Evaluating deep learning-based image segmentation for radiotherapy planning in pelvic and abdominal cancers

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• AI integration in radiotherapy planning enhances precision oncology.
• Traditional segmentation methods struggle with accuracy and variability.
• Deep learning models face challenges with diverse datasets and complex anatomy.

Purpose of the Study:

• To develop an attention-enhanced domain-adaptive segmentation framework for radiotherapy planning.
• To improve segmentation accuracy in complex anatomical regions for pelvic and abdominal malignancies.
• To address limitations of existing deep learning models in generalization and computational efficiency.

Main Methods:

• Proposed an attention-enhanced domain-adaptive segmentation framework.
• Incorporated a context-aware attention mechanism.
• Utilized a fine-tuned adaptation module for improved generalization.

Main Results:

• Achieved high segmentation accuracy in complex anatomical regions.
• Demonstrated improved performance on heterogeneous multi-center datasets.
• Maintained computational efficiency for clinical integration.

Conclusions:

• The proposed framework enhances radiotherapy planning accuracy and efficiency.
• It facilitates robust and reproducible contouring of organs and lesions.
• This contributes to more effective and individualized radiation therapy.