Synthetic CT generation from CBCT images via deep learning.

Area of Science:

• Medical Imaging
• Radiotherapy
• Deep Learning

Background:

• Cone-beam computed tomography (CBCT) is crucial for patient setup in image-guided radiotherapy.
• Inaccurate CT numbers in CBCT limit its use in dose calculation and treatment planning.

Purpose of the Study:

• To develop a deep learning model for synthesizing CT-like images with accurate CT numbers from on-treatment CBCT.
• To maintain the anatomical structure of CBCT while improving image quality for advanced radiotherapy applications.

Main Methods:

• A U-net based deep learning architecture (sCTU-net) was employed for CT synthesis.
• The model was trained using on-treatment CBCT and planning CT from 37 patients, with replanning CT images as reference.
• The sCTU-net's performance was evaluated on seven independent patient cases.

Main Results:

• The synthetic CT (sCT) images achieved an average accuracy of 18.98 HU (Mean Absolute Error) compared to reference CT.
• This represents a significant improvement over the original CBCT, which had an MAE of 44.38 HU.

Conclusions:

• The sCTU-net successfully synthesizes CT-quality images with accurate CT numbers from CBCT and planning CT.
• This technology holds potential for enabling advanced CBCT applications, including adaptive radiotherapy treatment planning.