Deep Learning: A Review for the Radiation Oncologist.

Area of Science:

• Radiation Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Deep Learning (DL), a subset of machine learning utilizing deep neural networks, offers advanced modeling capabilities.
• Radiation oncology applications for DL include image segmentation, outcome prediction, and treatment planning.
• The use of DL in radiation oncology has seen a significant surge, particularly from 2015 onwards.

Purpose of the Study:

• To review the methods and applications of Deep Learning in radiation oncology.
• To synthesize recent literature on DL in radiation oncology for a clinically-oriented audience.
• To identify trends and the current status of DL implementation in the field.

Main Methods:

• A literature review was conducted using PubMed/Medline with search terms 'radiotherapy' and 'deep learning'.
• The search identified studies published between 1997 and 2018, with a focus on recent publications.
• Reference lists of selected articles were hand-searched to ensure comprehensive coverage.

Main Results:

• DL applications in image segmentation were found across various cancer sites, including Brain, Head and Neck, Lung, Abdominal, and Pelvic cancers.
• DL has been utilized for predicting clinical outcomes, such as toxicity modeling, treatment response, survival, and aiding in treatment planning.
• The majority of identified studies were published from 2015 to 2018, indicating rapid recent development.

Conclusions:

• Deep Learning techniques show significant potential to enhance radiation oncology practices by improving efficiency and accuracy.
• DL-based systems can assist clinicians by reducing segmentation time and variability, and by predicting treatment outcomes and toxicities.
• Further evaluation is needed to determine the timeline for the routine clinical integration of these advanced DL technologies.