DART-Net: A Novel Deep Learning Framework for Precise Radiotherapy Planning with Automated Multiorgan Segmentation...

Area of Science:

• Medical Imaging and Radiation Oncology
• Artificial Intelligence in Healthcare
• Deep Learning for Medical Image Segmentation

Background:

• Manual delineation of pelvic organs in radiotherapy planning is time-consuming and prone to interobserver variability.
• Existing deep learning methods face challenges with complex anatomical boundaries and clinical workflow integration.
• There is a need for automated, accurate, and clinically integrated solutions for pelvic organ segmentation.

Purpose of the Study:

• To introduce the Dual-Attention Residual Technology Network (DART-Net), a novel framework for automated pelvic organ segmentation.
• To achieve direct generation of RTSTRUCT files for seamless integration into radiotherapy planning systems.
• To address limitations of existing deep learning approaches in accuracy and clinical workflow compatibility.

Main Methods:

• Developed DART-Net, a dual-encoder architecture incorporating attention mechanisms and residual connections.
• Processed raw and normalized CT volumes simultaneously for enhanced feature extraction.
• Trained the model on 125 expert-annotated pelvic CT scans with extensive data augmentation.

Main Results:

• Achieved state-of-the-art segmentation performance: Dice scores of 0.965 (bladder), 0.908 (prostate), and 0.840 (rectum).
• Reduced bladder boundary error by 18.6% with Hausdorff Distance values between 1.4-2.5 mm.
• Expert validation confirmed clinical acceptability; automated RTSTRUCT generation enabled seamless clinical integration.

Conclusions:

• DART-Net sets a new benchmark for AI-assisted radiotherapy planning, balancing technical innovation with clinical utility.
• The framework significantly reduces contouring time and improves anatomical precision in radiation oncology.
• DART-Net addresses critical workflow inefficiencies and has the potential to enhance treatment outcomes through consistent organ delineation.