Deep learning for automatic target volume segmentation in radiation therapy: a review.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiation Oncology

Background:

• Deep learning is a rapidly advancing field in machine learning, showing state-of-the-art performance in medical imaging applications.
• In radiation oncology, deep learning excels at automatic segmentation of Organs-At-Risks (OAR), improving contouring efficiency and reducing observer variability.
• Automatic segmentation of target volumes is crucial for radiation therapy planning, defining tumor boundaries and encompassing microscopic spread.

Purpose of the Study:

• To summarize major deep learning architectures for supervised target volume segmentation.
• To review the mechanisms of these deep learning models.
• To survey their application across various imaging modalities and clinical sites.

Main Methods:

• Review of supervised deep learning architectures for target volume segmentation.
• Analysis of model mechanisms and applications in CT, MRI, and PET imaging.
• Performance comparison using standard geometric evaluation metrics.

Main Results:

• Deep learning models show promise for automatic target volume segmentation in radiation oncology.
• Performance varies across different architectures, imaging modalities, and clinical sites.
• Standard evaluation metrics are used to compare model effectiveness.

Conclusions:

• Deep learning offers significant potential to improve target volume segmentation in radiation therapy.
• Further research is needed to address open challenges and optimize clinical application.
• Advancements in automatic segmentation can enhance treatment planning and patient outcomes.