offers potential cost reduction and error minimization in clinical settings.
• Existing segmentation methods struggle with the diverse sizes and blurred boundaries of liver tumors.

Purpose of the Study:

• To develop an advanced deep learning network for accurate liver tumor segmentation in CT images.
• To address challenges posed by varying tumor sizes and indistinct tumor margins.
• To improve the reliability and efficiency of computer-aided diagnosis for liver pathologies.

Main Methods:

• Proposed MAEG-Net, a multi-scale adaptive feature fusion network with edge guidance.
• Designed a multi-scale adaptive feature fusion module to integrate multi-scale information.
• Introduced an edge-aware guidance module to enhance feature learning for blurred boundaries.

Main Results:

• Achieved a Dice coefficient of 71.84% on the LiTS2017 liver tumor dataset.
• Obtained a Volumetric Overlap Error (VOE) of 38.64%.
• Demonstrated superior performance compared to existing methods for liver tumor segmentation.

Conclusions:

• MAEG-Net effectively segments liver tumors in CT images, outperforming current techniques.
• The network's multi-scale and edge-aware modules successfully handle variations in tumor size and boundary definition.
• This approach shows significant promise for improving automated liver tumor analysis in clinical practice.