Automated Tumor Segmentation in Radiotherapy.

Area of Science:

• Medical Physics
• Radiology
• Oncology

Background:

• Autosegmentation of gross tumor volumes (GTVs) offers potential to reduce clinical workload and improve consistency in radiation therapy planning.
• It also facilitates advanced imaging analyses like radiomics for large-scale patient studies.

Purpose of the Study:

• To review modern deep learning approaches for GTV autosegmentation across five major clinical sites.
• To highlight methods nearing clinical adoption, including competition winners and novel solutions to segmentation challenges.

Main Methods:

• Review of recent deep learning techniques applied to GTV autosegmentation.
• Focus on studies demonstrating progress towards clinical implementation.
• Analysis of network architectures and strategies for overcoming specific segmentation hurdles.

Main Results:

• Deep learning models show significant advancements in segmenting tumors in the brain, head and neck, thorax, abdomen, and pelvis.
• International competition results and novel architectural designs indicate increasing efficacy.
• Progress is being made in addressing challenges that hinder clinical integration.

Conclusions:

• Deep learning-based autosegmentation is a rapidly advancing field with the potential to transform radiation treatment planning and radiomics research.
• Further research and development are needed to overcome remaining barriers for widespread clinical adoption and integration.