Explainable and Robust Deep Learning for Liver Segmentation Through U-Net Network.

Area of Science:

• Medical imaging analysis
• Deep learning in radiology
• Computational anatomy

Background:

• Clinical imaging (MRI, CT) is crucial for diagnosing diseases like hepatocellular carcinoma.
• Accurate liver and tumor segmentation is vital for staging and treatment planning.
• Effective segmentation impacts diagnostic accuracy and patient outcomes.

Purpose of the Study:

• To develop a deep learning-based approach for accurate liver segmentation in medical images.
• To address the critical need for improved hepatic disease diagnosis and treatment planning.
• To incorporate prediction explainability into the segmentation process.

Main Methods:

• Utilized a U-Net architecture with residual connections for detailed anatomical feature capture.
• Trained two models on two distinct annotated computed tomography (CT) datasets.
• Implemented four experiments to evaluate model performance and robustness.

Main Results:

• Achieved high segmentation accuracy, ranging from 0.81 to 0.93.
• Demonstrated robustness and generalization across diverse datasets and imaging conditions.
• Provided explainability by highlighting image areas relevant to segmentation.

Conclusions:

• The proposed deep learning method offers a reliable and efficient solution for automated liver segmentation.
• This technology promises significant advancements in clinical workflows and precision medicine.
• Automated segmentation aids in optimizing diagnosis, staging, and treatment for liver diseases.