Related Experiment Video
Updated: Jan 5, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
Improving automatic delineation for head and neck organs at risk by Deep Learning Contouring
Lisanne V van Dijk1, Lisa Van den Bosch1, Paul Aljabar2
1Department of Radiation Oncology, University of Groningen, University Medical Center Groningen, The Netherlands.
Summary
Deep learning contouring (DLC) significantly improved head and neck (HN) organ-at-risk (OAR) auto-contouring compared to atlas-based contouring (ABAS). DLC offers a more accurate and efficient method for radiation therapy planning.
Area of Science:
- Medical Physics
- Radiotherapy
- Artificial Intelligence in Medicine
Background:
- Accurate delineation of head and neck (HN) organs-at-risk (OARs) is critical for effective radiotherapy and understanding treatment side effects.
- Current automatic contouring methods like atlas-based segmentation (ABAS) have limitations requiring improvement.
Purpose of the Study:
- To evaluate the performance of deep learning contouring (DLC) for HN OAR auto-contouring.
- To compare DLC against the established ABAS method using a comprehensive evaluation framework.
Main Methods:
- A deep learning neural network was trained on 589 HN cancer patients.
- DLC and ABAS were independently validated on a cohort of 104 manually contoured patients.
- Performance was assessed using Dice Similarity Coefficient (DICE) and dose differences for 22 OARs, with additional evaluation of time, inter-observer variation, and subjective preference for 7 OARs.
Main Results:
- DLC demonstrated superior or equal performance in 19 out of 22 OARs compared to ABAS (DICE/|Δmean dose|/|Δmax dose|: 0.74/1.1/0.8 Gy for DLC vs. 0.59/4.2/4.1 Gy for ABAS).
- Improvements were most notable in glandular and upper digestive tract OARs.
- DLC significantly reduced contouring time for inexperienced users and was subjectively preferred for precision and similarity to manual contours, often falling within inter-observer variability.
Conclusions:
- Deep learning contouring (DLC) significantly outperforms atlas-based auto-contouring (ABAS) for the majority of head and neck organs-at-risk.
- DLC represents a promising advancement in automated radiotherapy planning, offering improved accuracy and efficiency.
