Deep learning-based target tracking with X-ray images for radiotherapy: a narrative review

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiation Oncology

Background:

• Radiotherapy (RT) is a primary cancer treatment modality.
• Accurate localization of tumors and organs at risk (OARs) is crucial for effective RT.
• 2D kV X-ray images from LINACs offer potential for marker-less tracking, but face challenges in soft tissue contrast and organ overlap.

Purpose of the Study:

• To review the current development of deep learning-based target tracking using 2D kV X-ray images in RT.
• To discuss existing limitations, challenges, potential solutions, and future directions in this field.

Main Methods:

• A narrative review of English-language articles was conducted.
• Searches were performed on Web of Science, PubMed, and Google Scholar.
• Keywords included radiotherapy, motion tracking, X-ray images, and deep learning; 23 articles (March 2019-December 2023) were included.

Main Results:

• Deep learning methods are feasible for marker-less target tracking and real-time motion management in RT.
• Challenges persist in real-time tumor and OAR localization using 2D kV X-ray images.
• Further technical and clinical advancements are necessary.

Conclusions:

• Deep learning-based 2D kV X-ray image target tracking is a promising approach for RT motion management.
• Potential benefits include real-time motion recognition, reduced margins, and better normal tissue sparing.
• Significant development is still required before widespread clinical adoption.