Related Experiment Video
Updated: Oct 20, 2025
04:25
Author Spotlight: Bridging Gaps in Anatomy and Establishing a Foundation for Algorithmic Studies
Published on: December 15, 2023
3.1K
Evaluation of auto-segmentation accuracy of cloud-based artificial intelligence and atlas-based models
Yuka Urago1,2, Hiroyuki Okamoto3, Tomoya Kaneda4
1Department of Radiological Sciences, Graduate School of Human Health Sciences, Tokyo Metropolitan University, 7-2-10 Higashi-Ogu, Arakawa-ku, Tokyo, 116-8551, Japan.
Radiation Oncology (London, England)
|September 10, 2021
Summary
Artificial intelligence (AI) segmentation is more accurate than atlas-based methods for prostate cancer organs at risk (OARs), improving efficiency. For head and neck cancers, both AI and atlas-based methods show similar accuracy, but AI significantly reduces delineation time.
Area of Science:
- Medical imaging and radiation oncology.
- Development of automated segmentation techniques.
Background:
- Contour delineation in radiation oncology is critical but time-consuming and prone to inter-observer variability.
- Atlas-based and Artificial Intelligence (AI)-based auto-segmentation methods have been developed to improve efficiency and reduce variability.
Purpose of the Study:
- To compare the accuracy and efficiency of AI-based and atlas-based auto-segmentation for Organs at Risk (OARs) in prostate and head and neck cancer patients.
- To evaluate delineation accuracy using quantitative metrics and visual assessment by radiation oncologists.
Main Methods:
- Twenty-one prostate and 30 head and neck cancer patients were included.
- Atlas-based segmentation (MIM Maestro) and AI-based segmentation (MIM Contour ProtégéAI) were applied.
- Delineation accuracy was assessed using Dice Similarity Coefficient (DSC), Hausdorff Distance (HD), Mean Distance to Agreement (MDA), and visual evaluation.
Main Results:
- AI-based segmentation showed higher accuracy than atlas-based for prostate cancer OARs (bladder, rectum) based on DSC, HD, and MDA.
- Visual evaluation revealed errors in atlas-based delineations for prostate cancer, particularly at unclear boundaries.
- For head and neck cancers, no significant differences were found between AI and atlas-based models for most OARs, except for small structures like the optic chiasm.
Conclusions:
- AI-based segmentation offers significantly improved quantitative accuracy over atlas-based methods for prostate cancer.
- While quantitative accuracy was similar for head and neck cancers, AI-based segmentation demonstrated higher efficiency due to reduced manual correction needs.
- AI-based models are expected to enhance segmentation efficiency and substantially decrease delineation time in radiation oncology.
