A recursive ensemble organ segmentation (REOS) framework: application in brain radiotherapy.

Area of Science:

• Medical Imaging
• Radiotherapy
• Artificial Intelligence

Background:

• Accurate segmentation of organs-at-risk (OARs) is crucial for effective radiotherapy planning.
• Manual delineation of OARs is time-consuming and prone to inter-observer variability.
• Automated segmentation methods are needed to improve efficiency and consistency in radiotherapy.

Purpose of the Study:

• To develop and evaluate a novel recursive ensemble organs-at-risk segmentation (REOS) framework.
• To achieve accurate automatic segmentation of multiple OARs in brain radiotherapy.
• To assess the performance of the REOS framework on challenging anatomical structures.

Main Methods:

• Developed a REOS framework utilizing a 3D U-Net architecture for both organ localization and contour detection modules.
• Ensembled image features from localization and detection modules for recursive OAR segmentation.
• Applied the REOS framework to segment six OARs (eyes, brainstem, optical nerves, chiasm) using 80 T1-weighted MRI scans from brain cancer patients.

Main Results:

• Achieved high segmentation accuracy for major OARs: Dice Similarity Coefficient (DSC) of 93.9% ± 1.4% (left eye), 94.5% ± 2.0% (right eye), and 90.6% ± 2.7% (brainstem).
• Demonstrated effectiveness on smaller, challenging OARs: DSC of 78.0% ± 10.5% (left optical nerve), 82.2% ± 5.9% (right optical nerve), and 71.1% ± 9.1% (chiasm).
• The REOS framework showed robust performance across various OARs.

Conclusions:

• The developed REOS framework provides accurate and automated segmentation of OARs for brain radiotherapy.
• The recursive ensemble approach enhances segmentation precision, particularly for complex anatomical regions.
• This automated segmentation method holds significant potential for improving radiotherapy planning and delivery.