Deep Learning-Based Cascade 3D Kidney Segmentation Method.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early diagnosis and precise localization of renal tumors are critical for effective cancer treatment.
• Automated analysis of abdominal CT images can improve diagnostic accuracy and efficiency.
• Semantic segmentation of kidneys is a key step in renal tumor analysis.

Purpose of the Study:

• To develop and evaluate an automated system for renal tumor analysis in abdominal CT images.
• To enhance semantic kidney segmentation using a cascade 3D U-Net architecture.
• To address challenges in segmenting small objects and detecting tumor edges.

Main Methods:

• A cascade 3D U-Net architecture was employed for semantic kidney segmentation.
• Residual blocks were incorporated to improve model convergence and efficiency.
• Extensive training, preprocessing, and postprocessing strategies were utilized.
• The method was evaluated on the KiTS2019 dataset.

Main Results:

• The proposed method achieved high precision in semantic kidney segmentation.
• The cascade 3D U-Net demonstrated effectiveness in handling edge detection and small object segmentation.
• The system ranked 23rd on the KiTS2019 leaderboard (Nov 2024).

Conclusions:

• The enhanced cascade 3D U-Net is a promising approach for automated renal tumor analysis.
• Accurate semantic kidney segmentation is achievable with the proposed deep learning framework.
• This automated method can potentially improve early diagnosis and treatment planning for renal tumors.