Related Experiment Video
Updated: Jul 5, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Deep learning for automated segmentation in radiotherapy: a narrative review
Jean-Emmanuel Bibault1,2, Paul Giraud2,3
1Radiation Oncology Department, Georges Pompidou European Hospital, Assistance Publique-Hôpitaux de Paris, Université de Paris Cité, Paris, 75015, France.
The British Journal of Radiology
|January 24, 2024
Summary
Deep learning (DL) automates organ segmentation for radiation therapy planning, reducing manual effort and variability. U-net is the preferred deep neural network architecture, though standardization and external validation are needed.
Area of Science:
- Medical imaging and radiation oncology.
- Application of artificial intelligence in healthcare.
- Computational anatomy and image analysis.
Background:
- Manual segmentation of organs and structures is crucial but laborious for radiation therapy planning.
- Interobserver variability in manual segmentation can negatively affect radiation therapy outcomes.
- Deep learning (DL) offers automated solutions for segmentation, improving efficiency and consistency.
Approach:
- This article reviews the literature on DL techniques for segmentation in radiation therapy planning.
- Focuses on five clinical sub-sites: brain, head and neck, lung, abdominal, and pelvic cancers.
- Identifies commonly used DL architectures and validation methods.
Key Points:
- Convolutional Neural Networks (CNNs), particularly U-net, are the most prevalent DL architectures for segmentation in this field.
- The majority of studies reviewed concentrate on segmenting normal tissue structures.
- N-fold cross-validation is frequently used, but external validation is often lacking.
Conclusions:
- Deep learning shows significant promise for automating segmentation in radiation therapy planning.
- Standardization of evaluation metrics and independent external validation are essential for robust benchmarking.
- Further research is needed to fully integrate and validate DL methods in clinical practice.
