A 4D-CBCT correction network based on contrastive learning for dose calculation in lung cancer

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiation Oncology

Background:

• Four-dimensional cone beam computed tomography (4D-CBCT) is crucial for lung cancer radiotherapy but suffers from streak artifacts and inaccurate CT values.
• These image quality issues can compromise the accuracy of radiation dose calculations.

Purpose of the Study:

• To introduce a deep-learning network, contrastive learning-based cycle generative adversarial networks (CLCGAN), for artifact reduction and CT value correction in 4D-CBCT.
• To evaluate the impact of CLCGAN on image quality and dose calculation accuracy for lung cancer patients.

Main Methods:

• A CLCGAN model was trained using paired 4D-CBCT and 4D CT data from lung cancer patients.
• The model generated 4D synthetic CT (4D-sCT) images, which were quantitatively and qualitatively assessed against ground truth 4D CT.
• Dose distributions and calculations derived from 4D-CBCT and 4D-sCT were compared to 4D CT.

Main Results:

• CLCGAN significantly improved image quality metrics (SSIM, PSNR) of 4D-sCT compared to baseline methods.
• The network reduced absolute mean differences in CT values across various tissues (lungs, bones, soft tissues).
• Dose calculations using CLCGAN-corrected 4D-CBCT showed significant improvements, with reduced relative dose differences compared to standard 4D-CBCT.

Conclusions:

• CLCGAN effectively mitigates streak artifacts and corrects CT values in 4D-CBCT for lung cancer patients.
• The improved image quality and CT value accuracy facilitate more reliable dose calculations.
• CLCGAN-corrected 4D-CBCT shows potential for clinical use in lung cancer radiotherapy planning and delivery.