Automatic multiorgan segmentation in thorax CT images using U-net-GAN.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiation Oncology

Background:

• Accurate segmentation of organs at risk (OARs) is crucial for effective radiation therapy planning.
• Current manual segmentation methods are time-consuming and prone to inter-observer variability.
• Automated segmentation can significantly improve the efficiency and quality of radiotherapy.

Purpose of the Study:

• To develop and evaluate a deep learning-based method for automatic segmentation of multiple thoracic OARs on CT images.
• To improve the accuracy and efficiency of OAR segmentation for radiotherapy treatment planning.

Main Methods:

• An adversarial training strategy using a U-Net-generative adversarial network (U-Net-GAN) was proposed.
• The U-Net-GAN architecture comprises U-Nets as generators and fully convolutional networks (FCNs) as discriminators.
• The model was trained on 35 chest CT datasets and evaluated against manual segmentations.

Main Results:

• The U-Net-GAN achieved high segmentation accuracy for multiple thoracic OARs, with Dice similarity coefficients ranging from 0.75 to 0.97.
• Mean surface distances were low, ranging from 0.4 to 1.5 mm.
• Dosimetric evaluation on 20 SBRT lung plans showed minimal mean dose differences, indicating clinical relevance.

Conclusions:

• The developed deep learning approach demonstrates feasibility and reliability for segmenting multiple thoracic OARs.
• This automated segmentation method has the potential to significantly enhance the efficiency of chest radiotherapy planning.
• The U-Net-GAN strategy offers a promising tool for improving radiotherapy precision and patient outcomes.