Fan beam CT image synthesis from cone beam CT image using nested residual UNet based conditional generative...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiation Therapy
• Computational Imaging

Background:

• Image-Guided Radiation Therapy (IGRT) enhances treatment accuracy through frequent imaging.
• Fan Beam Computed Tomography (FBCT) and Cone Beam Computed Tomography (CBCT) are key IGRT imaging modalities.
• Replacing FBCT with CBCT could reduce radiation exposure and costs, but requires methods to maintain image quality.

Purpose of the Study:

• To develop a Conditional Generative Adversarial Network (CGAN) for synthesizing FBCT images from CBCT data.
• To improve image quality and treatment planning accuracy in IGRT by enabling CBCT-to-FBCT image translation.

Main Methods:

• A novel Nested Residual UNet (NR-UNet) architecture was designed as the generator within the CGAN.
• A composite loss function including adversarial loss, Mean Squared Error (MSE), and Gradient Difference Loss (GDL) was employed.
• The CGAN model processed three consecutive CBCT slices to generate a single FBCT slice, capturing inter-slice dependencies.

Main Results:

• Synthetic FBCT images achieved a Peak Signal-to-Noise Ratio of 34.04±0.93 dB and Structural Similarity Index Measure of 0.9751±0.001.
• The model improved Contrast-to-Noise Ratio by four times compared to input CBCT images, minimizing MSE and blurriness.
• Treatment plans based on synthetic images were closer to FBCT-based plans than those based on original CBCT images.

Conclusions:

• The proposed CGAN model effectively synthesizes high-quality FBCT images from CBCT data, preserving 3D contextual information.
• This approach offers a computationally efficient alternative to full 3D synthesis methods.
• The method demonstrates superior performance over existing state-of-the-art techniques, paving the way for reduced radiation dose and cost in IGRT.