Automatic Liver Segmentation Using EfficientNet and Attention-Based Residual U-Net in CT.

Area of Science:

• Medical Imaging
• Computer Vision
• Deep Learning

Background:

• Accurate liver segmentation is crucial for medical diagnosis and treatment planning.
• Existing methods face challenges in feature extraction and segmentation accuracy.

Purpose of the Study:

• To develop and validate a novel network framework for automatic and accurate liver segmentation.
• To improve upon existing deep learning models for liver segmentation tasks.

Main Methods:

• Utilized EfficientNetB4 as an encoder for enhanced feature extraction.
• Incorporated an attention gate in skip connections to focus on relevant features.
• Employed residual blocks in the decoder to mitigate gradient vanishing and boost accuracy.

Main Results:

• Achieved superior segmentation performance on the SLiver07 dataset across all five standard metrics.
• Demonstrated competitive results on the LiTS17 dataset, with minor inferiority only in RVD.
• Qualitative and quantitative analyses confirmed the method's effectiveness and applicability.

Conclusions:

• The proposed framework offers a robust solution for automatic liver segmentation.
• The integration of EfficientNetB4, attention gates, and residual learning significantly enhances segmentation accuracy.
• The method shows strong potential for computer-assisted liver segmentation in clinical settings.