Enhanced abdominal multi-organ segmentation with 3D UNet and UNet +  + deep neural networks utilizing the MONAI...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Anatomy

Background:

• Accurate abdominal organ segmentation is crucial for medical analysis and treatment planning.
• Challenges include anatomical variability, complex organ shapes, and noise in CT/MRI scans.

Purpose of the Study:

• To propose and evaluate 3D UNet and UNet++ architectures for multi-organ abdominal segmentation.
• To address segmentation challenges using advanced deep learning techniques and optimization.

Main Methods:

• Implementation of 3D UNet and UNet++ architectures within the MONAI framework.
• Utilizing skip and dense connections for enhanced feature extraction.
• Parameter optimization using Secretary Bird Optimization (SBO).
• Evaluation on Pancreas-CT, Liver-CT, and BTCV datasets.

Main Results:

• Achieved high Dice Similarity Coefficient (DSC) scores across datasets.
• 3D UNet++ outperformed 3D UNet on Pancreas-CT (95.62% vs 94.54%) and Liver-CT (97.36% vs 95.67%).
• Both models demonstrated robust performance on the BTCV dataset (DSC 93.42%–95.31%).

Conclusions:

• The proposed 3D UNet and UNet++ models are robust and efficient for multi-organ abdominal segmentation.
• The SBO-optimized architectures enhance accuracy in medical imaging.
• Validated for clinical applications and scalable solutions in complex abdominal imaging tasks.